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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = _mm_set1_ps(fr->ic->k_rf);
 116     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 117     crf              = _mm_set1_ps(fr->ic->c_rf);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 122     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 123     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     for(iidx=0;iidx<4*DIM;iidx++)
 136     {
 137         scratch[iidx] = 0.0;
 138     }
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 156                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 157
 158         fix1             = _mm_setzero_ps();
 159         fiy1             = _mm_setzero_ps();
 160         fiz1             = _mm_setzero_ps();
 161         fix2             = _mm_setzero_ps();
 162         fiy2             = _mm_setzero_ps();
 163         fiz2             = _mm_setzero_ps();
 164         fix3             = _mm_setzero_ps();
 165         fiy3             = _mm_setzero_ps();
 166         fiz3             = _mm_setzero_ps();
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm_setzero_ps();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             jnrC             = jjnr[jidx+2];
 179             jnrD             = jjnr[jidx+3];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               x+j_coord_offsetC,x+j_coord_offsetD,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx10             = _mm_sub_ps(ix1,jx0);
 192             dy10             = _mm_sub_ps(iy1,jy0);
 193             dz10             = _mm_sub_ps(iz1,jz0);
 194             dx20             = _mm_sub_ps(ix2,jx0);
 195             dy20             = _mm_sub_ps(iy2,jy0);
 196             dz20             = _mm_sub_ps(iz2,jz0);
 197             dx30             = _mm_sub_ps(ix3,jx0);
 198             dy30             = _mm_sub_ps(iy3,jy0);
 199             dz30             = _mm_sub_ps(iz3,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 203             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 204             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 205
 206             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 207             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 208             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 209
 210             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 211             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 212             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 216                                                               charge+jnrC+0,charge+jnrD+0);
 217
 218             fjx0             = _mm_setzero_ps();
 219             fjy0             = _mm_setzero_ps();
 220             fjz0             = _mm_setzero_ps();
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq10             = _mm_mul_ps(iq1,jq0);
 228
 229             /* REACTION-FIELD ELECTROSTATICS */
 230             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 231             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             velecsum         = _mm_add_ps(velecsum,velec);
 235
 236             fscal            = felec;
 237
 238              /* Update vectorial force */
 239             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 240             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 241             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 242
 243             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 244             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 245             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq20             = _mm_mul_ps(iq2,jq0);
 253
 254             /* REACTION-FIELD ELECTROSTATICS */
 255             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 256             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 257
 258             /* Update potential sum for this i atom from the interaction with this j atom. */
 259             velecsum         = _mm_add_ps(velecsum,velec);
 260
 261             fscal            = felec;
 262
 263              /* Update vectorial force */
 264             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 265             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 266             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 267
 268             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 269             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 270             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             qq30             = _mm_mul_ps(iq3,jq0);
 278
 279             /* REACTION-FIELD ELECTROSTATICS */
 280             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 281             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velecsum         = _mm_add_ps(velecsum,velec);
 285
 286             fscal            = felec;
 287
 288              /* Update vectorial force */
 289             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 290             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 291             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 292
 293             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 294             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 295             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 296
 297             fjptrA             = f+j_coord_offsetA;
 298             fjptrB             = f+j_coord_offsetB;
 299             fjptrC             = f+j_coord_offsetC;
 300             fjptrD             = f+j_coord_offsetD;
 301
 302             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 303
 304             /* Inner loop uses 105 flops */
 305         }
 306
 307         if(jidx<j_index_end)
 308         {
 309
 310             /* Get j neighbor index, and coordinate index */
 311             jnrlistA         = jjnr[jidx];
 312             jnrlistB         = jjnr[jidx+1];
 313             jnrlistC         = jjnr[jidx+2];
 314             jnrlistD         = jjnr[jidx+3];
 315             /* Sign of each element will be negative for non-real atoms.
 316              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 317              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 318              */
 319             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 320             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 321             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 322             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 323             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 324             j_coord_offsetA  = DIM*jnrA;
 325             j_coord_offsetB  = DIM*jnrB;
 326             j_coord_offsetC  = DIM*jnrC;
 327             j_coord_offsetD  = DIM*jnrD;
 328
 329             /* load j atom coordinates */
 330             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 331                                               x+j_coord_offsetC,x+j_coord_offsetD,
 332                                               &jx0,&jy0,&jz0);
 333
 334             /* Calculate displacement vector */
 335             dx10             = _mm_sub_ps(ix1,jx0);
 336             dy10             = _mm_sub_ps(iy1,jy0);
 337             dz10             = _mm_sub_ps(iz1,jz0);
 338             dx20             = _mm_sub_ps(ix2,jx0);
 339             dy20             = _mm_sub_ps(iy2,jy0);
 340             dz20             = _mm_sub_ps(iz2,jz0);
 341             dx30             = _mm_sub_ps(ix3,jx0);
 342             dy30             = _mm_sub_ps(iy3,jy0);
 343             dz30             = _mm_sub_ps(iz3,jz0);
 344
 345             /* Calculate squared distance and things based on it */
 346             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 347             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 348             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 349
 350             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 351             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 352             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 353
 354             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 355             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 356             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 357
 358             /* Load parameters for j particles */
 359             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 360                                                               charge+jnrC+0,charge+jnrD+0);
 361
 362             fjx0             = _mm_setzero_ps();
 363             fjy0             = _mm_setzero_ps();
 364             fjz0             = _mm_setzero_ps();
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq10             = _mm_mul_ps(iq1,jq0);
 372
 373             /* REACTION-FIELD ELECTROSTATICS */
 374             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 375             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm_andnot_ps(dummy_mask,velec);
 379             velecsum         = _mm_add_ps(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 384
 385              /* Update vectorial force */
 386             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 387             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 388             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 389
 390             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 391             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 392             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             /* Compute parameters for interactions between i and j atoms */
 399             qq20             = _mm_mul_ps(iq2,jq0);
 400
 401             /* REACTION-FIELD ELECTROSTATICS */
 402             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 403             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velec            = _mm_andnot_ps(dummy_mask,velec);
 407             velecsum         = _mm_add_ps(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 412
 413              /* Update vectorial force */
 414             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 415             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 416             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 417
 418             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 419             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 420             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             /* Compute parameters for interactions between i and j atoms */
 427             qq30             = _mm_mul_ps(iq3,jq0);
 428
 429             /* REACTION-FIELD ELECTROSTATICS */
 430             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 431             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velec            = _mm_andnot_ps(dummy_mask,velec);
 435             velecsum         = _mm_add_ps(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 440
 441              /* Update vectorial force */
 442             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 443             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 444             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 445
 446             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 447             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 448             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 449
 450             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 451             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 452             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 453             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 454
 455             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 456
 457             /* Inner loop uses 105 flops */
 458         }
 459
 460         /* End of innermost loop */
 461
 462         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 463                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 464
 465         ggid                        = gid[iidx];
 466         /* Update potential energies */
 467         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 468
 469         /* Increment number of inner iterations */
 470         inneriter                  += j_index_end - j_index_start;
 471
 472         /* Outer loop uses 19 flops */
 473     }
 474
 475     /* Increment number of outer iterations */
 476     outeriter        += nri;
 477
 478     /* Update outer/inner flops */
 479
 480     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*105);
 481 }
 482 /*
 483  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_single
 484  * Electrostatics interaction: ReactionField
 485  * VdW interaction:            None
 486  * Geometry:                   Water4-Particle
 487  * Calculate force/pot:        Force
 488  */
 489 void
 490 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_single
 491                     (t_nblist                    * gmx_restrict       nlist,
 492                      rvec                        * gmx_restrict          xx,
 493                      rvec                        * gmx_restrict          ff,
 494                      t_forcerec                  * gmx_restrict          fr,
 495                      t_mdatoms                   * gmx_restrict     mdatoms,
 496                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 497                      t_nrnb                      * gmx_restrict        nrnb)
 498 {
 499     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 500      * just 0 for non-waters.
 501      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 502      * jnr indices corresponding to data put in the four positions in the SIMD register.
 503      */
 504     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 505     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 506     int              jnrA,jnrB,jnrC,jnrD;
 507     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 508     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 509     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 510     real             rcutoff_scalar;
 511     real             *shiftvec,*fshift,*x,*f;
 512     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 513     real             scratch[4*DIM];
 514     __m128           fscal,rcutoff,rcutoff2,jidxall;
 515     int              vdwioffset1;
 516     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 517     int              vdwioffset2;
 518     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 519     int              vdwioffset3;
 520     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 521     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 522     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 523     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 524     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 525     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 526     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 527     real             *charge;
 528     __m128           dummy_mask,cutoff_mask;
 529     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 530     __m128           one     = _mm_set1_ps(1.0);
 531     __m128           two     = _mm_set1_ps(2.0);
 532     x                = xx[0];
 533     f                = ff[0];
 534
 535     nri              = nlist->nri;
 536     iinr             = nlist->iinr;
 537     jindex           = nlist->jindex;
 538     jjnr             = nlist->jjnr;
 539     shiftidx         = nlist->shift;
 540     gid              = nlist->gid;
 541     shiftvec         = fr->shift_vec[0];
 542     fshift           = fr->fshift[0];
 543     facel            = _mm_set1_ps(fr->epsfac);
 544     charge           = mdatoms->chargeA;
 545     krf              = _mm_set1_ps(fr->ic->k_rf);
 546     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 547     crf              = _mm_set1_ps(fr->ic->c_rf);
 548
 549     /* Setup water-specific parameters */
 550     inr              = nlist->iinr[0];
 551     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 552     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 553     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 554
 555     /* Avoid stupid compiler warnings */
 556     jnrA = jnrB = jnrC = jnrD = 0;
 557     j_coord_offsetA = 0;
 558     j_coord_offsetB = 0;
 559     j_coord_offsetC = 0;
 560     j_coord_offsetD = 0;
 561
 562     outeriter        = 0;
 563     inneriter        = 0;
 564
 565     for(iidx=0;iidx<4*DIM;iidx++)
 566     {
 567         scratch[iidx] = 0.0;
 568     }
 569
 570     /* Start outer loop over neighborlists */
 571     for(iidx=0; iidx<nri; iidx++)
 572     {
 573         /* Load shift vector for this list */
 574         i_shift_offset   = DIM*shiftidx[iidx];
 575
 576         /* Load limits for loop over neighbors */
 577         j_index_start    = jindex[iidx];
 578         j_index_end      = jindex[iidx+1];
 579
 580         /* Get outer coordinate index */
 581         inr              = iinr[iidx];
 582         i_coord_offset   = DIM*inr;
 583
 584         /* Load i particle coords and add shift vector */
 585         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 586                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 587
 588         fix1             = _mm_setzero_ps();
 589         fiy1             = _mm_setzero_ps();
 590         fiz1             = _mm_setzero_ps();
 591         fix2             = _mm_setzero_ps();
 592         fiy2             = _mm_setzero_ps();
 593         fiz2             = _mm_setzero_ps();
 594         fix3             = _mm_setzero_ps();
 595         fiy3             = _mm_setzero_ps();
 596         fiz3             = _mm_setzero_ps();
 597
 598         /* Start inner kernel loop */
 599         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 600         {
 601
 602             /* Get j neighbor index, and coordinate index */
 603             jnrA             = jjnr[jidx];
 604             jnrB             = jjnr[jidx+1];
 605             jnrC             = jjnr[jidx+2];
 606             jnrD             = jjnr[jidx+3];
 607             j_coord_offsetA  = DIM*jnrA;
 608             j_coord_offsetB  = DIM*jnrB;
 609             j_coord_offsetC  = DIM*jnrC;
 610             j_coord_offsetD  = DIM*jnrD;
 611
 612             /* load j atom coordinates */
 613             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 614                                               x+j_coord_offsetC,x+j_coord_offsetD,
 615                                               &jx0,&jy0,&jz0);
 616
 617             /* Calculate displacement vector */
 618             dx10             = _mm_sub_ps(ix1,jx0);
 619             dy10             = _mm_sub_ps(iy1,jy0);
 620             dz10             = _mm_sub_ps(iz1,jz0);
 621             dx20             = _mm_sub_ps(ix2,jx0);
 622             dy20             = _mm_sub_ps(iy2,jy0);
 623             dz20             = _mm_sub_ps(iz2,jz0);
 624             dx30             = _mm_sub_ps(ix3,jx0);
 625             dy30             = _mm_sub_ps(iy3,jy0);
 626             dz30             = _mm_sub_ps(iz3,jz0);
 627
 628             /* Calculate squared distance and things based on it */
 629             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 630             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 631             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 632
 633             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 634             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 635             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 636
 637             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 638             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 639             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 640
 641             /* Load parameters for j particles */
 642             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 643                                                               charge+jnrC+0,charge+jnrD+0);
 644
 645             fjx0             = _mm_setzero_ps();
 646             fjy0             = _mm_setzero_ps();
 647             fjz0             = _mm_setzero_ps();
 648
 649             /**************************
 650              * CALCULATE INTERACTIONS *
 651              **************************/
 652
 653             /* Compute parameters for interactions between i and j atoms */
 654             qq10             = _mm_mul_ps(iq1,jq0);
 655
 656             /* REACTION-FIELD ELECTROSTATICS */
 657             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 658
 659             fscal            = felec;
 660
 661              /* Update vectorial force */
 662             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 663             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 664             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 665
 666             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 667             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 668             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             /* Compute parameters for interactions between i and j atoms */
 675             qq20             = _mm_mul_ps(iq2,jq0);
 676
 677             /* REACTION-FIELD ELECTROSTATICS */
 678             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 679
 680             fscal            = felec;
 681
 682              /* Update vectorial force */
 683             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 684             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 685             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 686
 687             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 688             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 689             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 690
 691             /**************************
 692              * CALCULATE INTERACTIONS *
 693              **************************/
 694
 695             /* Compute parameters for interactions between i and j atoms */
 696             qq30             = _mm_mul_ps(iq3,jq0);
 697
 698             /* REACTION-FIELD ELECTROSTATICS */
 699             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 700
 701             fscal            = felec;
 702
 703              /* Update vectorial force */
 704             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 705             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 706             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 707
 708             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 709             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 710             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 711
 712             fjptrA             = f+j_coord_offsetA;
 713             fjptrB             = f+j_coord_offsetB;
 714             fjptrC             = f+j_coord_offsetC;
 715             fjptrD             = f+j_coord_offsetD;
 716
 717             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 718
 719             /* Inner loop uses 90 flops */
 720         }
 721
 722         if(jidx<j_index_end)
 723         {
 724
 725             /* Get j neighbor index, and coordinate index */
 726             jnrlistA         = jjnr[jidx];
 727             jnrlistB         = jjnr[jidx+1];
 728             jnrlistC         = jjnr[jidx+2];
 729             jnrlistD         = jjnr[jidx+3];
 730             /* Sign of each element will be negative for non-real atoms.
 731              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 732              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 733              */
 734             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 735             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 736             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 737             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 738             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 739             j_coord_offsetA  = DIM*jnrA;
 740             j_coord_offsetB  = DIM*jnrB;
 741             j_coord_offsetC  = DIM*jnrC;
 742             j_coord_offsetD  = DIM*jnrD;
 743
 744             /* load j atom coordinates */
 745             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 746                                               x+j_coord_offsetC,x+j_coord_offsetD,
 747                                               &jx0,&jy0,&jz0);
 748
 749             /* Calculate displacement vector */
 750             dx10             = _mm_sub_ps(ix1,jx0);
 751             dy10             = _mm_sub_ps(iy1,jy0);
 752             dz10             = _mm_sub_ps(iz1,jz0);
 753             dx20             = _mm_sub_ps(ix2,jx0);
 754             dy20             = _mm_sub_ps(iy2,jy0);
 755             dz20             = _mm_sub_ps(iz2,jz0);
 756             dx30             = _mm_sub_ps(ix3,jx0);
 757             dy30             = _mm_sub_ps(iy3,jy0);
 758             dz30             = _mm_sub_ps(iz3,jz0);
 759
 760             /* Calculate squared distance and things based on it */
 761             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 762             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 763             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 764
 765             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 766             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 767             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 768
 769             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 770             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 771             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 772
 773             /* Load parameters for j particles */
 774             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 775                                                               charge+jnrC+0,charge+jnrD+0);
 776
 777             fjx0             = _mm_setzero_ps();
 778             fjy0             = _mm_setzero_ps();
 779             fjz0             = _mm_setzero_ps();
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             /* Compute parameters for interactions between i and j atoms */
 786             qq10             = _mm_mul_ps(iq1,jq0);
 787
 788             /* REACTION-FIELD ELECTROSTATICS */
 789             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 790
 791             fscal            = felec;
 792
 793             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 794
 795              /* Update vectorial force */
 796             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 797             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 798             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 799
 800             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 801             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 802             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 803
 804             /**************************
 805              * CALCULATE INTERACTIONS *
 806              **************************/
 807
 808             /* Compute parameters for interactions between i and j atoms */
 809             qq20             = _mm_mul_ps(iq2,jq0);
 810
 811             /* REACTION-FIELD ELECTROSTATICS */
 812             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 813
 814             fscal            = felec;
 815
 816             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 817
 818              /* Update vectorial force */
 819             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 820             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 821             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 822
 823             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 824             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 825             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 826
 827             /**************************
 828              * CALCULATE INTERACTIONS *
 829              **************************/
 830
 831             /* Compute parameters for interactions between i and j atoms */
 832             qq30             = _mm_mul_ps(iq3,jq0);
 833
 834             /* REACTION-FIELD ELECTROSTATICS */
 835             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 836
 837             fscal            = felec;
 838
 839             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 840
 841              /* Update vectorial force */
 842             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 843             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 844             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 845
 846             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 847             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 848             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 849
 850             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 851             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 852             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 853             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 854
 855             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 856
 857             /* Inner loop uses 90 flops */
 858         }
 859
 860         /* End of innermost loop */
 861
 862         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 863                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 864
 865         /* Increment number of inner iterations */
 866         inneriter                  += j_index_end - j_index_start;
 867
 868         /* Outer loop uses 18 flops */
 869     }
 870
 871     /* Increment number of outer iterations */
 872     outeriter        += nri;
 873
 874     /* Update outer/inner flops */
 875
 876     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*90);
 877 }