src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwNone_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset1;
  84     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  85     int              vdwioffset2;
  86     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  87     int              vdwioffset3;
  88     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     __m128           dummy_mask,cutoff_mask;
  97     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  98     __m128           one     = _mm_set1_ps(1.0);
  99     __m128           two     = _mm_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     krf              = _mm_set1_ps(fr->ic->k_rf);
 114     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 115     crf              = _mm_set1_ps(fr->ic->c_rf);
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 120     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 121     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 154                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 155
 156         fix1             = _mm_setzero_ps();
 157         fiy1             = _mm_setzero_ps();
 158         fiz1             = _mm_setzero_ps();
 159         fix2             = _mm_setzero_ps();
 160         fiy2             = _mm_setzero_ps();
 161         fiz2             = _mm_setzero_ps();
 162         fix3             = _mm_setzero_ps();
 163         fiy3             = _mm_setzero_ps();
 164         fiz3             = _mm_setzero_ps();
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             jnrC             = jjnr[jidx+2];
 177             jnrD             = jjnr[jidx+3];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180             j_coord_offsetC  = DIM*jnrC;
 181             j_coord_offsetD  = DIM*jnrD;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                               x+j_coord_offsetC,x+j_coord_offsetD,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx10             = _mm_sub_ps(ix1,jx0);
 190             dy10             = _mm_sub_ps(iy1,jy0);
 191             dz10             = _mm_sub_ps(iz1,jz0);
 192             dx20             = _mm_sub_ps(ix2,jx0);
 193             dy20             = _mm_sub_ps(iy2,jy0);
 194             dz20             = _mm_sub_ps(iz2,jz0);
 195             dx30             = _mm_sub_ps(ix3,jx0);
 196             dy30             = _mm_sub_ps(iy3,jy0);
 197             dz30             = _mm_sub_ps(iz3,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 201             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 202             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 203
 204             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 205             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 206             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 207
 208             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 209             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 210             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 214                                                               charge+jnrC+0,charge+jnrD+0);
 215
 216             fjx0             = _mm_setzero_ps();
 217             fjy0             = _mm_setzero_ps();
 218             fjz0             = _mm_setzero_ps();
 219
 220             /**************************
 221              * CALCULATE INTERACTIONS *
 222              **************************/
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq10             = _mm_mul_ps(iq1,jq0);
 226
 227             /* REACTION-FIELD ELECTROSTATICS */
 228             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 229             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             velecsum         = _mm_add_ps(velecsum,velec);
 233
 234             fscal            = felec;
 235
 236              /* Update vectorial force */
 237             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 238             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 239             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 240
 241             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 242             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 243             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             /* Compute parameters for interactions between i and j atoms */
 250             qq20             = _mm_mul_ps(iq2,jq0);
 251
 252             /* REACTION-FIELD ELECTROSTATICS */
 253             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 254             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 255
 256             /* Update potential sum for this i atom from the interaction with this j atom. */
 257             velecsum         = _mm_add_ps(velecsum,velec);
 258
 259             fscal            = felec;
 260
 261              /* Update vectorial force */
 262             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 263             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 264             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 265
 266             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 267             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 268             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq30             = _mm_mul_ps(iq3,jq0);
 276
 277             /* REACTION-FIELD ELECTROSTATICS */
 278             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 279             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_ps(velecsum,velec);
 283
 284             fscal            = felec;
 285
 286              /* Update vectorial force */
 287             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 288             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 289             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 290
 291             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 292             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 293             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 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_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 301
 302             /* Inner loop uses 105 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_ps(mask,val) to clear dummy entries.
 316              */
 317             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 318             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 319             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 320             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 321             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 322             j_coord_offsetA  = DIM*jnrA;
 323             j_coord_offsetB  = DIM*jnrB;
 324             j_coord_offsetC  = DIM*jnrC;
 325             j_coord_offsetD  = DIM*jnrD;
 326
 327             /* load j atom coordinates */
 328             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 329                                               x+j_coord_offsetC,x+j_coord_offsetD,
 330                                               &jx0,&jy0,&jz0);
 331
 332             /* Calculate displacement vector */
 333             dx10             = _mm_sub_ps(ix1,jx0);
 334             dy10             = _mm_sub_ps(iy1,jy0);
 335             dz10             = _mm_sub_ps(iz1,jz0);
 336             dx20             = _mm_sub_ps(ix2,jx0);
 337             dy20             = _mm_sub_ps(iy2,jy0);
 338             dz20             = _mm_sub_ps(iz2,jz0);
 339             dx30             = _mm_sub_ps(ix3,jx0);
 340             dy30             = _mm_sub_ps(iy3,jy0);
 341             dz30             = _mm_sub_ps(iz3,jz0);
 342
 343             /* Calculate squared distance and things based on it */
 344             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 345             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 346             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 347
 348             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 349             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 350             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 351
 352             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 353             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 354             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 355
 356             /* Load parameters for j particles */
 357             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 358                                                               charge+jnrC+0,charge+jnrD+0);
 359
 360             fjx0             = _mm_setzero_ps();
 361             fjy0             = _mm_setzero_ps();
 362             fjz0             = _mm_setzero_ps();
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq10             = _mm_mul_ps(iq1,jq0);
 370
 371             /* REACTION-FIELD ELECTROSTATICS */
 372             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 373             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm_andnot_ps(dummy_mask,velec);
 377             velecsum         = _mm_add_ps(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 382
 383              /* Update vectorial force */
 384             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 385             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 386             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 387
 388             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 389             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 390             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             /* Compute parameters for interactions between i and j atoms */
 397             qq20             = _mm_mul_ps(iq2,jq0);
 398
 399             /* REACTION-FIELD ELECTROSTATICS */
 400             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 401             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velec            = _mm_andnot_ps(dummy_mask,velec);
 405             velecsum         = _mm_add_ps(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 410
 411              /* Update vectorial force */
 412             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 413             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 414             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 415
 416             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 417             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 418             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             /* Compute parameters for interactions between i and j atoms */
 425             qq30             = _mm_mul_ps(iq3,jq0);
 426
 427             /* REACTION-FIELD ELECTROSTATICS */
 428             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 429             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velec            = _mm_andnot_ps(dummy_mask,velec);
 433             velecsum         = _mm_add_ps(velecsum,velec);
 434
 435             fscal            = felec;
 436
 437             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 438
 439              /* Update vectorial force */
 440             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 441             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 442             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 443
 444             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 445             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 446             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 447
 448             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 449             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 450             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 451             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 452
 453             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 454
 455             /* Inner loop uses 105 flops */
 456         }
 457
 458         /* End of innermost loop */
 459
 460         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 461                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 462
 463         ggid                        = gid[iidx];
 464         /* Update potential energies */
 465         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 466
 467         /* Increment number of inner iterations */
 468         inneriter                  += j_index_end - j_index_start;
 469
 470         /* Outer loop uses 19 flops */
 471     }
 472
 473     /* Increment number of outer iterations */
 474     outeriter        += nri;
 475
 476     /* Update outer/inner flops */
 477
 478     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*105);
 479 }
 480 /*
 481  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_single
 482  * Electrostatics interaction: ReactionField
 483  * VdW interaction:            None
 484  * Geometry:                   Water4-Particle
 485  * Calculate force/pot:        Force
 486  */
 487 void
 488 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_single
 489                     (t_nblist                    * gmx_restrict       nlist,
 490                      rvec                        * gmx_restrict          xx,
 491                      rvec                        * gmx_restrict          ff,
 492                      t_forcerec                  * gmx_restrict          fr,
 493                      t_mdatoms                   * gmx_restrict     mdatoms,
 494                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 495                      t_nrnb                      * gmx_restrict        nrnb)
 496 {
 497     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 498      * just 0 for non-waters.
 499      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 500      * jnr indices corresponding to data put in the four positions in the SIMD register.
 501      */
 502     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 503     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 504     int              jnrA,jnrB,jnrC,jnrD;
 505     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 506     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 507     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 508     real             rcutoff_scalar;
 509     real             *shiftvec,*fshift,*x,*f;
 510     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 511     real             scratch[4*DIM];
 512     __m128           fscal,rcutoff,rcutoff2,jidxall;
 513     int              vdwioffset1;
 514     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 515     int              vdwioffset2;
 516     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 517     int              vdwioffset3;
 518     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 519     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 520     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 521     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 522     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 523     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 524     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 525     real             *charge;
 526     __m128           dummy_mask,cutoff_mask;
 527     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 528     __m128           one     = _mm_set1_ps(1.0);
 529     __m128           two     = _mm_set1_ps(2.0);
 530     x                = xx[0];
 531     f                = ff[0];
 532
 533     nri              = nlist->nri;
 534     iinr             = nlist->iinr;
 535     jindex           = nlist->jindex;
 536     jjnr             = nlist->jjnr;
 537     shiftidx         = nlist->shift;
 538     gid              = nlist->gid;
 539     shiftvec         = fr->shift_vec[0];
 540     fshift           = fr->fshift[0];
 541     facel            = _mm_set1_ps(fr->epsfac);
 542     charge           = mdatoms->chargeA;
 543     krf              = _mm_set1_ps(fr->ic->k_rf);
 544     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 545     crf              = _mm_set1_ps(fr->ic->c_rf);
 546
 547     /* Setup water-specific parameters */
 548     inr              = nlist->iinr[0];
 549     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 550     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 551     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 552
 553     /* Avoid stupid compiler warnings */
 554     jnrA = jnrB = jnrC = jnrD = 0;
 555     j_coord_offsetA = 0;
 556     j_coord_offsetB = 0;
 557     j_coord_offsetC = 0;
 558     j_coord_offsetD = 0;
 559
 560     outeriter        = 0;
 561     inneriter        = 0;
 562
 563     for(iidx=0;iidx<4*DIM;iidx++)
 564     {
 565         scratch[iidx] = 0.0;
 566     }
 567
 568     /* Start outer loop over neighborlists */
 569     for(iidx=0; iidx<nri; iidx++)
 570     {
 571         /* Load shift vector for this list */
 572         i_shift_offset   = DIM*shiftidx[iidx];
 573
 574         /* Load limits for loop over neighbors */
 575         j_index_start    = jindex[iidx];
 576         j_index_end      = jindex[iidx+1];
 577
 578         /* Get outer coordinate index */
 579         inr              = iinr[iidx];
 580         i_coord_offset   = DIM*inr;
 581
 582         /* Load i particle coords and add shift vector */
 583         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 584                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 585
 586         fix1             = _mm_setzero_ps();
 587         fiy1             = _mm_setzero_ps();
 588         fiz1             = _mm_setzero_ps();
 589         fix2             = _mm_setzero_ps();
 590         fiy2             = _mm_setzero_ps();
 591         fiz2             = _mm_setzero_ps();
 592         fix3             = _mm_setzero_ps();
 593         fiy3             = _mm_setzero_ps();
 594         fiz3             = _mm_setzero_ps();
 595
 596         /* Start inner kernel loop */
 597         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 598         {
 599
 600             /* Get j neighbor index, and coordinate index */
 601             jnrA             = jjnr[jidx];
 602             jnrB             = jjnr[jidx+1];
 603             jnrC             = jjnr[jidx+2];
 604             jnrD             = jjnr[jidx+3];
 605             j_coord_offsetA  = DIM*jnrA;
 606             j_coord_offsetB  = DIM*jnrB;
 607             j_coord_offsetC  = DIM*jnrC;
 608             j_coord_offsetD  = DIM*jnrD;
 609
 610             /* load j atom coordinates */
 611             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 612                                               x+j_coord_offsetC,x+j_coord_offsetD,
 613                                               &jx0,&jy0,&jz0);
 614
 615             /* Calculate displacement vector */
 616             dx10             = _mm_sub_ps(ix1,jx0);
 617             dy10             = _mm_sub_ps(iy1,jy0);
 618             dz10             = _mm_sub_ps(iz1,jz0);
 619             dx20             = _mm_sub_ps(ix2,jx0);
 620             dy20             = _mm_sub_ps(iy2,jy0);
 621             dz20             = _mm_sub_ps(iz2,jz0);
 622             dx30             = _mm_sub_ps(ix3,jx0);
 623             dy30             = _mm_sub_ps(iy3,jy0);
 624             dz30             = _mm_sub_ps(iz3,jz0);
 625
 626             /* Calculate squared distance and things based on it */
 627             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 628             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 629             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 630
 631             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 632             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 633             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 634
 635             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 636             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 637             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 638
 639             /* Load parameters for j particles */
 640             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 641                                                               charge+jnrC+0,charge+jnrD+0);
 642
 643             fjx0             = _mm_setzero_ps();
 644             fjy0             = _mm_setzero_ps();
 645             fjz0             = _mm_setzero_ps();
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             /* Compute parameters for interactions between i and j atoms */
 652             qq10             = _mm_mul_ps(iq1,jq0);
 653
 654             /* REACTION-FIELD ELECTROSTATICS */
 655             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 656
 657             fscal            = felec;
 658
 659              /* Update vectorial force */
 660             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 661             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 662             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 663
 664             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 665             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 666             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 667
 668             /**************************
 669              * CALCULATE INTERACTIONS *
 670              **************************/
 671
 672             /* Compute parameters for interactions between i and j atoms */
 673             qq20             = _mm_mul_ps(iq2,jq0);
 674
 675             /* REACTION-FIELD ELECTROSTATICS */
 676             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 677
 678             fscal            = felec;
 679
 680              /* Update vectorial force */
 681             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 682             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 683             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 684
 685             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 686             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 687             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 688
 689             /**************************
 690              * CALCULATE INTERACTIONS *
 691              **************************/
 692
 693             /* Compute parameters for interactions between i and j atoms */
 694             qq30             = _mm_mul_ps(iq3,jq0);
 695
 696             /* REACTION-FIELD ELECTROSTATICS */
 697             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 698
 699             fscal            = felec;
 700
 701              /* Update vectorial force */
 702             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 703             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 704             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 705
 706             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 707             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 708             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 709
 710             fjptrA             = f+j_coord_offsetA;
 711             fjptrB             = f+j_coord_offsetB;
 712             fjptrC             = f+j_coord_offsetC;
 713             fjptrD             = f+j_coord_offsetD;
 714
 715             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 716
 717             /* Inner loop uses 90 flops */
 718         }
 719
 720         if(jidx<j_index_end)
 721         {
 722
 723             /* Get j neighbor index, and coordinate index */
 724             jnrlistA         = jjnr[jidx];
 725             jnrlistB         = jjnr[jidx+1];
 726             jnrlistC         = jjnr[jidx+2];
 727             jnrlistD         = jjnr[jidx+3];
 728             /* Sign of each element will be negative for non-real atoms.
 729              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 730              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 731              */
 732             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 733             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 734             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 735             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 736             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 737             j_coord_offsetA  = DIM*jnrA;
 738             j_coord_offsetB  = DIM*jnrB;
 739             j_coord_offsetC  = DIM*jnrC;
 740             j_coord_offsetD  = DIM*jnrD;
 741
 742             /* load j atom coordinates */
 743             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 744                                               x+j_coord_offsetC,x+j_coord_offsetD,
 745                                               &jx0,&jy0,&jz0);
 746
 747             /* Calculate displacement vector */
 748             dx10             = _mm_sub_ps(ix1,jx0);
 749             dy10             = _mm_sub_ps(iy1,jy0);
 750             dz10             = _mm_sub_ps(iz1,jz0);
 751             dx20             = _mm_sub_ps(ix2,jx0);
 752             dy20             = _mm_sub_ps(iy2,jy0);
 753             dz20             = _mm_sub_ps(iz2,jz0);
 754             dx30             = _mm_sub_ps(ix3,jx0);
 755             dy30             = _mm_sub_ps(iy3,jy0);
 756             dz30             = _mm_sub_ps(iz3,jz0);
 757
 758             /* Calculate squared distance and things based on it */
 759             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 760             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 761             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 762
 763             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 764             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 765             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 766
 767             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 768             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 769             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 770
 771             /* Load parameters for j particles */
 772             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 773                                                               charge+jnrC+0,charge+jnrD+0);
 774
 775             fjx0             = _mm_setzero_ps();
 776             fjy0             = _mm_setzero_ps();
 777             fjz0             = _mm_setzero_ps();
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             /* Compute parameters for interactions between i and j atoms */
 784             qq10             = _mm_mul_ps(iq1,jq0);
 785
 786             /* REACTION-FIELD ELECTROSTATICS */
 787             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 788
 789             fscal            = felec;
 790
 791             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 792
 793              /* Update vectorial force */
 794             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 795             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 796             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 797
 798             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 799             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 800             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 801
 802             /**************************
 803              * CALCULATE INTERACTIONS *
 804              **************************/
 805
 806             /* Compute parameters for interactions between i and j atoms */
 807             qq20             = _mm_mul_ps(iq2,jq0);
 808
 809             /* REACTION-FIELD ELECTROSTATICS */
 810             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 811
 812             fscal            = felec;
 813
 814             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 815
 816              /* Update vectorial force */
 817             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 818             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 819             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 820
 821             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 822             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 823             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 824
 825             /**************************
 826              * CALCULATE INTERACTIONS *
 827              **************************/
 828
 829             /* Compute parameters for interactions between i and j atoms */
 830             qq30             = _mm_mul_ps(iq3,jq0);
 831
 832             /* REACTION-FIELD ELECTROSTATICS */
 833             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 834
 835             fscal            = felec;
 836
 837             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 838
 839              /* Update vectorial force */
 840             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 841             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 842             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 843
 844             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 845             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 846             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 847
 848             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 849             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 850             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 851             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 852
 853             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 854
 855             /* Inner loop uses 90 flops */
 856         }
 857
 858         /* End of innermost loop */
 859
 860         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 861                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 862
 863         /* Increment number of inner iterations */
 864         inneriter                  += j_index_end - j_index_start;
 865
 866         /* Outer loop uses 18 flops */
 867     }
 868
 869     /* Increment number of outer iterations */
 870     outeriter        += nri;
 871
 872     /* Update outer/inner flops */
 873
 874     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*90);
 875 }