src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_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_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_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_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_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             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 221             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 222             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 226                                                               charge+jnrC+0,charge+jnrD+0);
 227             vdwjidx0A        = 2*vdwtype[jnrA+0];
 228             vdwjidx0B        = 2*vdwtype[jnrB+0];
 229             vdwjidx0C        = 2*vdwtype[jnrC+0];
 230             vdwjidx0D        = 2*vdwtype[jnrD+0];
 231
 232             fjx0             = _mm_setzero_ps();
 233             fjy0             = _mm_setzero_ps();
 234             fjz0             = _mm_setzero_ps();
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             r00              = _mm_mul_ps(rsq00,rinv00);
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 244                                          vdwparam+vdwioffset0+vdwjidx0B,
 245                                          vdwparam+vdwioffset0+vdwjidx0C,
 246                                          vdwparam+vdwioffset0+vdwjidx0D,
 247                                          &c6_00,&c12_00);
 248
 249             /* Calculate table index by multiplying r with table scale and truncate to integer */
 250             rt               = _mm_mul_ps(r00,vftabscale);
 251             vfitab           = _mm_cvttps_epi32(rt);
 252 #ifdef __XOP__
 253             vfeps            = _mm_frcz_ps(rt);
 254 #else
 255             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 256 #endif
 257             twovfeps         = _mm_add_ps(vfeps,vfeps);
 258             vfitab           = _mm_slli_epi32(vfitab,3);
 259
 260             /* CUBIC SPLINE TABLE DISPERSION */
 261             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 262             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 263             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 264             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 265             _MM_TRANSPOSE4_PS(Y,F,G,H);
 266             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 267             VV               = _mm_macc_ps(vfeps,Fp,Y);
 268             vvdw6            = _mm_mul_ps(c6_00,VV);
 269             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 270             fvdw6            = _mm_mul_ps(c6_00,FF);
 271
 272             /* CUBIC SPLINE TABLE REPULSION */
 273             vfitab           = _mm_add_epi32(vfitab,ifour);
 274             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 276             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 277             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 278             _MM_TRANSPOSE4_PS(Y,F,G,H);
 279             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 280             VV               = _mm_macc_ps(vfeps,Fp,Y);
 281             vvdw12           = _mm_mul_ps(c12_00,VV);
 282             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 283             fvdw12           = _mm_mul_ps(c12_00,FF);
 284             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 285             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 289
 290             fscal            = fvdw;
 291
 292              /* Update vectorial force */
 293             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 294             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 295             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 296
 297             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 298             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 299             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_ps(iq1,jq0);
 307
 308             /* COULOMB ELECTROSTATICS */
 309             velec            = _mm_mul_ps(qq10,rinv10);
 310             felec            = _mm_mul_ps(velec,rinvsq10);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_ps(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317              /* Update vectorial force */
 318             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 319             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 320             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 321
 322             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 323             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 324             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq20             = _mm_mul_ps(iq2,jq0);
 332
 333             /* COULOMB ELECTROSTATICS */
 334             velec            = _mm_mul_ps(qq20,rinv20);
 335             felec            = _mm_mul_ps(velec,rinvsq20);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velecsum         = _mm_add_ps(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342              /* Update vectorial force */
 343             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 344             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 345             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 346
 347             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 348             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 349             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq30             = _mm_mul_ps(iq3,jq0);
 357
 358             /* COULOMB ELECTROSTATICS */
 359             velec            = _mm_mul_ps(qq30,rinv30);
 360             felec            = _mm_mul_ps(velec,rinvsq30);
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velecsum         = _mm_add_ps(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367              /* Update vectorial force */
 368             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 369             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 370             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 371
 372             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 373             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 374             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 375
 376             fjptrA             = f+j_coord_offsetA;
 377             fjptrB             = f+j_coord_offsetB;
 378             fjptrC             = f+j_coord_offsetC;
 379             fjptrD             = f+j_coord_offsetD;
 380
 381             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 382
 383             /* Inner loop uses 152 flops */
 384         }
 385
 386         if(jidx<j_index_end)
 387         {
 388
 389             /* Get j neighbor index, and coordinate index */
 390             jnrlistA         = jjnr[jidx];
 391             jnrlistB         = jjnr[jidx+1];
 392             jnrlistC         = jjnr[jidx+2];
 393             jnrlistD         = jjnr[jidx+3];
 394             /* Sign of each element will be negative for non-real atoms.
 395              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 396              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 397              */
 398             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 399             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 400             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 401             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 402             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 403             j_coord_offsetA  = DIM*jnrA;
 404             j_coord_offsetB  = DIM*jnrB;
 405             j_coord_offsetC  = DIM*jnrC;
 406             j_coord_offsetD  = DIM*jnrD;
 407
 408             /* load j atom coordinates */
 409             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 410                                               x+j_coord_offsetC,x+j_coord_offsetD,
 411                                               &jx0,&jy0,&jz0);
 412
 413             /* Calculate displacement vector */
 414             dx00             = _mm_sub_ps(ix0,jx0);
 415             dy00             = _mm_sub_ps(iy0,jy0);
 416             dz00             = _mm_sub_ps(iz0,jz0);
 417             dx10             = _mm_sub_ps(ix1,jx0);
 418             dy10             = _mm_sub_ps(iy1,jy0);
 419             dz10             = _mm_sub_ps(iz1,jz0);
 420             dx20             = _mm_sub_ps(ix2,jx0);
 421             dy20             = _mm_sub_ps(iy2,jy0);
 422             dz20             = _mm_sub_ps(iz2,jz0);
 423             dx30             = _mm_sub_ps(ix3,jx0);
 424             dy30             = _mm_sub_ps(iy3,jy0);
 425             dz30             = _mm_sub_ps(iz3,jz0);
 426
 427             /* Calculate squared distance and things based on it */
 428             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 429             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 430             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 431             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 432
 433             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 434             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 435             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 436             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 437
 438             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 439             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 440             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 441
 442             /* Load parameters for j particles */
 443             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 444                                                               charge+jnrC+0,charge+jnrD+0);
 445             vdwjidx0A        = 2*vdwtype[jnrA+0];
 446             vdwjidx0B        = 2*vdwtype[jnrB+0];
 447             vdwjidx0C        = 2*vdwtype[jnrC+0];
 448             vdwjidx0D        = 2*vdwtype[jnrD+0];
 449
 450             fjx0             = _mm_setzero_ps();
 451             fjy0             = _mm_setzero_ps();
 452             fjz0             = _mm_setzero_ps();
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             r00              = _mm_mul_ps(rsq00,rinv00);
 459             r00              = _mm_andnot_ps(dummy_mask,r00);
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 463                                          vdwparam+vdwioffset0+vdwjidx0B,
 464                                          vdwparam+vdwioffset0+vdwjidx0C,
 465                                          vdwparam+vdwioffset0+vdwjidx0D,
 466                                          &c6_00,&c12_00);
 467
 468             /* Calculate table index by multiplying r with table scale and truncate to integer */
 469             rt               = _mm_mul_ps(r00,vftabscale);
 470             vfitab           = _mm_cvttps_epi32(rt);
 471 #ifdef __XOP__
 472             vfeps            = _mm_frcz_ps(rt);
 473 #else
 474             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 475 #endif
 476             twovfeps         = _mm_add_ps(vfeps,vfeps);
 477             vfitab           = _mm_slli_epi32(vfitab,3);
 478
 479             /* CUBIC SPLINE TABLE DISPERSION */
 480             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 481             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 482             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 483             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 484             _MM_TRANSPOSE4_PS(Y,F,G,H);
 485             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 486             VV               = _mm_macc_ps(vfeps,Fp,Y);
 487             vvdw6            = _mm_mul_ps(c6_00,VV);
 488             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 489             fvdw6            = _mm_mul_ps(c6_00,FF);
 490
 491             /* CUBIC SPLINE TABLE REPULSION */
 492             vfitab           = _mm_add_epi32(vfitab,ifour);
 493             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 494             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 495             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 496             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 497             _MM_TRANSPOSE4_PS(Y,F,G,H);
 498             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 499             VV               = _mm_macc_ps(vfeps,Fp,Y);
 500             vvdw12           = _mm_mul_ps(c12_00,VV);
 501             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 502             fvdw12           = _mm_mul_ps(c12_00,FF);
 503             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 504             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 508             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 509
 510             fscal            = fvdw;
 511
 512             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 513
 514              /* Update vectorial force */
 515             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 516             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 517             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 518
 519             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 520             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 521             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             /* Compute parameters for interactions between i and j atoms */
 528             qq10             = _mm_mul_ps(iq1,jq0);
 529
 530             /* COULOMB ELECTROSTATICS */
 531             velec            = _mm_mul_ps(qq10,rinv10);
 532             felec            = _mm_mul_ps(velec,rinvsq10);
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm_andnot_ps(dummy_mask,velec);
 536             velecsum         = _mm_add_ps(velecsum,velec);
 537
 538             fscal            = felec;
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542              /* Update vectorial force */
 543             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 544             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 545             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 546
 547             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 548             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 549             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq20             = _mm_mul_ps(iq2,jq0);
 557
 558             /* COULOMB ELECTROSTATICS */
 559             velec            = _mm_mul_ps(qq20,rinv20);
 560             felec            = _mm_mul_ps(velec,rinvsq20);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velec            = _mm_andnot_ps(dummy_mask,velec);
 564             velecsum         = _mm_add_ps(velecsum,velec);
 565
 566             fscal            = felec;
 567
 568             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 569
 570              /* Update vectorial force */
 571             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 572             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 573             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 574
 575             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 576             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 577             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             qq30             = _mm_mul_ps(iq3,jq0);
 585
 586             /* COULOMB ELECTROSTATICS */
 587             velec            = _mm_mul_ps(qq30,rinv30);
 588             felec            = _mm_mul_ps(velec,rinvsq30);
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             velec            = _mm_andnot_ps(dummy_mask,velec);
 592             velecsum         = _mm_add_ps(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 597
 598              /* Update vectorial force */
 599             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 600             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 601             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 602
 603             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 604             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 605             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 606
 607             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 608             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 609             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 610             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 611
 612             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 613
 614             /* Inner loop uses 153 flops */
 615         }
 616
 617         /* End of innermost loop */
 618
 619         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 620                                               f+i_coord_offset,fshift+i_shift_offset);
 621
 622         ggid                        = gid[iidx];
 623         /* Update potential energies */
 624         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 625         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 626
 627         /* Increment number of inner iterations */
 628         inneriter                  += j_index_end - j_index_start;
 629
 630         /* Outer loop uses 26 flops */
 631     }
 632
 633     /* Increment number of outer iterations */
 634     outeriter        += nri;
 635
 636     /* Update outer/inner flops */
 637
 638     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*153);
 639 }
 640 /*
 641  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 642  * Electrostatics interaction: Coulomb
 643  * VdW interaction:            CubicSplineTable
 644  * Geometry:                   Water4-Particle
 645  * Calculate force/pot:        Force
 646  */
 647 void
 648 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 649                     (t_nblist * gmx_restrict                nlist,
 650                      rvec * gmx_restrict                    xx,
 651                      rvec * gmx_restrict                    ff,
 652                      t_forcerec * gmx_restrict              fr,
 653                      t_mdatoms * gmx_restrict               mdatoms,
 654                      nb_kernel_data_t * gmx_restrict        kernel_data,
 655                      t_nrnb * gmx_restrict                  nrnb)
 656 {
 657     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 658      * just 0 for non-waters.
 659      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 660      * jnr indices corresponding to data put in the four positions in the SIMD register.
 661      */
 662     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 663     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 664     int              jnrA,jnrB,jnrC,jnrD;
 665     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 666     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 667     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 668     real             rcutoff_scalar;
 669     real             *shiftvec,*fshift,*x,*f;
 670     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 671     real             scratch[4*DIM];
 672     __m128           fscal,rcutoff,rcutoff2,jidxall;
 673     int              vdwioffset0;
 674     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 675     int              vdwioffset1;
 676     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 677     int              vdwioffset2;
 678     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 679     int              vdwioffset3;
 680     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 681     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 682     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 683     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 684     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 685     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 686     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 687     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 688     real             *charge;
 689     int              nvdwtype;
 690     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 691     int              *vdwtype;
 692     real             *vdwparam;
 693     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 694     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 695     __m128i          vfitab;
 696     __m128i          ifour       = _mm_set1_epi32(4);
 697     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 698     real             *vftab;
 699     __m128           dummy_mask,cutoff_mask;
 700     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 701     __m128           one     = _mm_set1_ps(1.0);
 702     __m128           two     = _mm_set1_ps(2.0);
 703     x                = xx[0];
 704     f                = ff[0];
 705
 706     nri              = nlist->nri;
 707     iinr             = nlist->iinr;
 708     jindex           = nlist->jindex;
 709     jjnr             = nlist->jjnr;
 710     shiftidx         = nlist->shift;
 711     gid              = nlist->gid;
 712     shiftvec         = fr->shift_vec[0];
 713     fshift           = fr->fshift[0];
 714     facel            = _mm_set1_ps(fr->epsfac);
 715     charge           = mdatoms->chargeA;
 716     nvdwtype         = fr->ntype;
 717     vdwparam         = fr->nbfp;
 718     vdwtype          = mdatoms->typeA;
 719
 720     vftab            = kernel_data->table_vdw->data;
 721     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 722
 723     /* Setup water-specific parameters */
 724     inr              = nlist->iinr[0];
 725     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 726     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 727     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 728     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 729
 730     /* Avoid stupid compiler warnings */
 731     jnrA = jnrB = jnrC = jnrD = 0;
 732     j_coord_offsetA = 0;
 733     j_coord_offsetB = 0;
 734     j_coord_offsetC = 0;
 735     j_coord_offsetD = 0;
 736
 737     outeriter        = 0;
 738     inneriter        = 0;
 739
 740     for(iidx=0;iidx<4*DIM;iidx++)
 741     {
 742         scratch[iidx] = 0.0;
 743     }
 744
 745     /* Start outer loop over neighborlists */
 746     for(iidx=0; iidx<nri; iidx++)
 747     {
 748         /* Load shift vector for this list */
 749         i_shift_offset   = DIM*shiftidx[iidx];
 750
 751         /* Load limits for loop over neighbors */
 752         j_index_start    = jindex[iidx];
 753         j_index_end      = jindex[iidx+1];
 754
 755         /* Get outer coordinate index */
 756         inr              = iinr[iidx];
 757         i_coord_offset   = DIM*inr;
 758
 759         /* Load i particle coords and add shift vector */
 760         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 761                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 762
 763         fix0             = _mm_setzero_ps();
 764         fiy0             = _mm_setzero_ps();
 765         fiz0             = _mm_setzero_ps();
 766         fix1             = _mm_setzero_ps();
 767         fiy1             = _mm_setzero_ps();
 768         fiz1             = _mm_setzero_ps();
 769         fix2             = _mm_setzero_ps();
 770         fiy2             = _mm_setzero_ps();
 771         fiz2             = _mm_setzero_ps();
 772         fix3             = _mm_setzero_ps();
 773         fiy3             = _mm_setzero_ps();
 774         fiz3             = _mm_setzero_ps();
 775
 776         /* Start inner kernel loop */
 777         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 778         {
 779
 780             /* Get j neighbor index, and coordinate index */
 781             jnrA             = jjnr[jidx];
 782             jnrB             = jjnr[jidx+1];
 783             jnrC             = jjnr[jidx+2];
 784             jnrD             = jjnr[jidx+3];
 785             j_coord_offsetA  = DIM*jnrA;
 786             j_coord_offsetB  = DIM*jnrB;
 787             j_coord_offsetC  = DIM*jnrC;
 788             j_coord_offsetD  = DIM*jnrD;
 789
 790             /* load j atom coordinates */
 791             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 792                                               x+j_coord_offsetC,x+j_coord_offsetD,
 793                                               &jx0,&jy0,&jz0);
 794
 795             /* Calculate displacement vector */
 796             dx00             = _mm_sub_ps(ix0,jx0);
 797             dy00             = _mm_sub_ps(iy0,jy0);
 798             dz00             = _mm_sub_ps(iz0,jz0);
 799             dx10             = _mm_sub_ps(ix1,jx0);
 800             dy10             = _mm_sub_ps(iy1,jy0);
 801             dz10             = _mm_sub_ps(iz1,jz0);
 802             dx20             = _mm_sub_ps(ix2,jx0);
 803             dy20             = _mm_sub_ps(iy2,jy0);
 804             dz20             = _mm_sub_ps(iz2,jz0);
 805             dx30             = _mm_sub_ps(ix3,jx0);
 806             dy30             = _mm_sub_ps(iy3,jy0);
 807             dz30             = _mm_sub_ps(iz3,jz0);
 808
 809             /* Calculate squared distance and things based on it */
 810             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 811             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 812             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 813             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 814
 815             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 816             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 817             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 818             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 819
 820             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 821             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 822             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 823
 824             /* Load parameters for j particles */
 825             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 826                                                               charge+jnrC+0,charge+jnrD+0);
 827             vdwjidx0A        = 2*vdwtype[jnrA+0];
 828             vdwjidx0B        = 2*vdwtype[jnrB+0];
 829             vdwjidx0C        = 2*vdwtype[jnrC+0];
 830             vdwjidx0D        = 2*vdwtype[jnrD+0];
 831
 832             fjx0             = _mm_setzero_ps();
 833             fjy0             = _mm_setzero_ps();
 834             fjz0             = _mm_setzero_ps();
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             r00              = _mm_mul_ps(rsq00,rinv00);
 841
 842             /* Compute parameters for interactions between i and j atoms */
 843             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 844                                          vdwparam+vdwioffset0+vdwjidx0B,
 845                                          vdwparam+vdwioffset0+vdwjidx0C,
 846                                          vdwparam+vdwioffset0+vdwjidx0D,
 847                                          &c6_00,&c12_00);
 848
 849             /* Calculate table index by multiplying r with table scale and truncate to integer */
 850             rt               = _mm_mul_ps(r00,vftabscale);
 851             vfitab           = _mm_cvttps_epi32(rt);
 852 #ifdef __XOP__
 853             vfeps            = _mm_frcz_ps(rt);
 854 #else
 855             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 856 #endif
 857             twovfeps         = _mm_add_ps(vfeps,vfeps);
 858             vfitab           = _mm_slli_epi32(vfitab,3);
 859
 860             /* CUBIC SPLINE TABLE DISPERSION */
 861             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 862             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 863             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 864             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 865             _MM_TRANSPOSE4_PS(Y,F,G,H);
 866             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 867             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 868             fvdw6            = _mm_mul_ps(c6_00,FF);
 869
 870             /* CUBIC SPLINE TABLE REPULSION */
 871             vfitab           = _mm_add_epi32(vfitab,ifour);
 872             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 873             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 874             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 875             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 876             _MM_TRANSPOSE4_PS(Y,F,G,H);
 877             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 878             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 879             fvdw12           = _mm_mul_ps(c12_00,FF);
 880             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 881
 882             fscal            = fvdw;
 883
 884              /* Update vectorial force */
 885             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 886             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 887             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 888
 889             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 890             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 891             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 892
 893             /**************************
 894              * CALCULATE INTERACTIONS *
 895              **************************/
 896
 897             /* Compute parameters for interactions between i and j atoms */
 898             qq10             = _mm_mul_ps(iq1,jq0);
 899
 900             /* COULOMB ELECTROSTATICS */
 901             velec            = _mm_mul_ps(qq10,rinv10);
 902             felec            = _mm_mul_ps(velec,rinvsq10);
 903
 904             fscal            = felec;
 905
 906              /* Update vectorial force */
 907             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 908             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 909             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 910
 911             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 912             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 913             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 914
 915             /**************************
 916              * CALCULATE INTERACTIONS *
 917              **************************/
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq20             = _mm_mul_ps(iq2,jq0);
 921
 922             /* COULOMB ELECTROSTATICS */
 923             velec            = _mm_mul_ps(qq20,rinv20);
 924             felec            = _mm_mul_ps(velec,rinvsq20);
 925
 926             fscal            = felec;
 927
 928              /* Update vectorial force */
 929             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 930             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 931             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 932
 933             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 934             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 935             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 936
 937             /**************************
 938              * CALCULATE INTERACTIONS *
 939              **************************/
 940
 941             /* Compute parameters for interactions between i and j atoms */
 942             qq30             = _mm_mul_ps(iq3,jq0);
 943
 944             /* COULOMB ELECTROSTATICS */
 945             velec            = _mm_mul_ps(qq30,rinv30);
 946             felec            = _mm_mul_ps(velec,rinvsq30);
 947
 948             fscal            = felec;
 949
 950              /* Update vectorial force */
 951             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 952             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 953             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 954
 955             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 956             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 957             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 958
 959             fjptrA             = f+j_coord_offsetA;
 960             fjptrB             = f+j_coord_offsetB;
 961             fjptrC             = f+j_coord_offsetC;
 962             fjptrD             = f+j_coord_offsetD;
 963
 964             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 965
 966             /* Inner loop uses 141 flops */
 967         }
 968
 969         if(jidx<j_index_end)
 970         {
 971
 972             /* Get j neighbor index, and coordinate index */
 973             jnrlistA         = jjnr[jidx];
 974             jnrlistB         = jjnr[jidx+1];
 975             jnrlistC         = jjnr[jidx+2];
 976             jnrlistD         = jjnr[jidx+3];
 977             /* Sign of each element will be negative for non-real atoms.
 978              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 979              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 980              */
 981             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 982             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 983             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 984             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 985             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 986             j_coord_offsetA  = DIM*jnrA;
 987             j_coord_offsetB  = DIM*jnrB;
 988             j_coord_offsetC  = DIM*jnrC;
 989             j_coord_offsetD  = DIM*jnrD;
 990
 991             /* load j atom coordinates */
 992             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 993                                               x+j_coord_offsetC,x+j_coord_offsetD,
 994                                               &jx0,&jy0,&jz0);
 995
 996             /* Calculate displacement vector */
 997             dx00             = _mm_sub_ps(ix0,jx0);
 998             dy00             = _mm_sub_ps(iy0,jy0);
 999             dz00             = _mm_sub_ps(iz0,jz0);
1000             dx10             = _mm_sub_ps(ix1,jx0);
1001             dy10             = _mm_sub_ps(iy1,jy0);
1002             dz10             = _mm_sub_ps(iz1,jz0);
1003             dx20             = _mm_sub_ps(ix2,jx0);
1004             dy20             = _mm_sub_ps(iy2,jy0);
1005             dz20             = _mm_sub_ps(iz2,jz0);
1006             dx30             = _mm_sub_ps(ix3,jx0);
1007             dy30             = _mm_sub_ps(iy3,jy0);
1008             dz30             = _mm_sub_ps(iz3,jz0);
1009
1010             /* Calculate squared distance and things based on it */
1011             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1012             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1013             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1014             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1015
1016             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1017             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1018             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1019             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1020
1021             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1022             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1023             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1024
1025             /* Load parameters for j particles */
1026             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1027                                                               charge+jnrC+0,charge+jnrD+0);
1028             vdwjidx0A        = 2*vdwtype[jnrA+0];
1029             vdwjidx0B        = 2*vdwtype[jnrB+0];
1030             vdwjidx0C        = 2*vdwtype[jnrC+0];
1031             vdwjidx0D        = 2*vdwtype[jnrD+0];
1032
1033             fjx0             = _mm_setzero_ps();
1034             fjy0             = _mm_setzero_ps();
1035             fjz0             = _mm_setzero_ps();
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             r00              = _mm_mul_ps(rsq00,rinv00);
1042             r00              = _mm_andnot_ps(dummy_mask,r00);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1046                                          vdwparam+vdwioffset0+vdwjidx0B,
1047                                          vdwparam+vdwioffset0+vdwjidx0C,
1048                                          vdwparam+vdwioffset0+vdwjidx0D,
1049                                          &c6_00,&c12_00);
1050
1051             /* Calculate table index by multiplying r with table scale and truncate to integer */
1052             rt               = _mm_mul_ps(r00,vftabscale);
1053             vfitab           = _mm_cvttps_epi32(rt);
1054 #ifdef __XOP__
1055             vfeps            = _mm_frcz_ps(rt);
1056 #else
1057             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1058 #endif
1059             twovfeps         = _mm_add_ps(vfeps,vfeps);
1060             vfitab           = _mm_slli_epi32(vfitab,3);
1061
1062             /* CUBIC SPLINE TABLE DISPERSION */
1063             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1064             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1065             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1066             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1067             _MM_TRANSPOSE4_PS(Y,F,G,H);
1068             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1069             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1070             fvdw6            = _mm_mul_ps(c6_00,FF);
1071
1072             /* CUBIC SPLINE TABLE REPULSION */
1073             vfitab           = _mm_add_epi32(vfitab,ifour);
1074             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1075             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1076             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1077             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1078             _MM_TRANSPOSE4_PS(Y,F,G,H);
1079             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1080             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1081             fvdw12           = _mm_mul_ps(c12_00,FF);
1082             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1083
1084             fscal            = fvdw;
1085
1086             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1087
1088              /* Update vectorial force */
1089             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1090             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1091             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1092
1093             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1094             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1095             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq10             = _mm_mul_ps(iq1,jq0);
1103
1104             /* COULOMB ELECTROSTATICS */
1105             velec            = _mm_mul_ps(qq10,rinv10);
1106             felec            = _mm_mul_ps(velec,rinvsq10);
1107
1108             fscal            = felec;
1109
1110             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1111
1112              /* Update vectorial force */
1113             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1114             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1115             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1116
1117             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1118             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1119             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             /* Compute parameters for interactions between i and j atoms */
1126             qq20             = _mm_mul_ps(iq2,jq0);
1127
1128             /* COULOMB ELECTROSTATICS */
1129             velec            = _mm_mul_ps(qq20,rinv20);
1130             felec            = _mm_mul_ps(velec,rinvsq20);
1131
1132             fscal            = felec;
1133
1134             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1135
1136              /* Update vectorial force */
1137             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1138             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1139             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1140
1141             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1142             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1143             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1144
1145             /**************************
1146              * CALCULATE INTERACTIONS *
1147              **************************/
1148
1149             /* Compute parameters for interactions between i and j atoms */
1150             qq30             = _mm_mul_ps(iq3,jq0);
1151
1152             /* COULOMB ELECTROSTATICS */
1153             velec            = _mm_mul_ps(qq30,rinv30);
1154             felec            = _mm_mul_ps(velec,rinvsq30);
1155
1156             fscal            = felec;
1157
1158             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1159
1160              /* Update vectorial force */
1161             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1162             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1163             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1164
1165             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1166             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1167             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1168
1169             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1170             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1171             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1172             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1173
1174             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1175
1176             /* Inner loop uses 142 flops */
1177         }
1178
1179         /* End of innermost loop */
1180
1181         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1182                                               f+i_coord_offset,fshift+i_shift_offset);
1183
1184         /* Increment number of inner iterations */
1185         inneriter                  += j_index_end - j_index_start;
1186
1187         /* Outer loop uses 24 flops */
1188     }
1189
1190     /* Increment number of outer iterations */
1191     outeriter        += nri;
1192
1193     /* Update outer/inner flops */
1194
1195     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*142);
1196 }