src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwNone_GeomW3P1_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_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_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_256_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_256_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,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  79     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  80     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  81     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  82     real             *charge;
  83     __m256d          dummy_mask,cutoff_mask;
  84     __m128           tmpmask0,tmpmask1;
  85     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
  86     __m256d          one     = _mm256_set1_pd(1.0);
  87     __m256d          two     = _mm256_set1_pd(2.0);
  88     x                = xx[0];
  89     f                = ff[0];
  90
  91     nri              = nlist->nri;
  92     iinr             = nlist->iinr;
  93     jindex           = nlist->jindex;
  94     jjnr             = nlist->jjnr;
  95     shiftidx         = nlist->shift;
  96     gid              = nlist->gid;
  97     shiftvec         = fr->shift_vec[0];
  98     fshift           = fr->fshift[0];
  99     facel            = _mm256_set1_pd(fr->epsfac);
 100     charge           = mdatoms->chargeA;
 101
 102     /* Setup water-specific parameters */
 103     inr              = nlist->iinr[0];
 104     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 105     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 106     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 107
 108     /* Avoid stupid compiler warnings */
 109     jnrA = jnrB = jnrC = jnrD = 0;
 110     j_coord_offsetA = 0;
 111     j_coord_offsetB = 0;
 112     j_coord_offsetC = 0;
 113     j_coord_offsetD = 0;
 114
 115     outeriter        = 0;
 116     inneriter        = 0;
 117
 118     for(iidx=0;iidx<4*DIM;iidx++)
 119     {
 120         scratch[iidx] = 0.0;
 121     }
 122
 123     /* Start outer loop over neighborlists */
 124     for(iidx=0; iidx<nri; iidx++)
 125     {
 126         /* Load shift vector for this list */
 127         i_shift_offset   = DIM*shiftidx[iidx];
 128
 129         /* Load limits for loop over neighbors */
 130         j_index_start    = jindex[iidx];
 131         j_index_end      = jindex[iidx+1];
 132
 133         /* Get outer coordinate index */
 134         inr              = iinr[iidx];
 135         i_coord_offset   = DIM*inr;
 136
 137         /* Load i particle coords and add shift vector */
 138         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 139                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 140
 141         fix0             = _mm256_setzero_pd();
 142         fiy0             = _mm256_setzero_pd();
 143         fiz0             = _mm256_setzero_pd();
 144         fix1             = _mm256_setzero_pd();
 145         fiy1             = _mm256_setzero_pd();
 146         fiz1             = _mm256_setzero_pd();
 147         fix2             = _mm256_setzero_pd();
 148         fiy2             = _mm256_setzero_pd();
 149         fiz2             = _mm256_setzero_pd();
 150
 151         /* Reset potential sums */
 152         velecsum         = _mm256_setzero_pd();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             jnrC             = jjnr[jidx+2];
 162             jnrD             = jjnr[jidx+3];
 163             j_coord_offsetA  = DIM*jnrA;
 164             j_coord_offsetB  = DIM*jnrB;
 165             j_coord_offsetC  = DIM*jnrC;
 166             j_coord_offsetD  = DIM*jnrD;
 167
 168             /* load j atom coordinates */
 169             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 170                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 171                                                  &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _mm256_sub_pd(ix0,jx0);
 175             dy00             = _mm256_sub_pd(iy0,jy0);
 176             dz00             = _mm256_sub_pd(iz0,jz0);
 177             dx10             = _mm256_sub_pd(ix1,jx0);
 178             dy10             = _mm256_sub_pd(iy1,jy0);
 179             dz10             = _mm256_sub_pd(iz1,jz0);
 180             dx20             = _mm256_sub_pd(ix2,jx0);
 181             dy20             = _mm256_sub_pd(iy2,jy0);
 182             dz20             = _mm256_sub_pd(iz2,jz0);
 183
 184             /* Calculate squared distance and things based on it */
 185             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 186             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 187             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 188
 189             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 190             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 191             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 192
 193             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 194             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 195             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 196
 197             /* Load parameters for j particles */
 198             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 199                                                                  charge+jnrC+0,charge+jnrD+0);
 200
 201             fjx0             = _mm256_setzero_pd();
 202             fjy0             = _mm256_setzero_pd();
 203             fjz0             = _mm256_setzero_pd();
 204
 205             /**************************
 206              * CALCULATE INTERACTIONS *
 207              **************************/
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq00             = _mm256_mul_pd(iq0,jq0);
 211
 212             /* COULOMB ELECTROSTATICS */
 213             velec            = _mm256_mul_pd(qq00,rinv00);
 214             felec            = _mm256_mul_pd(velec,rinvsq00);
 215
 216             /* Update potential sum for this i atom from the interaction with this j atom. */
 217             velecsum         = _mm256_add_pd(velecsum,velec);
 218
 219             fscal            = felec;
 220
 221             /* Calculate temporary vectorial force */
 222             tx               = _mm256_mul_pd(fscal,dx00);
 223             ty               = _mm256_mul_pd(fscal,dy00);
 224             tz               = _mm256_mul_pd(fscal,dz00);
 225
 226             /* Update vectorial force */
 227             fix0             = _mm256_add_pd(fix0,tx);
 228             fiy0             = _mm256_add_pd(fiy0,ty);
 229             fiz0             = _mm256_add_pd(fiz0,tz);
 230
 231             fjx0             = _mm256_add_pd(fjx0,tx);
 232             fjy0             = _mm256_add_pd(fjy0,ty);
 233             fjz0             = _mm256_add_pd(fjz0,tz);
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq10             = _mm256_mul_pd(iq1,jq0);
 241
 242             /* COULOMB ELECTROSTATICS */
 243             velec            = _mm256_mul_pd(qq10,rinv10);
 244             felec            = _mm256_mul_pd(velec,rinvsq10);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velecsum         = _mm256_add_pd(velecsum,velec);
 248
 249             fscal            = felec;
 250
 251             /* Calculate temporary vectorial force */
 252             tx               = _mm256_mul_pd(fscal,dx10);
 253             ty               = _mm256_mul_pd(fscal,dy10);
 254             tz               = _mm256_mul_pd(fscal,dz10);
 255
 256             /* Update vectorial force */
 257             fix1             = _mm256_add_pd(fix1,tx);
 258             fiy1             = _mm256_add_pd(fiy1,ty);
 259             fiz1             = _mm256_add_pd(fiz1,tz);
 260
 261             fjx0             = _mm256_add_pd(fjx0,tx);
 262             fjy0             = _mm256_add_pd(fjy0,ty);
 263             fjz0             = _mm256_add_pd(fjz0,tz);
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq20             = _mm256_mul_pd(iq2,jq0);
 271
 272             /* COULOMB ELECTROSTATICS */
 273             velec            = _mm256_mul_pd(qq20,rinv20);
 274             felec            = _mm256_mul_pd(velec,rinvsq20);
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velecsum         = _mm256_add_pd(velecsum,velec);
 278
 279             fscal            = felec;
 280
 281             /* Calculate temporary vectorial force */
 282             tx               = _mm256_mul_pd(fscal,dx20);
 283             ty               = _mm256_mul_pd(fscal,dy20);
 284             tz               = _mm256_mul_pd(fscal,dz20);
 285
 286             /* Update vectorial force */
 287             fix2             = _mm256_add_pd(fix2,tx);
 288             fiy2             = _mm256_add_pd(fiy2,ty);
 289             fiz2             = _mm256_add_pd(fiz2,tz);
 290
 291             fjx0             = _mm256_add_pd(fjx0,tx);
 292             fjy0             = _mm256_add_pd(fjy0,ty);
 293             fjz0             = _mm256_add_pd(fjz0,tz);
 294
 295             fjptrA             = f+j_coord_offsetA;
 296             fjptrB             = f+j_coord_offsetB;
 297             fjptrC             = f+j_coord_offsetC;
 298             fjptrD             = f+j_coord_offsetD;
 299
 300             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 301
 302             /* Inner loop uses 84 flops */
 303         }
 304
 305         if(jidx<j_index_end)
 306         {
 307
 308             /* Get j neighbor index, and coordinate index */
 309             jnrlistA         = jjnr[jidx];
 310             jnrlistB         = jjnr[jidx+1];
 311             jnrlistC         = jjnr[jidx+2];
 312             jnrlistD         = jjnr[jidx+3];
 313             /* Sign of each element will be negative for non-real atoms.
 314              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 315              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 316              */
 317             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 318
 319             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 320             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 321             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 322
 323             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 324             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 325             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 326             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 327             j_coord_offsetA  = DIM*jnrA;
 328             j_coord_offsetB  = DIM*jnrB;
 329             j_coord_offsetC  = DIM*jnrC;
 330             j_coord_offsetD  = DIM*jnrD;
 331
 332             /* load j atom coordinates */
 333             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 334                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 335                                                  &jx0,&jy0,&jz0);
 336
 337             /* Calculate displacement vector */
 338             dx00             = _mm256_sub_pd(ix0,jx0);
 339             dy00             = _mm256_sub_pd(iy0,jy0);
 340             dz00             = _mm256_sub_pd(iz0,jz0);
 341             dx10             = _mm256_sub_pd(ix1,jx0);
 342             dy10             = _mm256_sub_pd(iy1,jy0);
 343             dz10             = _mm256_sub_pd(iz1,jz0);
 344             dx20             = _mm256_sub_pd(ix2,jx0);
 345             dy20             = _mm256_sub_pd(iy2,jy0);
 346             dz20             = _mm256_sub_pd(iz2,jz0);
 347
 348             /* Calculate squared distance and things based on it */
 349             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 350             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 351             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 352
 353             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 354             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 355             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 356
 357             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 358             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 359             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 360
 361             /* Load parameters for j particles */
 362             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 363                                                                  charge+jnrC+0,charge+jnrD+0);
 364
 365             fjx0             = _mm256_setzero_pd();
 366             fjy0             = _mm256_setzero_pd();
 367             fjz0             = _mm256_setzero_pd();
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq00             = _mm256_mul_pd(iq0,jq0);
 375
 376             /* COULOMB ELECTROSTATICS */
 377             velec            = _mm256_mul_pd(qq00,rinv00);
 378             felec            = _mm256_mul_pd(velec,rinvsq00);
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velec            = _mm256_andnot_pd(dummy_mask,velec);
 382             velecsum         = _mm256_add_pd(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 387
 388             /* Calculate temporary vectorial force */
 389             tx               = _mm256_mul_pd(fscal,dx00);
 390             ty               = _mm256_mul_pd(fscal,dy00);
 391             tz               = _mm256_mul_pd(fscal,dz00);
 392
 393             /* Update vectorial force */
 394             fix0             = _mm256_add_pd(fix0,tx);
 395             fiy0             = _mm256_add_pd(fiy0,ty);
 396             fiz0             = _mm256_add_pd(fiz0,tz);
 397
 398             fjx0             = _mm256_add_pd(fjx0,tx);
 399             fjy0             = _mm256_add_pd(fjy0,ty);
 400             fjz0             = _mm256_add_pd(fjz0,tz);
 401
 402             /**************************
 403              * CALCULATE INTERACTIONS *
 404              **************************/
 405
 406             /* Compute parameters for interactions between i and j atoms */
 407             qq10             = _mm256_mul_pd(iq1,jq0);
 408
 409             /* COULOMB ELECTROSTATICS */
 410             velec            = _mm256_mul_pd(qq10,rinv10);
 411             felec            = _mm256_mul_pd(velec,rinvsq10);
 412
 413             /* Update potential sum for this i atom from the interaction with this j atom. */
 414             velec            = _mm256_andnot_pd(dummy_mask,velec);
 415             velecsum         = _mm256_add_pd(velecsum,velec);
 416
 417             fscal            = felec;
 418
 419             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 420
 421             /* Calculate temporary vectorial force */
 422             tx               = _mm256_mul_pd(fscal,dx10);
 423             ty               = _mm256_mul_pd(fscal,dy10);
 424             tz               = _mm256_mul_pd(fscal,dz10);
 425
 426             /* Update vectorial force */
 427             fix1             = _mm256_add_pd(fix1,tx);
 428             fiy1             = _mm256_add_pd(fiy1,ty);
 429             fiz1             = _mm256_add_pd(fiz1,tz);
 430
 431             fjx0             = _mm256_add_pd(fjx0,tx);
 432             fjy0             = _mm256_add_pd(fjy0,ty);
 433             fjz0             = _mm256_add_pd(fjz0,tz);
 434
 435             /**************************
 436              * CALCULATE INTERACTIONS *
 437              **************************/
 438
 439             /* Compute parameters for interactions between i and j atoms */
 440             qq20             = _mm256_mul_pd(iq2,jq0);
 441
 442             /* COULOMB ELECTROSTATICS */
 443             velec            = _mm256_mul_pd(qq20,rinv20);
 444             felec            = _mm256_mul_pd(velec,rinvsq20);
 445
 446             /* Update potential sum for this i atom from the interaction with this j atom. */
 447             velec            = _mm256_andnot_pd(dummy_mask,velec);
 448             velecsum         = _mm256_add_pd(velecsum,velec);
 449
 450             fscal            = felec;
 451
 452             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 453
 454             /* Calculate temporary vectorial force */
 455             tx               = _mm256_mul_pd(fscal,dx20);
 456             ty               = _mm256_mul_pd(fscal,dy20);
 457             tz               = _mm256_mul_pd(fscal,dz20);
 458
 459             /* Update vectorial force */
 460             fix2             = _mm256_add_pd(fix2,tx);
 461             fiy2             = _mm256_add_pd(fiy2,ty);
 462             fiz2             = _mm256_add_pd(fiz2,tz);
 463
 464             fjx0             = _mm256_add_pd(fjx0,tx);
 465             fjy0             = _mm256_add_pd(fjy0,ty);
 466             fjz0             = _mm256_add_pd(fjz0,tz);
 467
 468             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 469             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 470             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 471             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 472
 473             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 474
 475             /* Inner loop uses 84 flops */
 476         }
 477
 478         /* End of innermost loop */
 479
 480         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 481                                                  f+i_coord_offset,fshift+i_shift_offset);
 482
 483         ggid                        = gid[iidx];
 484         /* Update potential energies */
 485         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 486
 487         /* Increment number of inner iterations */
 488         inneriter                  += j_index_end - j_index_start;
 489
 490         /* Outer loop uses 19 flops */
 491     }
 492
 493     /* Increment number of outer iterations */
 494     outeriter        += nri;
 495
 496     /* Update outer/inner flops */
 497
 498     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*84);
 499 }
 500 /*
 501  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_256_double
 502  * Electrostatics interaction: Coulomb
 503  * VdW interaction:            None
 504  * Geometry:                   Water3-Particle
 505  * Calculate force/pot:        Force
 506  */
 507 void
 508 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_256_double
 509                     (t_nblist * gmx_restrict                nlist,
 510                      rvec * gmx_restrict                    xx,
 511                      rvec * gmx_restrict                    ff,
 512                      t_forcerec * gmx_restrict              fr,
 513                      t_mdatoms * gmx_restrict               mdatoms,
 514                      nb_kernel_data_t * gmx_restrict        kernel_data,
 515                      t_nrnb * gmx_restrict                  nrnb)
 516 {
 517     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 518      * just 0 for non-waters.
 519      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 520      * jnr indices corresponding to data put in the four positions in the SIMD register.
 521      */
 522     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 523     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 524     int              jnrA,jnrB,jnrC,jnrD;
 525     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 526     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 527     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 528     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 529     real             rcutoff_scalar;
 530     real             *shiftvec,*fshift,*x,*f;
 531     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 532     real             scratch[4*DIM];
 533     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 534     real *           vdwioffsetptr0;
 535     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 536     real *           vdwioffsetptr1;
 537     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 538     real *           vdwioffsetptr2;
 539     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 540     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 541     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 542     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 543     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 544     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 545     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 546     real             *charge;
 547     __m256d          dummy_mask,cutoff_mask;
 548     __m128           tmpmask0,tmpmask1;
 549     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 550     __m256d          one     = _mm256_set1_pd(1.0);
 551     __m256d          two     = _mm256_set1_pd(2.0);
 552     x                = xx[0];
 553     f                = ff[0];
 554
 555     nri              = nlist->nri;
 556     iinr             = nlist->iinr;
 557     jindex           = nlist->jindex;
 558     jjnr             = nlist->jjnr;
 559     shiftidx         = nlist->shift;
 560     gid              = nlist->gid;
 561     shiftvec         = fr->shift_vec[0];
 562     fshift           = fr->fshift[0];
 563     facel            = _mm256_set1_pd(fr->epsfac);
 564     charge           = mdatoms->chargeA;
 565
 566     /* Setup water-specific parameters */
 567     inr              = nlist->iinr[0];
 568     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 569     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 570     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 571
 572     /* Avoid stupid compiler warnings */
 573     jnrA = jnrB = jnrC = jnrD = 0;
 574     j_coord_offsetA = 0;
 575     j_coord_offsetB = 0;
 576     j_coord_offsetC = 0;
 577     j_coord_offsetD = 0;
 578
 579     outeriter        = 0;
 580     inneriter        = 0;
 581
 582     for(iidx=0;iidx<4*DIM;iidx++)
 583     {
 584         scratch[iidx] = 0.0;
 585     }
 586
 587     /* Start outer loop over neighborlists */
 588     for(iidx=0; iidx<nri; iidx++)
 589     {
 590         /* Load shift vector for this list */
 591         i_shift_offset   = DIM*shiftidx[iidx];
 592
 593         /* Load limits for loop over neighbors */
 594         j_index_start    = jindex[iidx];
 595         j_index_end      = jindex[iidx+1];
 596
 597         /* Get outer coordinate index */
 598         inr              = iinr[iidx];
 599         i_coord_offset   = DIM*inr;
 600
 601         /* Load i particle coords and add shift vector */
 602         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 603                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 604
 605         fix0             = _mm256_setzero_pd();
 606         fiy0             = _mm256_setzero_pd();
 607         fiz0             = _mm256_setzero_pd();
 608         fix1             = _mm256_setzero_pd();
 609         fiy1             = _mm256_setzero_pd();
 610         fiz1             = _mm256_setzero_pd();
 611         fix2             = _mm256_setzero_pd();
 612         fiy2             = _mm256_setzero_pd();
 613         fiz2             = _mm256_setzero_pd();
 614
 615         /* Start inner kernel loop */
 616         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 617         {
 618
 619             /* Get j neighbor index, and coordinate index */
 620             jnrA             = jjnr[jidx];
 621             jnrB             = jjnr[jidx+1];
 622             jnrC             = jjnr[jidx+2];
 623             jnrD             = jjnr[jidx+3];
 624             j_coord_offsetA  = DIM*jnrA;
 625             j_coord_offsetB  = DIM*jnrB;
 626             j_coord_offsetC  = DIM*jnrC;
 627             j_coord_offsetD  = DIM*jnrD;
 628
 629             /* load j atom coordinates */
 630             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 631                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 632                                                  &jx0,&jy0,&jz0);
 633
 634             /* Calculate displacement vector */
 635             dx00             = _mm256_sub_pd(ix0,jx0);
 636             dy00             = _mm256_sub_pd(iy0,jy0);
 637             dz00             = _mm256_sub_pd(iz0,jz0);
 638             dx10             = _mm256_sub_pd(ix1,jx0);
 639             dy10             = _mm256_sub_pd(iy1,jy0);
 640             dz10             = _mm256_sub_pd(iz1,jz0);
 641             dx20             = _mm256_sub_pd(ix2,jx0);
 642             dy20             = _mm256_sub_pd(iy2,jy0);
 643             dz20             = _mm256_sub_pd(iz2,jz0);
 644
 645             /* Calculate squared distance and things based on it */
 646             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 647             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 648             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 649
 650             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 651             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 652             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 653
 654             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 655             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 656             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 657
 658             /* Load parameters for j particles */
 659             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 660                                                                  charge+jnrC+0,charge+jnrD+0);
 661
 662             fjx0             = _mm256_setzero_pd();
 663             fjy0             = _mm256_setzero_pd();
 664             fjz0             = _mm256_setzero_pd();
 665
 666             /**************************
 667              * CALCULATE INTERACTIONS *
 668              **************************/
 669
 670             /* Compute parameters for interactions between i and j atoms */
 671             qq00             = _mm256_mul_pd(iq0,jq0);
 672
 673             /* COULOMB ELECTROSTATICS */
 674             velec            = _mm256_mul_pd(qq00,rinv00);
 675             felec            = _mm256_mul_pd(velec,rinvsq00);
 676
 677             fscal            = felec;
 678
 679             /* Calculate temporary vectorial force */
 680             tx               = _mm256_mul_pd(fscal,dx00);
 681             ty               = _mm256_mul_pd(fscal,dy00);
 682             tz               = _mm256_mul_pd(fscal,dz00);
 683
 684             /* Update vectorial force */
 685             fix0             = _mm256_add_pd(fix0,tx);
 686             fiy0             = _mm256_add_pd(fiy0,ty);
 687             fiz0             = _mm256_add_pd(fiz0,tz);
 688
 689             fjx0             = _mm256_add_pd(fjx0,tx);
 690             fjy0             = _mm256_add_pd(fjy0,ty);
 691             fjz0             = _mm256_add_pd(fjz0,tz);
 692
 693             /**************************
 694              * CALCULATE INTERACTIONS *
 695              **************************/
 696
 697             /* Compute parameters for interactions between i and j atoms */
 698             qq10             = _mm256_mul_pd(iq1,jq0);
 699
 700             /* COULOMB ELECTROSTATICS */
 701             velec            = _mm256_mul_pd(qq10,rinv10);
 702             felec            = _mm256_mul_pd(velec,rinvsq10);
 703
 704             fscal            = felec;
 705
 706             /* Calculate temporary vectorial force */
 707             tx               = _mm256_mul_pd(fscal,dx10);
 708             ty               = _mm256_mul_pd(fscal,dy10);
 709             tz               = _mm256_mul_pd(fscal,dz10);
 710
 711             /* Update vectorial force */
 712             fix1             = _mm256_add_pd(fix1,tx);
 713             fiy1             = _mm256_add_pd(fiy1,ty);
 714             fiz1             = _mm256_add_pd(fiz1,tz);
 715
 716             fjx0             = _mm256_add_pd(fjx0,tx);
 717             fjy0             = _mm256_add_pd(fjy0,ty);
 718             fjz0             = _mm256_add_pd(fjz0,tz);
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             /* Compute parameters for interactions between i and j atoms */
 725             qq20             = _mm256_mul_pd(iq2,jq0);
 726
 727             /* COULOMB ELECTROSTATICS */
 728             velec            = _mm256_mul_pd(qq20,rinv20);
 729             felec            = _mm256_mul_pd(velec,rinvsq20);
 730
 731             fscal            = felec;
 732
 733             /* Calculate temporary vectorial force */
 734             tx               = _mm256_mul_pd(fscal,dx20);
 735             ty               = _mm256_mul_pd(fscal,dy20);
 736             tz               = _mm256_mul_pd(fscal,dz20);
 737
 738             /* Update vectorial force */
 739             fix2             = _mm256_add_pd(fix2,tx);
 740             fiy2             = _mm256_add_pd(fiy2,ty);
 741             fiz2             = _mm256_add_pd(fiz2,tz);
 742
 743             fjx0             = _mm256_add_pd(fjx0,tx);
 744             fjy0             = _mm256_add_pd(fjy0,ty);
 745             fjz0             = _mm256_add_pd(fjz0,tz);
 746
 747             fjptrA             = f+j_coord_offsetA;
 748             fjptrB             = f+j_coord_offsetB;
 749             fjptrC             = f+j_coord_offsetC;
 750             fjptrD             = f+j_coord_offsetD;
 751
 752             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 753
 754             /* Inner loop uses 81 flops */
 755         }
 756
 757         if(jidx<j_index_end)
 758         {
 759
 760             /* Get j neighbor index, and coordinate index */
 761             jnrlistA         = jjnr[jidx];
 762             jnrlistB         = jjnr[jidx+1];
 763             jnrlistC         = jjnr[jidx+2];
 764             jnrlistD         = jjnr[jidx+3];
 765             /* Sign of each element will be negative for non-real atoms.
 766              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 767              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 768              */
 769             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 770
 771             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 772             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 773             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 774
 775             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 776             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 777             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 778             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 779             j_coord_offsetA  = DIM*jnrA;
 780             j_coord_offsetB  = DIM*jnrB;
 781             j_coord_offsetC  = DIM*jnrC;
 782             j_coord_offsetD  = DIM*jnrD;
 783
 784             /* load j atom coordinates */
 785             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 786                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 787                                                  &jx0,&jy0,&jz0);
 788
 789             /* Calculate displacement vector */
 790             dx00             = _mm256_sub_pd(ix0,jx0);
 791             dy00             = _mm256_sub_pd(iy0,jy0);
 792             dz00             = _mm256_sub_pd(iz0,jz0);
 793             dx10             = _mm256_sub_pd(ix1,jx0);
 794             dy10             = _mm256_sub_pd(iy1,jy0);
 795             dz10             = _mm256_sub_pd(iz1,jz0);
 796             dx20             = _mm256_sub_pd(ix2,jx0);
 797             dy20             = _mm256_sub_pd(iy2,jy0);
 798             dz20             = _mm256_sub_pd(iz2,jz0);
 799
 800             /* Calculate squared distance and things based on it */
 801             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 802             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 803             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 804
 805             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 806             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 807             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 808
 809             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 810             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 811             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 812
 813             /* Load parameters for j particles */
 814             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 815                                                                  charge+jnrC+0,charge+jnrD+0);
 816
 817             fjx0             = _mm256_setzero_pd();
 818             fjy0             = _mm256_setzero_pd();
 819             fjz0             = _mm256_setzero_pd();
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             /* Compute parameters for interactions between i and j atoms */
 826             qq00             = _mm256_mul_pd(iq0,jq0);
 827
 828             /* COULOMB ELECTROSTATICS */
 829             velec            = _mm256_mul_pd(qq00,rinv00);
 830             felec            = _mm256_mul_pd(velec,rinvsq00);
 831
 832             fscal            = felec;
 833
 834             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 835
 836             /* Calculate temporary vectorial force */
 837             tx               = _mm256_mul_pd(fscal,dx00);
 838             ty               = _mm256_mul_pd(fscal,dy00);
 839             tz               = _mm256_mul_pd(fscal,dz00);
 840
 841             /* Update vectorial force */
 842             fix0             = _mm256_add_pd(fix0,tx);
 843             fiy0             = _mm256_add_pd(fiy0,ty);
 844             fiz0             = _mm256_add_pd(fiz0,tz);
 845
 846             fjx0             = _mm256_add_pd(fjx0,tx);
 847             fjy0             = _mm256_add_pd(fjy0,ty);
 848             fjz0             = _mm256_add_pd(fjz0,tz);
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             /* Compute parameters for interactions between i and j atoms */
 855             qq10             = _mm256_mul_pd(iq1,jq0);
 856
 857             /* COULOMB ELECTROSTATICS */
 858             velec            = _mm256_mul_pd(qq10,rinv10);
 859             felec            = _mm256_mul_pd(velec,rinvsq10);
 860
 861             fscal            = felec;
 862
 863             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 864
 865             /* Calculate temporary vectorial force */
 866             tx               = _mm256_mul_pd(fscal,dx10);
 867             ty               = _mm256_mul_pd(fscal,dy10);
 868             tz               = _mm256_mul_pd(fscal,dz10);
 869
 870             /* Update vectorial force */
 871             fix1             = _mm256_add_pd(fix1,tx);
 872             fiy1             = _mm256_add_pd(fiy1,ty);
 873             fiz1             = _mm256_add_pd(fiz1,tz);
 874
 875             fjx0             = _mm256_add_pd(fjx0,tx);
 876             fjy0             = _mm256_add_pd(fjy0,ty);
 877             fjz0             = _mm256_add_pd(fjz0,tz);
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq20             = _mm256_mul_pd(iq2,jq0);
 885
 886             /* COULOMB ELECTROSTATICS */
 887             velec            = _mm256_mul_pd(qq20,rinv20);
 888             felec            = _mm256_mul_pd(velec,rinvsq20);
 889
 890             fscal            = felec;
 891
 892             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 893
 894             /* Calculate temporary vectorial force */
 895             tx               = _mm256_mul_pd(fscal,dx20);
 896             ty               = _mm256_mul_pd(fscal,dy20);
 897             tz               = _mm256_mul_pd(fscal,dz20);
 898
 899             /* Update vectorial force */
 900             fix2             = _mm256_add_pd(fix2,tx);
 901             fiy2             = _mm256_add_pd(fiy2,ty);
 902             fiz2             = _mm256_add_pd(fiz2,tz);
 903
 904             fjx0             = _mm256_add_pd(fjx0,tx);
 905             fjy0             = _mm256_add_pd(fjy0,ty);
 906             fjz0             = _mm256_add_pd(fjz0,tz);
 907
 908             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 909             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 910             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 911             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 912
 913             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 914
 915             /* Inner loop uses 81 flops */
 916         }
 917
 918         /* End of innermost loop */
 919
 920         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 921                                                  f+i_coord_offset,fshift+i_shift_offset);
 922
 923         /* Increment number of inner iterations */
 924         inneriter                  += j_index_end - j_index_start;
 925
 926         /* Outer loop uses 18 flops */
 927     }
 928
 929     /* Increment number of outer iterations */
 930     outeriter        += nri;
 931
 932     /* Update outer/inner flops */
 933
 934     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*81);
 935 }