src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSh_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_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              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  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           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     krf              = _mm_set1_ps(fr->ic->k_rf);
 123     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 124     crf              = _mm_set1_ps(fr->ic->c_rf);
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 132     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 133     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 137     rcutoff_scalar   = fr->rcoulomb;
 138     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 139     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 140
 141     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 142     rvdw             = _mm_set1_ps(fr->rvdw);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 176
 177         fix0             = _mm_setzero_ps();
 178         fiy0             = _mm_setzero_ps();
 179         fiz0             = _mm_setzero_ps();
 180         fix1             = _mm_setzero_ps();
 181         fiy1             = _mm_setzero_ps();
 182         fiz1             = _mm_setzero_ps();
 183         fix2             = _mm_setzero_ps();
 184         fiy2             = _mm_setzero_ps();
 185         fiz2             = _mm_setzero_ps();
 186         fix3             = _mm_setzero_ps();
 187         fiy3             = _mm_setzero_ps();
 188         fiz3             = _mm_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm_setzero_ps();
 192         vvdwsum          = _mm_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                               x+j_coord_offsetC,x+j_coord_offsetD,
 211                                               &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm_sub_ps(ix0,jx0);
 215             dy00             = _mm_sub_ps(iy0,jy0);
 216             dz00             = _mm_sub_ps(iz0,jz0);
 217             dx10             = _mm_sub_ps(ix1,jx0);
 218             dy10             = _mm_sub_ps(iy1,jy0);
 219             dz10             = _mm_sub_ps(iz1,jz0);
 220             dx20             = _mm_sub_ps(ix2,jx0);
 221             dy20             = _mm_sub_ps(iy2,jy0);
 222             dz20             = _mm_sub_ps(iz2,jz0);
 223             dx30             = _mm_sub_ps(ix3,jx0);
 224             dy30             = _mm_sub_ps(iy3,jy0);
 225             dz30             = _mm_sub_ps(iz3,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 229             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 230             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 231             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 232
 233             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 234             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 235             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 236
 237             rinvsq00         = gmx_mm_inv_ps(rsq00);
 238             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 239             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 240             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 244                                                               charge+jnrC+0,charge+jnrD+0);
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246             vdwjidx0B        = 2*vdwtype[jnrB+0];
 247             vdwjidx0C        = 2*vdwtype[jnrC+0];
 248             vdwjidx0D        = 2*vdwtype[jnrD+0];
 249
 250             fjx0             = _mm_setzero_ps();
 251             fjy0             = _mm_setzero_ps();
 252             fjz0             = _mm_setzero_ps();
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             if (gmx_mm_any_lt(rsq00,rcutoff2))
 259             {
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 263                                          vdwparam+vdwioffset0+vdwjidx0B,
 264                                          vdwparam+vdwioffset0+vdwjidx0C,
 265                                          vdwparam+vdwioffset0+vdwjidx0D,
 266                                          &c6_00,&c12_00);
 267
 268             /* LENNARD-JONES DISPERSION/REPULSION */
 269
 270             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 271             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 272             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 273             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 274                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 275             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 276
 277             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 281             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 282
 283             fscal            = fvdw;
 284
 285             fscal            = _mm_and_ps(fscal,cutoff_mask);
 286
 287              /* Update vectorial force */
 288             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 289             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 290             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 291
 292             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 293             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 294             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 295
 296             }
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             if (gmx_mm_any_lt(rsq10,rcutoff2))
 303             {
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_ps(iq1,jq0);
 307
 308             /* REACTION-FIELD ELECTROSTATICS */
 309             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 310             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 311
 312             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velec            = _mm_and_ps(velec,cutoff_mask);
 316             velecsum         = _mm_add_ps(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320             fscal            = _mm_and_ps(fscal,cutoff_mask);
 321
 322              /* Update vectorial force */
 323             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 324             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 325             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 326
 327             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 328             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 329             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 330
 331             }
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             if (gmx_mm_any_lt(rsq20,rcutoff2))
 338             {
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq20             = _mm_mul_ps(iq2,jq0);
 342
 343             /* REACTION-FIELD ELECTROSTATICS */
 344             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 345             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 346
 347             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velec            = _mm_and_ps(velec,cutoff_mask);
 351             velecsum         = _mm_add_ps(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             fscal            = _mm_and_ps(fscal,cutoff_mask);
 356
 357              /* Update vectorial force */
 358             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 359             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 360             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 361
 362             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 363             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 364             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 365
 366             }
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             if (gmx_mm_any_lt(rsq30,rcutoff2))
 373             {
 374
 375             /* Compute parameters for interactions between i and j atoms */
 376             qq30             = _mm_mul_ps(iq3,jq0);
 377
 378             /* REACTION-FIELD ELECTROSTATICS */
 379             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 380             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 381
 382             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velec            = _mm_and_ps(velec,cutoff_mask);
 386             velecsum         = _mm_add_ps(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             fscal            = _mm_and_ps(fscal,cutoff_mask);
 391
 392              /* Update vectorial force */
 393             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 394             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 395             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 396
 397             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 398             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 399             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 400
 401             }
 402
 403             fjptrA             = f+j_coord_offsetA;
 404             fjptrB             = f+j_coord_offsetB;
 405             fjptrC             = f+j_coord_offsetC;
 406             fjptrD             = f+j_coord_offsetD;
 407
 408             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 409
 410             /* Inner loop uses 161 flops */
 411         }
 412
 413         if(jidx<j_index_end)
 414         {
 415
 416             /* Get j neighbor index, and coordinate index */
 417             jnrlistA         = jjnr[jidx];
 418             jnrlistB         = jjnr[jidx+1];
 419             jnrlistC         = jjnr[jidx+2];
 420             jnrlistD         = jjnr[jidx+3];
 421             /* Sign of each element will be negative for non-real atoms.
 422              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 423              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 424              */
 425             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 426             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 427             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 428             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 429             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 430             j_coord_offsetA  = DIM*jnrA;
 431             j_coord_offsetB  = DIM*jnrB;
 432             j_coord_offsetC  = DIM*jnrC;
 433             j_coord_offsetD  = DIM*jnrD;
 434
 435             /* load j atom coordinates */
 436             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 437                                               x+j_coord_offsetC,x+j_coord_offsetD,
 438                                               &jx0,&jy0,&jz0);
 439
 440             /* Calculate displacement vector */
 441             dx00             = _mm_sub_ps(ix0,jx0);
 442             dy00             = _mm_sub_ps(iy0,jy0);
 443             dz00             = _mm_sub_ps(iz0,jz0);
 444             dx10             = _mm_sub_ps(ix1,jx0);
 445             dy10             = _mm_sub_ps(iy1,jy0);
 446             dz10             = _mm_sub_ps(iz1,jz0);
 447             dx20             = _mm_sub_ps(ix2,jx0);
 448             dy20             = _mm_sub_ps(iy2,jy0);
 449             dz20             = _mm_sub_ps(iz2,jz0);
 450             dx30             = _mm_sub_ps(ix3,jx0);
 451             dy30             = _mm_sub_ps(iy3,jy0);
 452             dz30             = _mm_sub_ps(iz3,jz0);
 453
 454             /* Calculate squared distance and things based on it */
 455             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 456             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 457             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 458             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 459
 460             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 461             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 462             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 463
 464             rinvsq00         = gmx_mm_inv_ps(rsq00);
 465             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 466             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 467             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 468
 469             /* Load parameters for j particles */
 470             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 471                                                               charge+jnrC+0,charge+jnrD+0);
 472             vdwjidx0A        = 2*vdwtype[jnrA+0];
 473             vdwjidx0B        = 2*vdwtype[jnrB+0];
 474             vdwjidx0C        = 2*vdwtype[jnrC+0];
 475             vdwjidx0D        = 2*vdwtype[jnrD+0];
 476
 477             fjx0             = _mm_setzero_ps();
 478             fjy0             = _mm_setzero_ps();
 479             fjz0             = _mm_setzero_ps();
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             if (gmx_mm_any_lt(rsq00,rcutoff2))
 486             {
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 490                                          vdwparam+vdwioffset0+vdwjidx0B,
 491                                          vdwparam+vdwioffset0+vdwjidx0C,
 492                                          vdwparam+vdwioffset0+vdwjidx0D,
 493                                          &c6_00,&c12_00);
 494
 495             /* LENNARD-JONES DISPERSION/REPULSION */
 496
 497             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 498             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 499             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 500             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 501                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 502             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 503
 504             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 508             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 509             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 510
 511             fscal            = fvdw;
 512
 513             fscal            = _mm_and_ps(fscal,cutoff_mask);
 514
 515             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 516
 517              /* Update vectorial force */
 518             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 519             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 520             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 521
 522             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 523             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 524             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 525
 526             }
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             if (gmx_mm_any_lt(rsq10,rcutoff2))
 533             {
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq10             = _mm_mul_ps(iq1,jq0);
 537
 538             /* REACTION-FIELD ELECTROSTATICS */
 539             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 540             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 541
 542             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm_and_ps(velec,cutoff_mask);
 546             velec            = _mm_andnot_ps(dummy_mask,velec);
 547             velecsum         = _mm_add_ps(velecsum,velec);
 548
 549             fscal            = felec;
 550
 551             fscal            = _mm_and_ps(fscal,cutoff_mask);
 552
 553             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 554
 555              /* Update vectorial force */
 556             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 557             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 558             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 559
 560             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 561             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 562             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 563
 564             }
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             if (gmx_mm_any_lt(rsq20,rcutoff2))
 571             {
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq20             = _mm_mul_ps(iq2,jq0);
 575
 576             /* REACTION-FIELD ELECTROSTATICS */
 577             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 578             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 579
 580             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm_and_ps(velec,cutoff_mask);
 584             velec            = _mm_andnot_ps(dummy_mask,velec);
 585             velecsum         = _mm_add_ps(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm_and_ps(fscal,cutoff_mask);
 590
 591             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 592
 593              /* Update vectorial force */
 594             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 595             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 596             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 597
 598             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 599             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 600             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 601
 602             }
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             if (gmx_mm_any_lt(rsq30,rcutoff2))
 609             {
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq30             = _mm_mul_ps(iq3,jq0);
 613
 614             /* REACTION-FIELD ELECTROSTATICS */
 615             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 616             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 617
 618             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 619
 620             /* Update potential sum for this i atom from the interaction with this j atom. */
 621             velec            = _mm_and_ps(velec,cutoff_mask);
 622             velec            = _mm_andnot_ps(dummy_mask,velec);
 623             velecsum         = _mm_add_ps(velecsum,velec);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm_and_ps(fscal,cutoff_mask);
 628
 629             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 630
 631              /* Update vectorial force */
 632             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 633             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 634             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 635
 636             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 637             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 638             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 639
 640             }
 641
 642             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 643             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 644             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 645             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 646
 647             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 648
 649             /* Inner loop uses 161 flops */
 650         }
 651
 652         /* End of innermost loop */
 653
 654         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 655                                               f+i_coord_offset,fshift+i_shift_offset);
 656
 657         ggid                        = gid[iidx];
 658         /* Update potential energies */
 659         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 660         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 661
 662         /* Increment number of inner iterations */
 663         inneriter                  += j_index_end - j_index_start;
 664
 665         /* Outer loop uses 26 flops */
 666     }
 667
 668     /* Increment number of outer iterations */
 669     outeriter        += nri;
 670
 671     /* Update outer/inner flops */
 672
 673     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*161);
 674 }
 675 /*
 676  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_128_fma_single
 677  * Electrostatics interaction: ReactionField
 678  * VdW interaction:            LennardJones
 679  * Geometry:                   Water4-Particle
 680  * Calculate force/pot:        Force
 681  */
 682 void
 683 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_128_fma_single
 684                     (t_nblist                    * gmx_restrict       nlist,
 685                      rvec                        * gmx_restrict          xx,
 686                      rvec                        * gmx_restrict          ff,
 687                      t_forcerec                  * gmx_restrict          fr,
 688                      t_mdatoms                   * gmx_restrict     mdatoms,
 689                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 690                      t_nrnb                      * gmx_restrict        nrnb)
 691 {
 692     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 693      * just 0 for non-waters.
 694      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 695      * jnr indices corresponding to data put in the four positions in the SIMD register.
 696      */
 697     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 698     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 699     int              jnrA,jnrB,jnrC,jnrD;
 700     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 701     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 702     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 703     real             rcutoff_scalar;
 704     real             *shiftvec,*fshift,*x,*f;
 705     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 706     real             scratch[4*DIM];
 707     __m128           fscal,rcutoff,rcutoff2,jidxall;
 708     int              vdwioffset0;
 709     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 710     int              vdwioffset1;
 711     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 712     int              vdwioffset2;
 713     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 714     int              vdwioffset3;
 715     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 716     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 717     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 718     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 719     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 720     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 721     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 722     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 723     real             *charge;
 724     int              nvdwtype;
 725     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 726     int              *vdwtype;
 727     real             *vdwparam;
 728     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 729     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 730     __m128           dummy_mask,cutoff_mask;
 731     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 732     __m128           one     = _mm_set1_ps(1.0);
 733     __m128           two     = _mm_set1_ps(2.0);
 734     x                = xx[0];
 735     f                = ff[0];
 736
 737     nri              = nlist->nri;
 738     iinr             = nlist->iinr;
 739     jindex           = nlist->jindex;
 740     jjnr             = nlist->jjnr;
 741     shiftidx         = nlist->shift;
 742     gid              = nlist->gid;
 743     shiftvec         = fr->shift_vec[0];
 744     fshift           = fr->fshift[0];
 745     facel            = _mm_set1_ps(fr->epsfac);
 746     charge           = mdatoms->chargeA;
 747     krf              = _mm_set1_ps(fr->ic->k_rf);
 748     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 749     crf              = _mm_set1_ps(fr->ic->c_rf);
 750     nvdwtype         = fr->ntype;
 751     vdwparam         = fr->nbfp;
 752     vdwtype          = mdatoms->typeA;
 753
 754     /* Setup water-specific parameters */
 755     inr              = nlist->iinr[0];
 756     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 757     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 758     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 759     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 760
 761     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 762     rcutoff_scalar   = fr->rcoulomb;
 763     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 764     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 765
 766     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 767     rvdw             = _mm_set1_ps(fr->rvdw);
 768
 769     /* Avoid stupid compiler warnings */
 770     jnrA = jnrB = jnrC = jnrD = 0;
 771     j_coord_offsetA = 0;
 772     j_coord_offsetB = 0;
 773     j_coord_offsetC = 0;
 774     j_coord_offsetD = 0;
 775
 776     outeriter        = 0;
 777     inneriter        = 0;
 778
 779     for(iidx=0;iidx<4*DIM;iidx++)
 780     {
 781         scratch[iidx] = 0.0;
 782     }
 783
 784     /* Start outer loop over neighborlists */
 785     for(iidx=0; iidx<nri; iidx++)
 786     {
 787         /* Load shift vector for this list */
 788         i_shift_offset   = DIM*shiftidx[iidx];
 789
 790         /* Load limits for loop over neighbors */
 791         j_index_start    = jindex[iidx];
 792         j_index_end      = jindex[iidx+1];
 793
 794         /* Get outer coordinate index */
 795         inr              = iinr[iidx];
 796         i_coord_offset   = DIM*inr;
 797
 798         /* Load i particle coords and add shift vector */
 799         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 800                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 801
 802         fix0             = _mm_setzero_ps();
 803         fiy0             = _mm_setzero_ps();
 804         fiz0             = _mm_setzero_ps();
 805         fix1             = _mm_setzero_ps();
 806         fiy1             = _mm_setzero_ps();
 807         fiz1             = _mm_setzero_ps();
 808         fix2             = _mm_setzero_ps();
 809         fiy2             = _mm_setzero_ps();
 810         fiz2             = _mm_setzero_ps();
 811         fix3             = _mm_setzero_ps();
 812         fiy3             = _mm_setzero_ps();
 813         fiz3             = _mm_setzero_ps();
 814
 815         /* Start inner kernel loop */
 816         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 817         {
 818
 819             /* Get j neighbor index, and coordinate index */
 820             jnrA             = jjnr[jidx];
 821             jnrB             = jjnr[jidx+1];
 822             jnrC             = jjnr[jidx+2];
 823             jnrD             = jjnr[jidx+3];
 824             j_coord_offsetA  = DIM*jnrA;
 825             j_coord_offsetB  = DIM*jnrB;
 826             j_coord_offsetC  = DIM*jnrC;
 827             j_coord_offsetD  = DIM*jnrD;
 828
 829             /* load j atom coordinates */
 830             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 831                                               x+j_coord_offsetC,x+j_coord_offsetD,
 832                                               &jx0,&jy0,&jz0);
 833
 834             /* Calculate displacement vector */
 835             dx00             = _mm_sub_ps(ix0,jx0);
 836             dy00             = _mm_sub_ps(iy0,jy0);
 837             dz00             = _mm_sub_ps(iz0,jz0);
 838             dx10             = _mm_sub_ps(ix1,jx0);
 839             dy10             = _mm_sub_ps(iy1,jy0);
 840             dz10             = _mm_sub_ps(iz1,jz0);
 841             dx20             = _mm_sub_ps(ix2,jx0);
 842             dy20             = _mm_sub_ps(iy2,jy0);
 843             dz20             = _mm_sub_ps(iz2,jz0);
 844             dx30             = _mm_sub_ps(ix3,jx0);
 845             dy30             = _mm_sub_ps(iy3,jy0);
 846             dz30             = _mm_sub_ps(iz3,jz0);
 847
 848             /* Calculate squared distance and things based on it */
 849             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 850             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 851             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 852             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 853
 854             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 855             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 856             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 857
 858             rinvsq00         = gmx_mm_inv_ps(rsq00);
 859             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 860             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 861             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 862
 863             /* Load parameters for j particles */
 864             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 865                                                               charge+jnrC+0,charge+jnrD+0);
 866             vdwjidx0A        = 2*vdwtype[jnrA+0];
 867             vdwjidx0B        = 2*vdwtype[jnrB+0];
 868             vdwjidx0C        = 2*vdwtype[jnrC+0];
 869             vdwjidx0D        = 2*vdwtype[jnrD+0];
 870
 871             fjx0             = _mm_setzero_ps();
 872             fjy0             = _mm_setzero_ps();
 873             fjz0             = _mm_setzero_ps();
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             if (gmx_mm_any_lt(rsq00,rcutoff2))
 880             {
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 884                                          vdwparam+vdwioffset0+vdwjidx0B,
 885                                          vdwparam+vdwioffset0+vdwjidx0C,
 886                                          vdwparam+vdwioffset0+vdwjidx0D,
 887                                          &c6_00,&c12_00);
 888
 889             /* LENNARD-JONES DISPERSION/REPULSION */
 890
 891             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 892             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 893
 894             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 895
 896             fscal            = fvdw;
 897
 898             fscal            = _mm_and_ps(fscal,cutoff_mask);
 899
 900              /* Update vectorial force */
 901             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 902             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 903             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 904
 905             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 906             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 907             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 908
 909             }
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             if (gmx_mm_any_lt(rsq10,rcutoff2))
 916             {
 917
 918             /* Compute parameters for interactions between i and j atoms */
 919             qq10             = _mm_mul_ps(iq1,jq0);
 920
 921             /* REACTION-FIELD ELECTROSTATICS */
 922             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 923
 924             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 925
 926             fscal            = felec;
 927
 928             fscal            = _mm_and_ps(fscal,cutoff_mask);
 929
 930              /* Update vectorial force */
 931             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 932             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 933             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 934
 935             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 936             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 937             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 938
 939             }
 940
 941             /**************************
 942              * CALCULATE INTERACTIONS *
 943              **************************/
 944
 945             if (gmx_mm_any_lt(rsq20,rcutoff2))
 946             {
 947
 948             /* Compute parameters for interactions between i and j atoms */
 949             qq20             = _mm_mul_ps(iq2,jq0);
 950
 951             /* REACTION-FIELD ELECTROSTATICS */
 952             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 953
 954             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 955
 956             fscal            = felec;
 957
 958             fscal            = _mm_and_ps(fscal,cutoff_mask);
 959
 960              /* Update vectorial force */
 961             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 962             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 963             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 964
 965             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 966             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 967             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 968
 969             }
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             if (gmx_mm_any_lt(rsq30,rcutoff2))
 976             {
 977
 978             /* Compute parameters for interactions between i and j atoms */
 979             qq30             = _mm_mul_ps(iq3,jq0);
 980
 981             /* REACTION-FIELD ELECTROSTATICS */
 982             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 983
 984             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 985
 986             fscal            = felec;
 987
 988             fscal            = _mm_and_ps(fscal,cutoff_mask);
 989
 990              /* Update vectorial force */
 991             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 992             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 993             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 994
 995             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 996             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 997             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 998
 999             }
1000
1001             fjptrA             = f+j_coord_offsetA;
1002             fjptrB             = f+j_coord_offsetB;
1003             fjptrC             = f+j_coord_offsetC;
1004             fjptrD             = f+j_coord_offsetD;
1005
1006             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1007
1008             /* Inner loop uses 132 flops */
1009         }
1010
1011         if(jidx<j_index_end)
1012         {
1013
1014             /* Get j neighbor index, and coordinate index */
1015             jnrlistA         = jjnr[jidx];
1016             jnrlistB         = jjnr[jidx+1];
1017             jnrlistC         = jjnr[jidx+2];
1018             jnrlistD         = jjnr[jidx+3];
1019             /* Sign of each element will be negative for non-real atoms.
1020              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1021              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1022              */
1023             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1024             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1025             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1026             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1027             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1028             j_coord_offsetA  = DIM*jnrA;
1029             j_coord_offsetB  = DIM*jnrB;
1030             j_coord_offsetC  = DIM*jnrC;
1031             j_coord_offsetD  = DIM*jnrD;
1032
1033             /* load j atom coordinates */
1034             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1035                                               x+j_coord_offsetC,x+j_coord_offsetD,
1036                                               &jx0,&jy0,&jz0);
1037
1038             /* Calculate displacement vector */
1039             dx00             = _mm_sub_ps(ix0,jx0);
1040             dy00             = _mm_sub_ps(iy0,jy0);
1041             dz00             = _mm_sub_ps(iz0,jz0);
1042             dx10             = _mm_sub_ps(ix1,jx0);
1043             dy10             = _mm_sub_ps(iy1,jy0);
1044             dz10             = _mm_sub_ps(iz1,jz0);
1045             dx20             = _mm_sub_ps(ix2,jx0);
1046             dy20             = _mm_sub_ps(iy2,jy0);
1047             dz20             = _mm_sub_ps(iz2,jz0);
1048             dx30             = _mm_sub_ps(ix3,jx0);
1049             dy30             = _mm_sub_ps(iy3,jy0);
1050             dz30             = _mm_sub_ps(iz3,jz0);
1051
1052             /* Calculate squared distance and things based on it */
1053             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1054             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1055             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1056             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1057
1058             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1059             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1060             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1061
1062             rinvsq00         = gmx_mm_inv_ps(rsq00);
1063             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1064             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1065             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1066
1067             /* Load parameters for j particles */
1068             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1069                                                               charge+jnrC+0,charge+jnrD+0);
1070             vdwjidx0A        = 2*vdwtype[jnrA+0];
1071             vdwjidx0B        = 2*vdwtype[jnrB+0];
1072             vdwjidx0C        = 2*vdwtype[jnrC+0];
1073             vdwjidx0D        = 2*vdwtype[jnrD+0];
1074
1075             fjx0             = _mm_setzero_ps();
1076             fjy0             = _mm_setzero_ps();
1077             fjz0             = _mm_setzero_ps();
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             if (gmx_mm_any_lt(rsq00,rcutoff2))
1084             {
1085
1086             /* Compute parameters for interactions between i and j atoms */
1087             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1088                                          vdwparam+vdwioffset0+vdwjidx0B,
1089                                          vdwparam+vdwioffset0+vdwjidx0C,
1090                                          vdwparam+vdwioffset0+vdwjidx0D,
1091                                          &c6_00,&c12_00);
1092
1093             /* LENNARD-JONES DISPERSION/REPULSION */
1094
1095             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1096             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1097
1098             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1099
1100             fscal            = fvdw;
1101
1102             fscal            = _mm_and_ps(fscal,cutoff_mask);
1103
1104             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1105
1106              /* Update vectorial force */
1107             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1108             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1109             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1110
1111             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1112             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1113             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1114
1115             }
1116
1117             /**************************
1118              * CALCULATE INTERACTIONS *
1119              **************************/
1120
1121             if (gmx_mm_any_lt(rsq10,rcutoff2))
1122             {
1123
1124             /* Compute parameters for interactions between i and j atoms */
1125             qq10             = _mm_mul_ps(iq1,jq0);
1126
1127             /* REACTION-FIELD ELECTROSTATICS */
1128             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1129
1130             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1131
1132             fscal            = felec;
1133
1134             fscal            = _mm_and_ps(fscal,cutoff_mask);
1135
1136             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1137
1138              /* Update vectorial force */
1139             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1140             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1141             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1142
1143             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1144             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1145             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1146
1147             }
1148
1149             /**************************
1150              * CALCULATE INTERACTIONS *
1151              **************************/
1152
1153             if (gmx_mm_any_lt(rsq20,rcutoff2))
1154             {
1155
1156             /* Compute parameters for interactions between i and j atoms */
1157             qq20             = _mm_mul_ps(iq2,jq0);
1158
1159             /* REACTION-FIELD ELECTROSTATICS */
1160             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1161
1162             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1163
1164             fscal            = felec;
1165
1166             fscal            = _mm_and_ps(fscal,cutoff_mask);
1167
1168             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1169
1170              /* Update vectorial force */
1171             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1172             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1173             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1174
1175             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1176             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1177             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1178
1179             }
1180
1181             /**************************
1182              * CALCULATE INTERACTIONS *
1183              **************************/
1184
1185             if (gmx_mm_any_lt(rsq30,rcutoff2))
1186             {
1187
1188             /* Compute parameters for interactions between i and j atoms */
1189             qq30             = _mm_mul_ps(iq3,jq0);
1190
1191             /* REACTION-FIELD ELECTROSTATICS */
1192             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1193
1194             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1195
1196             fscal            = felec;
1197
1198             fscal            = _mm_and_ps(fscal,cutoff_mask);
1199
1200             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1201
1202              /* Update vectorial force */
1203             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1204             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1205             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1206
1207             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1208             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1209             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1210
1211             }
1212
1213             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1214             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1215             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1216             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1217
1218             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1219
1220             /* Inner loop uses 132 flops */
1221         }
1222
1223         /* End of innermost loop */
1224
1225         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1226                                               f+i_coord_offset,fshift+i_shift_offset);
1227
1228         /* Increment number of inner iterations */
1229         inneriter                  += j_index_end - j_index_start;
1230
1231         /* Outer loop uses 24 flops */
1232     }
1233
1234     /* Increment number of outer iterations */
1235     outeriter        += nri;
1236
1237     /* Update outer/inner flops */
1238
1239     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1240 }