src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwLJ_GeomW4P1_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_ElecCoul_VdwLJ_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_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              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128d          dummy_mask,cutoff_mask;
  89     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  90     __m128d          one     = _mm_set1_pd(1.0);
  91     __m128d          two     = _mm_set1_pd(2.0);
  92     x                = xx[0];
  93     f                = ff[0];
  94
  95     nri              = nlist->nri;
  96     iinr             = nlist->iinr;
  97     jindex           = nlist->jindex;
  98     jjnr             = nlist->jjnr;
  99     shiftidx         = nlist->shift;
 100     gid              = nlist->gid;
 101     shiftvec         = fr->shift_vec[0];
 102     fshift           = fr->fshift[0];
 103     facel            = _mm_set1_pd(fr->epsfac);
 104     charge           = mdatoms->chargeA;
 105     nvdwtype         = fr->ntype;
 106     vdwparam         = fr->nbfp;
 107     vdwtype          = mdatoms->typeA;
 108
 109     /* Setup water-specific parameters */
 110     inr              = nlist->iinr[0];
 111     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 112     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 113     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 114     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 140                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 141
 142         fix0             = _mm_setzero_pd();
 143         fiy0             = _mm_setzero_pd();
 144         fiz0             = _mm_setzero_pd();
 145         fix1             = _mm_setzero_pd();
 146         fiy1             = _mm_setzero_pd();
 147         fiz1             = _mm_setzero_pd();
 148         fix2             = _mm_setzero_pd();
 149         fiy2             = _mm_setzero_pd();
 150         fiz2             = _mm_setzero_pd();
 151         fix3             = _mm_setzero_pd();
 152         fiy3             = _mm_setzero_pd();
 153         fiz3             = _mm_setzero_pd();
 154
 155         /* Reset potential sums */
 156         velecsum         = _mm_setzero_pd();
 157         vvdwsum          = _mm_setzero_pd();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _mm_sub_pd(ix0,jx0);
 175             dy00             = _mm_sub_pd(iy0,jy0);
 176             dz00             = _mm_sub_pd(iz0,jz0);
 177             dx10             = _mm_sub_pd(ix1,jx0);
 178             dy10             = _mm_sub_pd(iy1,jy0);
 179             dz10             = _mm_sub_pd(iz1,jz0);
 180             dx20             = _mm_sub_pd(ix2,jx0);
 181             dy20             = _mm_sub_pd(iy2,jy0);
 182             dz20             = _mm_sub_pd(iz2,jz0);
 183             dx30             = _mm_sub_pd(ix3,jx0);
 184             dy30             = _mm_sub_pd(iy3,jy0);
 185             dz30             = _mm_sub_pd(iz3,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 189             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 190             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 191             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 192
 193             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 194             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 195             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 196
 197             rinvsq00         = gmx_mm_inv_pd(rsq00);
 198             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 199             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 200             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 201
 202             /* Load parameters for j particles */
 203             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 204             vdwjidx0A        = 2*vdwtype[jnrA+0];
 205             vdwjidx0B        = 2*vdwtype[jnrB+0];
 206
 207             fjx0             = _mm_setzero_pd();
 208             fjy0             = _mm_setzero_pd();
 209             fjz0             = _mm_setzero_pd();
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             /* Compute parameters for interactions between i and j atoms */
 216             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 217                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 218
 219             /* LENNARD-JONES DISPERSION/REPULSION */
 220
 221             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 222             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 223             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 224             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 225             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 229
 230             fscal            = fvdw;
 231
 232             /* Update vectorial force */
 233             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 234             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 235             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 236
 237             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 238             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 239             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq10             = _mm_mul_pd(iq1,jq0);
 247
 248             /* COULOMB ELECTROSTATICS */
 249             velec            = _mm_mul_pd(qq10,rinv10);
 250             felec            = _mm_mul_pd(velec,rinvsq10);
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             velecsum         = _mm_add_pd(velecsum,velec);
 254
 255             fscal            = felec;
 256
 257             /* Update vectorial force */
 258             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 259             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 260             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 261
 262             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 263             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 264             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq20             = _mm_mul_pd(iq2,jq0);
 272
 273             /* COULOMB ELECTROSTATICS */
 274             velec            = _mm_mul_pd(qq20,rinv20);
 275             felec            = _mm_mul_pd(velec,rinvsq20);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velecsum         = _mm_add_pd(velecsum,velec);
 279
 280             fscal            = felec;
 281
 282             /* Update vectorial force */
 283             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 284             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 285             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 286
 287             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 288             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 289             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             /* Compute parameters for interactions between i and j atoms */
 296             qq30             = _mm_mul_pd(iq3,jq0);
 297
 298             /* COULOMB ELECTROSTATICS */
 299             velec            = _mm_mul_pd(qq30,rinv30);
 300             felec            = _mm_mul_pd(velec,rinvsq30);
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velecsum         = _mm_add_pd(velecsum,velec);
 304
 305             fscal            = felec;
 306
 307             /* Update vectorial force */
 308             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 309             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 310             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 311
 312             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 313             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 314             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 315
 316             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 317
 318             /* Inner loop uses 131 flops */
 319         }
 320
 321         if(jidx<j_index_end)
 322         {
 323
 324             jnrA             = jjnr[jidx];
 325             j_coord_offsetA  = DIM*jnrA;
 326
 327             /* load j atom coordinates */
 328             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 329                                               &jx0,&jy0,&jz0);
 330
 331             /* Calculate displacement vector */
 332             dx00             = _mm_sub_pd(ix0,jx0);
 333             dy00             = _mm_sub_pd(iy0,jy0);
 334             dz00             = _mm_sub_pd(iz0,jz0);
 335             dx10             = _mm_sub_pd(ix1,jx0);
 336             dy10             = _mm_sub_pd(iy1,jy0);
 337             dz10             = _mm_sub_pd(iz1,jz0);
 338             dx20             = _mm_sub_pd(ix2,jx0);
 339             dy20             = _mm_sub_pd(iy2,jy0);
 340             dz20             = _mm_sub_pd(iz2,jz0);
 341             dx30             = _mm_sub_pd(ix3,jx0);
 342             dy30             = _mm_sub_pd(iy3,jy0);
 343             dz30             = _mm_sub_pd(iz3,jz0);
 344
 345             /* Calculate squared distance and things based on it */
 346             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 347             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 348             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 349             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 350
 351             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 352             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 353             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 354
 355             rinvsq00         = gmx_mm_inv_pd(rsq00);
 356             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 357             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 358             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 359
 360             /* Load parameters for j particles */
 361             jq0              = _mm_load_sd(charge+jnrA+0);
 362             vdwjidx0A        = 2*vdwtype[jnrA+0];
 363
 364             fjx0             = _mm_setzero_pd();
 365             fjy0             = _mm_setzero_pd();
 366             fjz0             = _mm_setzero_pd();
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 374
 375             /* LENNARD-JONES DISPERSION/REPULSION */
 376
 377             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 378             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 379             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 380             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 381             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 382
 383             /* Update potential sum for this i atom from the interaction with this j atom. */
 384             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 385             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 386
 387             fscal            = fvdw;
 388
 389             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 390
 391             /* Update vectorial force */
 392             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 393             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 394             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 395
 396             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 397             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 398             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 399
 400             /**************************
 401              * CALCULATE INTERACTIONS *
 402              **************************/
 403
 404             /* Compute parameters for interactions between i and j atoms */
 405             qq10             = _mm_mul_pd(iq1,jq0);
 406
 407             /* COULOMB ELECTROSTATICS */
 408             velec            = _mm_mul_pd(qq10,rinv10);
 409             felec            = _mm_mul_pd(velec,rinvsq10);
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 413             velecsum         = _mm_add_pd(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 418
 419             /* Update vectorial force */
 420             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 421             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 422             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 423
 424             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 425             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 426             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             qq20             = _mm_mul_pd(iq2,jq0);
 434
 435             /* COULOMB ELECTROSTATICS */
 436             velec            = _mm_mul_pd(qq20,rinv20);
 437             felec            = _mm_mul_pd(velec,rinvsq20);
 438
 439             /* Update potential sum for this i atom from the interaction with this j atom. */
 440             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 441             velecsum         = _mm_add_pd(velecsum,velec);
 442
 443             fscal            = felec;
 444
 445             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 446
 447             /* Update vectorial force */
 448             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 449             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 450             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 451
 452             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 453             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 454             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             /* Compute parameters for interactions between i and j atoms */
 461             qq30             = _mm_mul_pd(iq3,jq0);
 462
 463             /* COULOMB ELECTROSTATICS */
 464             velec            = _mm_mul_pd(qq30,rinv30);
 465             felec            = _mm_mul_pd(velec,rinvsq30);
 466
 467             /* Update potential sum for this i atom from the interaction with this j atom. */
 468             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 469             velecsum         = _mm_add_pd(velecsum,velec);
 470
 471             fscal            = felec;
 472
 473             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 474
 475             /* Update vectorial force */
 476             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 477             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 478             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 479
 480             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 481             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 482             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 483
 484             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 485
 486             /* Inner loop uses 131 flops */
 487         }
 488
 489         /* End of innermost loop */
 490
 491         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 492                                               f+i_coord_offset,fshift+i_shift_offset);
 493
 494         ggid                        = gid[iidx];
 495         /* Update potential energies */
 496         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 497         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 498
 499         /* Increment number of inner iterations */
 500         inneriter                  += j_index_end - j_index_start;
 501
 502         /* Outer loop uses 26 flops */
 503     }
 504
 505     /* Increment number of outer iterations */
 506     outeriter        += nri;
 507
 508     /* Update outer/inner flops */
 509
 510     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*131);
 511 }
 512 /*
 513  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_128_fma_double
 514  * Electrostatics interaction: Coulomb
 515  * VdW interaction:            LennardJones
 516  * Geometry:                   Water4-Particle
 517  * Calculate force/pot:        Force
 518  */
 519 void
 520 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_128_fma_double
 521                     (t_nblist * gmx_restrict                nlist,
 522                      rvec * gmx_restrict                    xx,
 523                      rvec * gmx_restrict                    ff,
 524                      t_forcerec * gmx_restrict              fr,
 525                      t_mdatoms * gmx_restrict               mdatoms,
 526                      nb_kernel_data_t * gmx_restrict        kernel_data,
 527                      t_nrnb * gmx_restrict                  nrnb)
 528 {
 529     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 530      * just 0 for non-waters.
 531      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 532      * jnr indices corresponding to data put in the four positions in the SIMD register.
 533      */
 534     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 535     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 536     int              jnrA,jnrB;
 537     int              j_coord_offsetA,j_coord_offsetB;
 538     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 539     real             rcutoff_scalar;
 540     real             *shiftvec,*fshift,*x,*f;
 541     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 542     int              vdwioffset0;
 543     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 544     int              vdwioffset1;
 545     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 546     int              vdwioffset2;
 547     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 548     int              vdwioffset3;
 549     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 550     int              vdwjidx0A,vdwjidx0B;
 551     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 552     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 553     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 554     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 555     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 556     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 557     real             *charge;
 558     int              nvdwtype;
 559     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 560     int              *vdwtype;
 561     real             *vdwparam;
 562     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 563     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 564     __m128d          dummy_mask,cutoff_mask;
 565     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 566     __m128d          one     = _mm_set1_pd(1.0);
 567     __m128d          two     = _mm_set1_pd(2.0);
 568     x                = xx[0];
 569     f                = ff[0];
 570
 571     nri              = nlist->nri;
 572     iinr             = nlist->iinr;
 573     jindex           = nlist->jindex;
 574     jjnr             = nlist->jjnr;
 575     shiftidx         = nlist->shift;
 576     gid              = nlist->gid;
 577     shiftvec         = fr->shift_vec[0];
 578     fshift           = fr->fshift[0];
 579     facel            = _mm_set1_pd(fr->epsfac);
 580     charge           = mdatoms->chargeA;
 581     nvdwtype         = fr->ntype;
 582     vdwparam         = fr->nbfp;
 583     vdwtype          = mdatoms->typeA;
 584
 585     /* Setup water-specific parameters */
 586     inr              = nlist->iinr[0];
 587     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 588     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 589     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 590     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 591
 592     /* Avoid stupid compiler warnings */
 593     jnrA = jnrB = 0;
 594     j_coord_offsetA = 0;
 595     j_coord_offsetB = 0;
 596
 597     outeriter        = 0;
 598     inneriter        = 0;
 599
 600     /* Start outer loop over neighborlists */
 601     for(iidx=0; iidx<nri; iidx++)
 602     {
 603         /* Load shift vector for this list */
 604         i_shift_offset   = DIM*shiftidx[iidx];
 605
 606         /* Load limits for loop over neighbors */
 607         j_index_start    = jindex[iidx];
 608         j_index_end      = jindex[iidx+1];
 609
 610         /* Get outer coordinate index */
 611         inr              = iinr[iidx];
 612         i_coord_offset   = DIM*inr;
 613
 614         /* Load i particle coords and add shift vector */
 615         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 616                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 617
 618         fix0             = _mm_setzero_pd();
 619         fiy0             = _mm_setzero_pd();
 620         fiz0             = _mm_setzero_pd();
 621         fix1             = _mm_setzero_pd();
 622         fiy1             = _mm_setzero_pd();
 623         fiz1             = _mm_setzero_pd();
 624         fix2             = _mm_setzero_pd();
 625         fiy2             = _mm_setzero_pd();
 626         fiz2             = _mm_setzero_pd();
 627         fix3             = _mm_setzero_pd();
 628         fiy3             = _mm_setzero_pd();
 629         fiz3             = _mm_setzero_pd();
 630
 631         /* Start inner kernel loop */
 632         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 633         {
 634
 635             /* Get j neighbor index, and coordinate index */
 636             jnrA             = jjnr[jidx];
 637             jnrB             = jjnr[jidx+1];
 638             j_coord_offsetA  = DIM*jnrA;
 639             j_coord_offsetB  = DIM*jnrB;
 640
 641             /* load j atom coordinates */
 642             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 643                                               &jx0,&jy0,&jz0);
 644
 645             /* Calculate displacement vector */
 646             dx00             = _mm_sub_pd(ix0,jx0);
 647             dy00             = _mm_sub_pd(iy0,jy0);
 648             dz00             = _mm_sub_pd(iz0,jz0);
 649             dx10             = _mm_sub_pd(ix1,jx0);
 650             dy10             = _mm_sub_pd(iy1,jy0);
 651             dz10             = _mm_sub_pd(iz1,jz0);
 652             dx20             = _mm_sub_pd(ix2,jx0);
 653             dy20             = _mm_sub_pd(iy2,jy0);
 654             dz20             = _mm_sub_pd(iz2,jz0);
 655             dx30             = _mm_sub_pd(ix3,jx0);
 656             dy30             = _mm_sub_pd(iy3,jy0);
 657             dz30             = _mm_sub_pd(iz3,jz0);
 658
 659             /* Calculate squared distance and things based on it */
 660             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 661             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 662             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 663             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 664
 665             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 666             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 667             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 668
 669             rinvsq00         = gmx_mm_inv_pd(rsq00);
 670             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 671             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 672             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 673
 674             /* Load parameters for j particles */
 675             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 676             vdwjidx0A        = 2*vdwtype[jnrA+0];
 677             vdwjidx0B        = 2*vdwtype[jnrB+0];
 678
 679             fjx0             = _mm_setzero_pd();
 680             fjy0             = _mm_setzero_pd();
 681             fjz0             = _mm_setzero_pd();
 682
 683             /**************************
 684              * CALCULATE INTERACTIONS *
 685              **************************/
 686
 687             /* Compute parameters for interactions between i and j atoms */
 688             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 689                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 690
 691             /* LENNARD-JONES DISPERSION/REPULSION */
 692
 693             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 694             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 695
 696             fscal            = fvdw;
 697
 698             /* Update vectorial force */
 699             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 700             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 701             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 702
 703             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 704             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 705             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             /* Compute parameters for interactions between i and j atoms */
 712             qq10             = _mm_mul_pd(iq1,jq0);
 713
 714             /* COULOMB ELECTROSTATICS */
 715             velec            = _mm_mul_pd(qq10,rinv10);
 716             felec            = _mm_mul_pd(velec,rinvsq10);
 717
 718             fscal            = felec;
 719
 720             /* Update vectorial force */
 721             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 722             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 723             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 724
 725             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 726             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 727             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 728
 729             /**************************
 730              * CALCULATE INTERACTIONS *
 731              **************************/
 732
 733             /* Compute parameters for interactions between i and j atoms */
 734             qq20             = _mm_mul_pd(iq2,jq0);
 735
 736             /* COULOMB ELECTROSTATICS */
 737             velec            = _mm_mul_pd(qq20,rinv20);
 738             felec            = _mm_mul_pd(velec,rinvsq20);
 739
 740             fscal            = felec;
 741
 742             /* Update vectorial force */
 743             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 744             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 745             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 746
 747             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 748             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 749             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 750
 751             /**************************
 752              * CALCULATE INTERACTIONS *
 753              **************************/
 754
 755             /* Compute parameters for interactions between i and j atoms */
 756             qq30             = _mm_mul_pd(iq3,jq0);
 757
 758             /* COULOMB ELECTROSTATICS */
 759             velec            = _mm_mul_pd(qq30,rinv30);
 760             felec            = _mm_mul_pd(velec,rinvsq30);
 761
 762             fscal            = felec;
 763
 764             /* Update vectorial force */
 765             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 766             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 767             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 768
 769             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 770             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 771             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 772
 773             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 774
 775             /* Inner loop uses 123 flops */
 776         }
 777
 778         if(jidx<j_index_end)
 779         {
 780
 781             jnrA             = jjnr[jidx];
 782             j_coord_offsetA  = DIM*jnrA;
 783
 784             /* load j atom coordinates */
 785             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 786                                               &jx0,&jy0,&jz0);
 787
 788             /* Calculate displacement vector */
 789             dx00             = _mm_sub_pd(ix0,jx0);
 790             dy00             = _mm_sub_pd(iy0,jy0);
 791             dz00             = _mm_sub_pd(iz0,jz0);
 792             dx10             = _mm_sub_pd(ix1,jx0);
 793             dy10             = _mm_sub_pd(iy1,jy0);
 794             dz10             = _mm_sub_pd(iz1,jz0);
 795             dx20             = _mm_sub_pd(ix2,jx0);
 796             dy20             = _mm_sub_pd(iy2,jy0);
 797             dz20             = _mm_sub_pd(iz2,jz0);
 798             dx30             = _mm_sub_pd(ix3,jx0);
 799             dy30             = _mm_sub_pd(iy3,jy0);
 800             dz30             = _mm_sub_pd(iz3,jz0);
 801
 802             /* Calculate squared distance and things based on it */
 803             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 804             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 805             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 806             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 807
 808             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 809             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 810             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 811
 812             rinvsq00         = gmx_mm_inv_pd(rsq00);
 813             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 814             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 815             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 816
 817             /* Load parameters for j particles */
 818             jq0              = _mm_load_sd(charge+jnrA+0);
 819             vdwjidx0A        = 2*vdwtype[jnrA+0];
 820
 821             fjx0             = _mm_setzero_pd();
 822             fjy0             = _mm_setzero_pd();
 823             fjz0             = _mm_setzero_pd();
 824
 825             /**************************
 826              * CALCULATE INTERACTIONS *
 827              **************************/
 828
 829             /* Compute parameters for interactions between i and j atoms */
 830             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 831
 832             /* LENNARD-JONES DISPERSION/REPULSION */
 833
 834             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 835             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 836
 837             fscal            = fvdw;
 838
 839             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 840
 841             /* Update vectorial force */
 842             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 843             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 844             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 845
 846             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 847             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 848             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             /* Compute parameters for interactions between i and j atoms */
 855             qq10             = _mm_mul_pd(iq1,jq0);
 856
 857             /* COULOMB ELECTROSTATICS */
 858             velec            = _mm_mul_pd(qq10,rinv10);
 859             felec            = _mm_mul_pd(velec,rinvsq10);
 860
 861             fscal            = felec;
 862
 863             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 864
 865             /* Update vectorial force */
 866             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 867             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 868             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 869
 870             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 871             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 872             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 873
 874             /**************************
 875              * CALCULATE INTERACTIONS *
 876              **************************/
 877
 878             /* Compute parameters for interactions between i and j atoms */
 879             qq20             = _mm_mul_pd(iq2,jq0);
 880
 881             /* COULOMB ELECTROSTATICS */
 882             velec            = _mm_mul_pd(qq20,rinv20);
 883             felec            = _mm_mul_pd(velec,rinvsq20);
 884
 885             fscal            = felec;
 886
 887             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 888
 889             /* Update vectorial force */
 890             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 891             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 892             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 893
 894             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 895             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 896             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             qq30             = _mm_mul_pd(iq3,jq0);
 904
 905             /* COULOMB ELECTROSTATICS */
 906             velec            = _mm_mul_pd(qq30,rinv30);
 907             felec            = _mm_mul_pd(velec,rinvsq30);
 908
 909             fscal            = felec;
 910
 911             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 912
 913             /* Update vectorial force */
 914             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 915             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 916             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 917
 918             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 919             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 920             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 921
 922             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 923
 924             /* Inner loop uses 123 flops */
 925         }
 926
 927         /* End of innermost loop */
 928
 929         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 930                                               f+i_coord_offset,fshift+i_shift_offset);
 931
 932         /* Increment number of inner iterations */
 933         inneriter                  += j_index_end - j_index_start;
 934
 935         /* Outer loop uses 24 flops */
 936     }
 937
 938     /* Increment number of outer iterations */
 939     outeriter        += nri;
 940
 941     /* Update outer/inner flops */
 942
 943     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
 944 }