src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double 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_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  84     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  85     __m128i          vfitab;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  88     real             *vftab;
  89     __m128d          dummy_mask,cutoff_mask;
  90     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  91     __m128d          one     = _mm_set1_pd(1.0);
  92     __m128d          two     = _mm_set1_pd(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_pd(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     nvdwtype         = fr->ntype;
 107     vdwparam         = fr->nbfp;
 108     vdwtype          = mdatoms->typeA;
 109
 110     vftab            = kernel_data->table_elec->data;
 111     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 112
 113     /* Setup water-specific parameters */
 114     inr              = nlist->iinr[0];
 115     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 116     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 117     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 118     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 144                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 145
 146         fix0             = _mm_setzero_pd();
 147         fiy0             = _mm_setzero_pd();
 148         fiz0             = _mm_setzero_pd();
 149         fix1             = _mm_setzero_pd();
 150         fiy1             = _mm_setzero_pd();
 151         fiz1             = _mm_setzero_pd();
 152         fix2             = _mm_setzero_pd();
 153         fiy2             = _mm_setzero_pd();
 154         fiz2             = _mm_setzero_pd();
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_pd();
 158         vvdwsum          = _mm_setzero_pd();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169
 170             /* load j atom coordinates */
 171             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_pd(ix0,jx0);
 176             dy00             = _mm_sub_pd(iy0,jy0);
 177             dz00             = _mm_sub_pd(iz0,jz0);
 178             dx10             = _mm_sub_pd(ix1,jx0);
 179             dy10             = _mm_sub_pd(iy1,jy0);
 180             dz10             = _mm_sub_pd(iz1,jz0);
 181             dx20             = _mm_sub_pd(ix2,jx0);
 182             dy20             = _mm_sub_pd(iy2,jy0);
 183             dz20             = _mm_sub_pd(iz2,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 187             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 188             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 189
 190             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 191             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 192             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 193
 194             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 198             vdwjidx0A        = 2*vdwtype[jnrA+0];
 199             vdwjidx0B        = 2*vdwtype[jnrB+0];
 200
 201             fjx0             = _mm_setzero_pd();
 202             fjy0             = _mm_setzero_pd();
 203             fjz0             = _mm_setzero_pd();
 204
 205             /**************************
 206              * CALCULATE INTERACTIONS *
 207              **************************/
 208
 209             r00              = _mm_mul_pd(rsq00,rinv00);
 210
 211             /* Compute parameters for interactions between i and j atoms */
 212             qq00             = _mm_mul_pd(iq0,jq0);
 213             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 214                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 215
 216             /* Calculate table index by multiplying r with table scale and truncate to integer */
 217             rt               = _mm_mul_pd(r00,vftabscale);
 218             vfitab           = _mm_cvttpd_epi32(rt);
 219 #ifdef __XOP__
 220             vfeps            = _mm_frcz_pd(rt);
 221 #else
 222             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 223 #endif
 224             twovfeps         = _mm_add_pd(vfeps,vfeps);
 225             vfitab           = _mm_slli_epi32(vfitab,2);
 226
 227             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 228             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 229             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 230             GMX_MM_TRANSPOSE2_PD(Y,F);
 231             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 232             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 233             GMX_MM_TRANSPOSE2_PD(G,H);
 234             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 235             VV               = _mm_macc_pd(vfeps,Fp,Y);
 236             velec            = _mm_mul_pd(qq00,VV);
 237             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 238             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 239
 240             /* LENNARD-JONES DISPERSION/REPULSION */
 241
 242             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 243             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 244             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 245             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 246             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             velecsum         = _mm_add_pd(velecsum,velec);
 250             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 251
 252             fscal            = _mm_add_pd(felec,fvdw);
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 256             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 257             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 258
 259             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 260             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 261             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r10              = _mm_mul_pd(rsq10,rinv10);
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq10             = _mm_mul_pd(iq1,jq0);
 271
 272             /* Calculate table index by multiplying r with table scale and truncate to integer */
 273             rt               = _mm_mul_pd(r10,vftabscale);
 274             vfitab           = _mm_cvttpd_epi32(rt);
 275 #ifdef __XOP__
 276             vfeps            = _mm_frcz_pd(rt);
 277 #else
 278             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 279 #endif
 280             twovfeps         = _mm_add_pd(vfeps,vfeps);
 281             vfitab           = _mm_slli_epi32(vfitab,2);
 282
 283             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 284             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 285             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 286             GMX_MM_TRANSPOSE2_PD(Y,F);
 287             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 288             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 289             GMX_MM_TRANSPOSE2_PD(G,H);
 290             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 291             VV               = _mm_macc_pd(vfeps,Fp,Y);
 292             velec            = _mm_mul_pd(qq10,VV);
 293             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 294             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velecsum         = _mm_add_pd(velecsum,velec);
 298
 299             fscal            = felec;
 300
 301             /* Update vectorial force */
 302             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 303             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 304             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 305
 306             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 307             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 308             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             r20              = _mm_mul_pd(rsq20,rinv20);
 315
 316             /* Compute parameters for interactions between i and j atoms */
 317             qq20             = _mm_mul_pd(iq2,jq0);
 318
 319             /* Calculate table index by multiplying r with table scale and truncate to integer */
 320             rt               = _mm_mul_pd(r20,vftabscale);
 321             vfitab           = _mm_cvttpd_epi32(rt);
 322 #ifdef __XOP__
 323             vfeps            = _mm_frcz_pd(rt);
 324 #else
 325             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 326 #endif
 327             twovfeps         = _mm_add_pd(vfeps,vfeps);
 328             vfitab           = _mm_slli_epi32(vfitab,2);
 329
 330             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 331             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 332             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 333             GMX_MM_TRANSPOSE2_PD(Y,F);
 334             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 335             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 336             GMX_MM_TRANSPOSE2_PD(G,H);
 337             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 338             VV               = _mm_macc_pd(vfeps,Fp,Y);
 339             velec            = _mm_mul_pd(qq20,VV);
 340             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 341             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_pd(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Update vectorial force */
 349             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 350             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 351             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 352
 353             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 354             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 355             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 356
 357             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 358
 359             /* Inner loop uses 154 flops */
 360         }
 361
 362         if(jidx<j_index_end)
 363         {
 364
 365             jnrA             = jjnr[jidx];
 366             j_coord_offsetA  = DIM*jnrA;
 367
 368             /* load j atom coordinates */
 369             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 370                                               &jx0,&jy0,&jz0);
 371
 372             /* Calculate displacement vector */
 373             dx00             = _mm_sub_pd(ix0,jx0);
 374             dy00             = _mm_sub_pd(iy0,jy0);
 375             dz00             = _mm_sub_pd(iz0,jz0);
 376             dx10             = _mm_sub_pd(ix1,jx0);
 377             dy10             = _mm_sub_pd(iy1,jy0);
 378             dz10             = _mm_sub_pd(iz1,jz0);
 379             dx20             = _mm_sub_pd(ix2,jx0);
 380             dy20             = _mm_sub_pd(iy2,jy0);
 381             dz20             = _mm_sub_pd(iz2,jz0);
 382
 383             /* Calculate squared distance and things based on it */
 384             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 385             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 386             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 387
 388             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 389             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 390             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 391
 392             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 393
 394             /* Load parameters for j particles */
 395             jq0              = _mm_load_sd(charge+jnrA+0);
 396             vdwjidx0A        = 2*vdwtype[jnrA+0];
 397
 398             fjx0             = _mm_setzero_pd();
 399             fjy0             = _mm_setzero_pd();
 400             fjz0             = _mm_setzero_pd();
 401
 402             /**************************
 403              * CALCULATE INTERACTIONS *
 404              **************************/
 405
 406             r00              = _mm_mul_pd(rsq00,rinv00);
 407
 408             /* Compute parameters for interactions between i and j atoms */
 409             qq00             = _mm_mul_pd(iq0,jq0);
 410             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 411
 412             /* Calculate table index by multiplying r with table scale and truncate to integer */
 413             rt               = _mm_mul_pd(r00,vftabscale);
 414             vfitab           = _mm_cvttpd_epi32(rt);
 415 #ifdef __XOP__
 416             vfeps            = _mm_frcz_pd(rt);
 417 #else
 418             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 419 #endif
 420             twovfeps         = _mm_add_pd(vfeps,vfeps);
 421             vfitab           = _mm_slli_epi32(vfitab,2);
 422
 423             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 424             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 425             F                = _mm_setzero_pd();
 426             GMX_MM_TRANSPOSE2_PD(Y,F);
 427             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 428             H                = _mm_setzero_pd();
 429             GMX_MM_TRANSPOSE2_PD(G,H);
 430             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 431             VV               = _mm_macc_pd(vfeps,Fp,Y);
 432             velec            = _mm_mul_pd(qq00,VV);
 433             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 434             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 435
 436             /* LENNARD-JONES DISPERSION/REPULSION */
 437
 438             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 439             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 440             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 441             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 442             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 443
 444             /* Update potential sum for this i atom from the interaction with this j atom. */
 445             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 446             velecsum         = _mm_add_pd(velecsum,velec);
 447             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 448             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 449
 450             fscal            = _mm_add_pd(felec,fvdw);
 451
 452             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 453
 454             /* Update vectorial force */
 455             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 456             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 457             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 458
 459             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 460             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 461             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             r10              = _mm_mul_pd(rsq10,rinv10);
 468
 469             /* Compute parameters for interactions between i and j atoms */
 470             qq10             = _mm_mul_pd(iq1,jq0);
 471
 472             /* Calculate table index by multiplying r with table scale and truncate to integer */
 473             rt               = _mm_mul_pd(r10,vftabscale);
 474             vfitab           = _mm_cvttpd_epi32(rt);
 475 #ifdef __XOP__
 476             vfeps            = _mm_frcz_pd(rt);
 477 #else
 478             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 479 #endif
 480             twovfeps         = _mm_add_pd(vfeps,vfeps);
 481             vfitab           = _mm_slli_epi32(vfitab,2);
 482
 483             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 484             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 485             F                = _mm_setzero_pd();
 486             GMX_MM_TRANSPOSE2_PD(Y,F);
 487             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 488             H                = _mm_setzero_pd();
 489             GMX_MM_TRANSPOSE2_PD(G,H);
 490             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 491             VV               = _mm_macc_pd(vfeps,Fp,Y);
 492             velec            = _mm_mul_pd(qq10,VV);
 493             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 494             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 495
 496             /* Update potential sum for this i atom from the interaction with this j atom. */
 497             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 498             velecsum         = _mm_add_pd(velecsum,velec);
 499
 500             fscal            = felec;
 501
 502             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 503
 504             /* Update vectorial force */
 505             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 506             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 507             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 508
 509             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 510             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 511             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             r20              = _mm_mul_pd(rsq20,rinv20);
 518
 519             /* Compute parameters for interactions between i and j atoms */
 520             qq20             = _mm_mul_pd(iq2,jq0);
 521
 522             /* Calculate table index by multiplying r with table scale and truncate to integer */
 523             rt               = _mm_mul_pd(r20,vftabscale);
 524             vfitab           = _mm_cvttpd_epi32(rt);
 525 #ifdef __XOP__
 526             vfeps            = _mm_frcz_pd(rt);
 527 #else
 528             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 529 #endif
 530             twovfeps         = _mm_add_pd(vfeps,vfeps);
 531             vfitab           = _mm_slli_epi32(vfitab,2);
 532
 533             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 534             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 535             F                = _mm_setzero_pd();
 536             GMX_MM_TRANSPOSE2_PD(Y,F);
 537             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 538             H                = _mm_setzero_pd();
 539             GMX_MM_TRANSPOSE2_PD(G,H);
 540             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 541             VV               = _mm_macc_pd(vfeps,Fp,Y);
 542             velec            = _mm_mul_pd(qq20,VV);
 543             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 544             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 545
 546             /* Update potential sum for this i atom from the interaction with this j atom. */
 547             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 548             velecsum         = _mm_add_pd(velecsum,velec);
 549
 550             fscal            = felec;
 551
 552             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 553
 554             /* Update vectorial force */
 555             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 556             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 557             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 558
 559             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 560             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 561             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 562
 563             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 564
 565             /* Inner loop uses 154 flops */
 566         }
 567
 568         /* End of innermost loop */
 569
 570         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 571                                               f+i_coord_offset,fshift+i_shift_offset);
 572
 573         ggid                        = gid[iidx];
 574         /* Update potential energies */
 575         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 576         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 577
 578         /* Increment number of inner iterations */
 579         inneriter                  += j_index_end - j_index_start;
 580
 581         /* Outer loop uses 20 flops */
 582     }
 583
 584     /* Increment number of outer iterations */
 585     outeriter        += nri;
 586
 587     /* Update outer/inner flops */
 588
 589     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 590 }
 591 /*
 592  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_double
 593  * Electrostatics interaction: CubicSplineTable
 594  * VdW interaction:            LennardJones
 595  * Geometry:                   Water3-Particle
 596  * Calculate force/pot:        Force
 597  */
 598 void
 599 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_double
 600                     (t_nblist * gmx_restrict                nlist,
 601                      rvec * gmx_restrict                    xx,
 602                      rvec * gmx_restrict                    ff,
 603                      t_forcerec * gmx_restrict              fr,
 604                      t_mdatoms * gmx_restrict               mdatoms,
 605                      nb_kernel_data_t * gmx_restrict        kernel_data,
 606                      t_nrnb * gmx_restrict                  nrnb)
 607 {
 608     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 609      * just 0 for non-waters.
 610      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 611      * jnr indices corresponding to data put in the four positions in the SIMD register.
 612      */
 613     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 614     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 615     int              jnrA,jnrB;
 616     int              j_coord_offsetA,j_coord_offsetB;
 617     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 618     real             rcutoff_scalar;
 619     real             *shiftvec,*fshift,*x,*f;
 620     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 621     int              vdwioffset0;
 622     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 623     int              vdwioffset1;
 624     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 625     int              vdwioffset2;
 626     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 627     int              vdwjidx0A,vdwjidx0B;
 628     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 629     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 630     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 631     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 632     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 633     real             *charge;
 634     int              nvdwtype;
 635     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 636     int              *vdwtype;
 637     real             *vdwparam;
 638     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 639     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 640     __m128i          vfitab;
 641     __m128i          ifour       = _mm_set1_epi32(4);
 642     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 643     real             *vftab;
 644     __m128d          dummy_mask,cutoff_mask;
 645     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 646     __m128d          one     = _mm_set1_pd(1.0);
 647     __m128d          two     = _mm_set1_pd(2.0);
 648     x                = xx[0];
 649     f                = ff[0];
 650
 651     nri              = nlist->nri;
 652     iinr             = nlist->iinr;
 653     jindex           = nlist->jindex;
 654     jjnr             = nlist->jjnr;
 655     shiftidx         = nlist->shift;
 656     gid              = nlist->gid;
 657     shiftvec         = fr->shift_vec[0];
 658     fshift           = fr->fshift[0];
 659     facel            = _mm_set1_pd(fr->epsfac);
 660     charge           = mdatoms->chargeA;
 661     nvdwtype         = fr->ntype;
 662     vdwparam         = fr->nbfp;
 663     vdwtype          = mdatoms->typeA;
 664
 665     vftab            = kernel_data->table_elec->data;
 666     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 667
 668     /* Setup water-specific parameters */
 669     inr              = nlist->iinr[0];
 670     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 671     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 672     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 673     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 674
 675     /* Avoid stupid compiler warnings */
 676     jnrA = jnrB = 0;
 677     j_coord_offsetA = 0;
 678     j_coord_offsetB = 0;
 679
 680     outeriter        = 0;
 681     inneriter        = 0;
 682
 683     /* Start outer loop over neighborlists */
 684     for(iidx=0; iidx<nri; iidx++)
 685     {
 686         /* Load shift vector for this list */
 687         i_shift_offset   = DIM*shiftidx[iidx];
 688
 689         /* Load limits for loop over neighbors */
 690         j_index_start    = jindex[iidx];
 691         j_index_end      = jindex[iidx+1];
 692
 693         /* Get outer coordinate index */
 694         inr              = iinr[iidx];
 695         i_coord_offset   = DIM*inr;
 696
 697         /* Load i particle coords and add shift vector */
 698         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 699                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 700
 701         fix0             = _mm_setzero_pd();
 702         fiy0             = _mm_setzero_pd();
 703         fiz0             = _mm_setzero_pd();
 704         fix1             = _mm_setzero_pd();
 705         fiy1             = _mm_setzero_pd();
 706         fiz1             = _mm_setzero_pd();
 707         fix2             = _mm_setzero_pd();
 708         fiy2             = _mm_setzero_pd();
 709         fiz2             = _mm_setzero_pd();
 710
 711         /* Start inner kernel loop */
 712         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 713         {
 714
 715             /* Get j neighbor index, and coordinate index */
 716             jnrA             = jjnr[jidx];
 717             jnrB             = jjnr[jidx+1];
 718             j_coord_offsetA  = DIM*jnrA;
 719             j_coord_offsetB  = DIM*jnrB;
 720
 721             /* load j atom coordinates */
 722             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 723                                               &jx0,&jy0,&jz0);
 724
 725             /* Calculate displacement vector */
 726             dx00             = _mm_sub_pd(ix0,jx0);
 727             dy00             = _mm_sub_pd(iy0,jy0);
 728             dz00             = _mm_sub_pd(iz0,jz0);
 729             dx10             = _mm_sub_pd(ix1,jx0);
 730             dy10             = _mm_sub_pd(iy1,jy0);
 731             dz10             = _mm_sub_pd(iz1,jz0);
 732             dx20             = _mm_sub_pd(ix2,jx0);
 733             dy20             = _mm_sub_pd(iy2,jy0);
 734             dz20             = _mm_sub_pd(iz2,jz0);
 735
 736             /* Calculate squared distance and things based on it */
 737             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 738             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 739             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 740
 741             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 742             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 743             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 744
 745             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 746
 747             /* Load parameters for j particles */
 748             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 749             vdwjidx0A        = 2*vdwtype[jnrA+0];
 750             vdwjidx0B        = 2*vdwtype[jnrB+0];
 751
 752             fjx0             = _mm_setzero_pd();
 753             fjy0             = _mm_setzero_pd();
 754             fjz0             = _mm_setzero_pd();
 755
 756             /**************************
 757              * CALCULATE INTERACTIONS *
 758              **************************/
 759
 760             r00              = _mm_mul_pd(rsq00,rinv00);
 761
 762             /* Compute parameters for interactions between i and j atoms */
 763             qq00             = _mm_mul_pd(iq0,jq0);
 764             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 765                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 766
 767             /* Calculate table index by multiplying r with table scale and truncate to integer */
 768             rt               = _mm_mul_pd(r00,vftabscale);
 769             vfitab           = _mm_cvttpd_epi32(rt);
 770 #ifdef __XOP__
 771             vfeps            = _mm_frcz_pd(rt);
 772 #else
 773             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 774 #endif
 775             twovfeps         = _mm_add_pd(vfeps,vfeps);
 776             vfitab           = _mm_slli_epi32(vfitab,2);
 777
 778             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 779             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 780             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 781             GMX_MM_TRANSPOSE2_PD(Y,F);
 782             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 783             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 784             GMX_MM_TRANSPOSE2_PD(G,H);
 785             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 786             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 787             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 788
 789             /* LENNARD-JONES DISPERSION/REPULSION */
 790
 791             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 792             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 793
 794             fscal            = _mm_add_pd(felec,fvdw);
 795
 796             /* Update vectorial force */
 797             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 798             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 799             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 800
 801             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 802             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 803             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 804
 805             /**************************
 806              * CALCULATE INTERACTIONS *
 807              **************************/
 808
 809             r10              = _mm_mul_pd(rsq10,rinv10);
 810
 811             /* Compute parameters for interactions between i and j atoms */
 812             qq10             = _mm_mul_pd(iq1,jq0);
 813
 814             /* Calculate table index by multiplying r with table scale and truncate to integer */
 815             rt               = _mm_mul_pd(r10,vftabscale);
 816             vfitab           = _mm_cvttpd_epi32(rt);
 817 #ifdef __XOP__
 818             vfeps            = _mm_frcz_pd(rt);
 819 #else
 820             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 821 #endif
 822             twovfeps         = _mm_add_pd(vfeps,vfeps);
 823             vfitab           = _mm_slli_epi32(vfitab,2);
 824
 825             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 826             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 827             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 828             GMX_MM_TRANSPOSE2_PD(Y,F);
 829             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 830             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 831             GMX_MM_TRANSPOSE2_PD(G,H);
 832             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 833             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 834             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 835
 836             fscal            = felec;
 837
 838             /* Update vectorial force */
 839             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 840             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 841             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 842
 843             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 844             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 845             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 846
 847             /**************************
 848              * CALCULATE INTERACTIONS *
 849              **************************/
 850
 851             r20              = _mm_mul_pd(rsq20,rinv20);
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             qq20             = _mm_mul_pd(iq2,jq0);
 855
 856             /* Calculate table index by multiplying r with table scale and truncate to integer */
 857             rt               = _mm_mul_pd(r20,vftabscale);
 858             vfitab           = _mm_cvttpd_epi32(rt);
 859 #ifdef __XOP__
 860             vfeps            = _mm_frcz_pd(rt);
 861 #else
 862             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 863 #endif
 864             twovfeps         = _mm_add_pd(vfeps,vfeps);
 865             vfitab           = _mm_slli_epi32(vfitab,2);
 866
 867             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 868             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 869             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 870             GMX_MM_TRANSPOSE2_PD(Y,F);
 871             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 872             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 873             GMX_MM_TRANSPOSE2_PD(G,H);
 874             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 875             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 876             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 877
 878             fscal            = felec;
 879
 880             /* Update vectorial force */
 881             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 882             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 883             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 884
 885             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 886             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 887             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 888
 889             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 890
 891             /* Inner loop uses 137 flops */
 892         }
 893
 894         if(jidx<j_index_end)
 895         {
 896
 897             jnrA             = jjnr[jidx];
 898             j_coord_offsetA  = DIM*jnrA;
 899
 900             /* load j atom coordinates */
 901             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 902                                               &jx0,&jy0,&jz0);
 903
 904             /* Calculate displacement vector */
 905             dx00             = _mm_sub_pd(ix0,jx0);
 906             dy00             = _mm_sub_pd(iy0,jy0);
 907             dz00             = _mm_sub_pd(iz0,jz0);
 908             dx10             = _mm_sub_pd(ix1,jx0);
 909             dy10             = _mm_sub_pd(iy1,jy0);
 910             dz10             = _mm_sub_pd(iz1,jz0);
 911             dx20             = _mm_sub_pd(ix2,jx0);
 912             dy20             = _mm_sub_pd(iy2,jy0);
 913             dz20             = _mm_sub_pd(iz2,jz0);
 914
 915             /* Calculate squared distance and things based on it */
 916             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 917             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 918             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 919
 920             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 921             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 922             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 923
 924             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 925
 926             /* Load parameters for j particles */
 927             jq0              = _mm_load_sd(charge+jnrA+0);
 928             vdwjidx0A        = 2*vdwtype[jnrA+0];
 929
 930             fjx0             = _mm_setzero_pd();
 931             fjy0             = _mm_setzero_pd();
 932             fjz0             = _mm_setzero_pd();
 933
 934             /**************************
 935              * CALCULATE INTERACTIONS *
 936              **************************/
 937
 938             r00              = _mm_mul_pd(rsq00,rinv00);
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             qq00             = _mm_mul_pd(iq0,jq0);
 942             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 943
 944             /* Calculate table index by multiplying r with table scale and truncate to integer */
 945             rt               = _mm_mul_pd(r00,vftabscale);
 946             vfitab           = _mm_cvttpd_epi32(rt);
 947 #ifdef __XOP__
 948             vfeps            = _mm_frcz_pd(rt);
 949 #else
 950             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 951 #endif
 952             twovfeps         = _mm_add_pd(vfeps,vfeps);
 953             vfitab           = _mm_slli_epi32(vfitab,2);
 954
 955             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 956             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 957             F                = _mm_setzero_pd();
 958             GMX_MM_TRANSPOSE2_PD(Y,F);
 959             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 960             H                = _mm_setzero_pd();
 961             GMX_MM_TRANSPOSE2_PD(G,H);
 962             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 963             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 964             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 965
 966             /* LENNARD-JONES DISPERSION/REPULSION */
 967
 968             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 969             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 970
 971             fscal            = _mm_add_pd(felec,fvdw);
 972
 973             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 974
 975             /* Update vectorial force */
 976             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 977             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 978             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 979
 980             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 981             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 982             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             r10              = _mm_mul_pd(rsq10,rinv10);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq10             = _mm_mul_pd(iq1,jq0);
 992
 993             /* Calculate table index by multiplying r with table scale and truncate to integer */
 994             rt               = _mm_mul_pd(r10,vftabscale);
 995             vfitab           = _mm_cvttpd_epi32(rt);
 996 #ifdef __XOP__
 997             vfeps            = _mm_frcz_pd(rt);
 998 #else
 999             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1000 #endif
1001             twovfeps         = _mm_add_pd(vfeps,vfeps);
1002             vfitab           = _mm_slli_epi32(vfitab,2);
1003
1004             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1005             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1006             F                = _mm_setzero_pd();
1007             GMX_MM_TRANSPOSE2_PD(Y,F);
1008             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1009             H                = _mm_setzero_pd();
1010             GMX_MM_TRANSPOSE2_PD(G,H);
1011             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1012             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1013             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1014
1015             fscal            = felec;
1016
1017             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1018
1019             /* Update vectorial force */
1020             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1021             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1022             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1023
1024             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1025             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1026             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             r20              = _mm_mul_pd(rsq20,rinv20);
1033
1034             /* Compute parameters for interactions between i and j atoms */
1035             qq20             = _mm_mul_pd(iq2,jq0);
1036
1037             /* Calculate table index by multiplying r with table scale and truncate to integer */
1038             rt               = _mm_mul_pd(r20,vftabscale);
1039             vfitab           = _mm_cvttpd_epi32(rt);
1040 #ifdef __XOP__
1041             vfeps            = _mm_frcz_pd(rt);
1042 #else
1043             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1044 #endif
1045             twovfeps         = _mm_add_pd(vfeps,vfeps);
1046             vfitab           = _mm_slli_epi32(vfitab,2);
1047
1048             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1049             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1050             F                = _mm_setzero_pd();
1051             GMX_MM_TRANSPOSE2_PD(Y,F);
1052             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1053             H                = _mm_setzero_pd();
1054             GMX_MM_TRANSPOSE2_PD(G,H);
1055             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1056             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1057             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1058
1059             fscal            = felec;
1060
1061             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1062
1063             /* Update vectorial force */
1064             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1065             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1066             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1067
1068             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1069             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1070             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1071
1072             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1073
1074             /* Inner loop uses 137 flops */
1075         }
1076
1077         /* End of innermost loop */
1078
1079         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1080                                               f+i_coord_offset,fshift+i_shift_offset);
1081
1082         /* Increment number of inner iterations */
1083         inneriter                  += j_index_end - j_index_start;
1084
1085         /* Outer loop uses 18 flops */
1086     }
1087
1088     /* Increment number of outer iterations */
1089     outeriter        += nri;
1090
1091     /* Update outer/inner flops */
1092
1093     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1094 }