src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwNone_GeomW3W3_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_ElecRF_VdwNone_GeomW3W3_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwNone_GeomW3W3_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  92     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  93     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  94     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  97     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  98     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  99     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 100     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 101     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 102     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 103     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 104     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 105     real             *charge;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm_set1_ps(fr->ic->k_rf);
 124     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_ps(fr->ic->c_rf);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 130     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 131     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 132
 133     jq0              = _mm_set1_ps(charge[inr+0]);
 134     jq1              = _mm_set1_ps(charge[inr+1]);
 135     jq2              = _mm_set1_ps(charge[inr+2]);
 136     qq00             = _mm_mul_ps(iq0,jq0);
 137     qq01             = _mm_mul_ps(iq0,jq1);
 138     qq02             = _mm_mul_ps(iq0,jq2);
 139     qq10             = _mm_mul_ps(iq1,jq0);
 140     qq11             = _mm_mul_ps(iq1,jq1);
 141     qq12             = _mm_mul_ps(iq1,jq2);
 142     qq20             = _mm_mul_ps(iq2,jq0);
 143     qq21             = _mm_mul_ps(iq2,jq1);
 144     qq22             = _mm_mul_ps(iq2,jq2);
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     for(iidx=0;iidx<4*DIM;iidx++)
 157     {
 158         scratch[iidx] = 0.0;
 159     }
 160
 161     /* Start outer loop over neighborlists */
 162     for(iidx=0; iidx<nri; iidx++)
 163     {
 164         /* Load shift vector for this list */
 165         i_shift_offset   = DIM*shiftidx[iidx];
 166
 167         /* Load limits for loop over neighbors */
 168         j_index_start    = jindex[iidx];
 169         j_index_end      = jindex[iidx+1];
 170
 171         /* Get outer coordinate index */
 172         inr              = iinr[iidx];
 173         i_coord_offset   = DIM*inr;
 174
 175         /* Load i particle coords and add shift vector */
 176         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 177                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 178
 179         fix0             = _mm_setzero_ps();
 180         fiy0             = _mm_setzero_ps();
 181         fiz0             = _mm_setzero_ps();
 182         fix1             = _mm_setzero_ps();
 183         fiy1             = _mm_setzero_ps();
 184         fiz1             = _mm_setzero_ps();
 185         fix2             = _mm_setzero_ps();
 186         fiy2             = _mm_setzero_ps();
 187         fiz2             = _mm_setzero_ps();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               x+j_coord_offsetC,x+j_coord_offsetD,
 209                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm_sub_ps(ix0,jx0);
 213             dy00             = _mm_sub_ps(iy0,jy0);
 214             dz00             = _mm_sub_ps(iz0,jz0);
 215             dx01             = _mm_sub_ps(ix0,jx1);
 216             dy01             = _mm_sub_ps(iy0,jy1);
 217             dz01             = _mm_sub_ps(iz0,jz1);
 218             dx02             = _mm_sub_ps(ix0,jx2);
 219             dy02             = _mm_sub_ps(iy0,jy2);
 220             dz02             = _mm_sub_ps(iz0,jz2);
 221             dx10             = _mm_sub_ps(ix1,jx0);
 222             dy10             = _mm_sub_ps(iy1,jy0);
 223             dz10             = _mm_sub_ps(iz1,jz0);
 224             dx11             = _mm_sub_ps(ix1,jx1);
 225             dy11             = _mm_sub_ps(iy1,jy1);
 226             dz11             = _mm_sub_ps(iz1,jz1);
 227             dx12             = _mm_sub_ps(ix1,jx2);
 228             dy12             = _mm_sub_ps(iy1,jy2);
 229             dz12             = _mm_sub_ps(iz1,jz2);
 230             dx20             = _mm_sub_ps(ix2,jx0);
 231             dy20             = _mm_sub_ps(iy2,jy0);
 232             dz20             = _mm_sub_ps(iz2,jz0);
 233             dx21             = _mm_sub_ps(ix2,jx1);
 234             dy21             = _mm_sub_ps(iy2,jy1);
 235             dz21             = _mm_sub_ps(iz2,jz1);
 236             dx22             = _mm_sub_ps(ix2,jx2);
 237             dy22             = _mm_sub_ps(iy2,jy2);
 238             dz22             = _mm_sub_ps(iz2,jz2);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 242             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 243             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 244             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 245             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 246             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 247             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 248             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 249             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 250
 251             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 252             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 253             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 254             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 255             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 256             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 257             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 258             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 259             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 260
 261             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 262             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 263             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 264             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 265             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 266             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 267             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 268             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 269             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 270
 271             fjx0             = _mm_setzero_ps();
 272             fjy0             = _mm_setzero_ps();
 273             fjz0             = _mm_setzero_ps();
 274             fjx1             = _mm_setzero_ps();
 275             fjy1             = _mm_setzero_ps();
 276             fjz1             = _mm_setzero_ps();
 277             fjx2             = _mm_setzero_ps();
 278             fjy2             = _mm_setzero_ps();
 279             fjz2             = _mm_setzero_ps();
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             /* REACTION-FIELD ELECTROSTATICS */
 286             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 287             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             velecsum         = _mm_add_ps(velecsum,velec);
 291
 292             fscal            = felec;
 293
 294              /* Update vectorial force */
 295             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 296             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 297             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 298
 299             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 300             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 301             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             /* REACTION-FIELD ELECTROSTATICS */
 308             velec            = _mm_mul_ps(qq01,_mm_sub_ps(_mm_macc_ps(krf,rsq01,rinv01),crf));
 309             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velecsum         = _mm_add_ps(velecsum,velec);
 313
 314             fscal            = felec;
 315
 316              /* Update vectorial force */
 317             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 318             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 319             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 320
 321             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 322             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 323             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             /* REACTION-FIELD ELECTROSTATICS */
 330             velec            = _mm_mul_ps(qq02,_mm_sub_ps(_mm_macc_ps(krf,rsq02,rinv02),crf));
 331             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velecsum         = _mm_add_ps(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338              /* Update vectorial force */
 339             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 340             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 341             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 342
 343             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 344             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 345             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             /* REACTION-FIELD ELECTROSTATICS */
 352             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 353             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm_add_ps(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360              /* Update vectorial force */
 361             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 362             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 363             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 364
 365             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 366             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 367             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             /* REACTION-FIELD ELECTROSTATICS */
 374             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 375             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm_add_ps(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382              /* Update vectorial force */
 383             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 384             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 385             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 386
 387             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 388             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 389             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 390
 391             /**************************
 392              * CALCULATE INTERACTIONS *
 393              **************************/
 394
 395             /* REACTION-FIELD ELECTROSTATICS */
 396             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 397             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 398
 399             /* Update potential sum for this i atom from the interaction with this j atom. */
 400             velecsum         = _mm_add_ps(velecsum,velec);
 401
 402             fscal            = felec;
 403
 404              /* Update vectorial force */
 405             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 406             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 407             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 408
 409             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 410             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 411             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 412
 413             /**************************
 414              * CALCULATE INTERACTIONS *
 415              **************************/
 416
 417             /* REACTION-FIELD ELECTROSTATICS */
 418             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 419             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 420
 421             /* Update potential sum for this i atom from the interaction with this j atom. */
 422             velecsum         = _mm_add_ps(velecsum,velec);
 423
 424             fscal            = felec;
 425
 426              /* Update vectorial force */
 427             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 428             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 429             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 430
 431             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 432             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 433             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 434
 435             /**************************
 436              * CALCULATE INTERACTIONS *
 437              **************************/
 438
 439             /* REACTION-FIELD ELECTROSTATICS */
 440             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 441             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 442
 443             /* Update potential sum for this i atom from the interaction with this j atom. */
 444             velecsum         = _mm_add_ps(velecsum,velec);
 445
 446             fscal            = felec;
 447
 448              /* Update vectorial force */
 449             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 450             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 451             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 452
 453             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 454             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 455             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             /* REACTION-FIELD ELECTROSTATICS */
 462             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 463             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 464
 465             /* Update potential sum for this i atom from the interaction with this j atom. */
 466             velecsum         = _mm_add_ps(velecsum,velec);
 467
 468             fscal            = felec;
 469
 470              /* Update vectorial force */
 471             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 472             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 473             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 474
 475             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 476             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 477             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 478
 479             fjptrA             = f+j_coord_offsetA;
 480             fjptrB             = f+j_coord_offsetB;
 481             fjptrC             = f+j_coord_offsetC;
 482             fjptrD             = f+j_coord_offsetD;
 483
 484             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 485                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 486
 487             /* Inner loop uses 315 flops */
 488         }
 489
 490         if(jidx<j_index_end)
 491         {
 492
 493             /* Get j neighbor index, and coordinate index */
 494             jnrlistA         = jjnr[jidx];
 495             jnrlistB         = jjnr[jidx+1];
 496             jnrlistC         = jjnr[jidx+2];
 497             jnrlistD         = jjnr[jidx+3];
 498             /* Sign of each element will be negative for non-real atoms.
 499              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 500              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 501              */
 502             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 503             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 504             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 505             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 506             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 507             j_coord_offsetA  = DIM*jnrA;
 508             j_coord_offsetB  = DIM*jnrB;
 509             j_coord_offsetC  = DIM*jnrC;
 510             j_coord_offsetD  = DIM*jnrD;
 511
 512             /* load j atom coordinates */
 513             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 514                                               x+j_coord_offsetC,x+j_coord_offsetD,
 515                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 516
 517             /* Calculate displacement vector */
 518             dx00             = _mm_sub_ps(ix0,jx0);
 519             dy00             = _mm_sub_ps(iy0,jy0);
 520             dz00             = _mm_sub_ps(iz0,jz0);
 521             dx01             = _mm_sub_ps(ix0,jx1);
 522             dy01             = _mm_sub_ps(iy0,jy1);
 523             dz01             = _mm_sub_ps(iz0,jz1);
 524             dx02             = _mm_sub_ps(ix0,jx2);
 525             dy02             = _mm_sub_ps(iy0,jy2);
 526             dz02             = _mm_sub_ps(iz0,jz2);
 527             dx10             = _mm_sub_ps(ix1,jx0);
 528             dy10             = _mm_sub_ps(iy1,jy0);
 529             dz10             = _mm_sub_ps(iz1,jz0);
 530             dx11             = _mm_sub_ps(ix1,jx1);
 531             dy11             = _mm_sub_ps(iy1,jy1);
 532             dz11             = _mm_sub_ps(iz1,jz1);
 533             dx12             = _mm_sub_ps(ix1,jx2);
 534             dy12             = _mm_sub_ps(iy1,jy2);
 535             dz12             = _mm_sub_ps(iz1,jz2);
 536             dx20             = _mm_sub_ps(ix2,jx0);
 537             dy20             = _mm_sub_ps(iy2,jy0);
 538             dz20             = _mm_sub_ps(iz2,jz0);
 539             dx21             = _mm_sub_ps(ix2,jx1);
 540             dy21             = _mm_sub_ps(iy2,jy1);
 541             dz21             = _mm_sub_ps(iz2,jz1);
 542             dx22             = _mm_sub_ps(ix2,jx2);
 543             dy22             = _mm_sub_ps(iy2,jy2);
 544             dz22             = _mm_sub_ps(iz2,jz2);
 545
 546             /* Calculate squared distance and things based on it */
 547             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 548             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 549             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 550             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 551             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 552             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 553             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 554             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 555             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 556
 557             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 558             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 559             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 560             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 561             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 562             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 563             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 564             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 565             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 566
 567             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 568             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 569             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 570             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 571             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 572             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 573             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 574             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 575             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 576
 577             fjx0             = _mm_setzero_ps();
 578             fjy0             = _mm_setzero_ps();
 579             fjz0             = _mm_setzero_ps();
 580             fjx1             = _mm_setzero_ps();
 581             fjy1             = _mm_setzero_ps();
 582             fjz1             = _mm_setzero_ps();
 583             fjx2             = _mm_setzero_ps();
 584             fjy2             = _mm_setzero_ps();
 585             fjz2             = _mm_setzero_ps();
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             /* REACTION-FIELD ELECTROSTATICS */
 592             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 593             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 594
 595             /* Update potential sum for this i atom from the interaction with this j atom. */
 596             velec            = _mm_andnot_ps(dummy_mask,velec);
 597             velecsum         = _mm_add_ps(velecsum,velec);
 598
 599             fscal            = felec;
 600
 601             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 602
 603              /* Update vectorial force */
 604             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 605             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 606             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 607
 608             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 609             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 610             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             /* REACTION-FIELD ELECTROSTATICS */
 617             velec            = _mm_mul_ps(qq01,_mm_sub_ps(_mm_macc_ps(krf,rsq01,rinv01),crf));
 618             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
 619
 620             /* Update potential sum for this i atom from the interaction with this j atom. */
 621             velec            = _mm_andnot_ps(dummy_mask,velec);
 622             velecsum         = _mm_add_ps(velecsum,velec);
 623
 624             fscal            = felec;
 625
 626             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 627
 628              /* Update vectorial force */
 629             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 630             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 631             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 632
 633             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 634             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 635             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 636
 637             /**************************
 638              * CALCULATE INTERACTIONS *
 639              **************************/
 640
 641             /* REACTION-FIELD ELECTROSTATICS */
 642             velec            = _mm_mul_ps(qq02,_mm_sub_ps(_mm_macc_ps(krf,rsq02,rinv02),crf));
 643             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
 644
 645             /* Update potential sum for this i atom from the interaction with this j atom. */
 646             velec            = _mm_andnot_ps(dummy_mask,velec);
 647             velecsum         = _mm_add_ps(velecsum,velec);
 648
 649             fscal            = felec;
 650
 651             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 652
 653              /* Update vectorial force */
 654             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 655             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 656             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 657
 658             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 659             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 660             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 661
 662             /**************************
 663              * CALCULATE INTERACTIONS *
 664              **************************/
 665
 666             /* REACTION-FIELD ELECTROSTATICS */
 667             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 668             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 669
 670             /* Update potential sum for this i atom from the interaction with this j atom. */
 671             velec            = _mm_andnot_ps(dummy_mask,velec);
 672             velecsum         = _mm_add_ps(velecsum,velec);
 673
 674             fscal            = felec;
 675
 676             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 677
 678              /* Update vectorial force */
 679             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 680             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 681             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 682
 683             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 684             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 685             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 686
 687             /**************************
 688              * CALCULATE INTERACTIONS *
 689              **************************/
 690
 691             /* REACTION-FIELD ELECTROSTATICS */
 692             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 693             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 694
 695             /* Update potential sum for this i atom from the interaction with this j atom. */
 696             velec            = _mm_andnot_ps(dummy_mask,velec);
 697             velecsum         = _mm_add_ps(velecsum,velec);
 698
 699             fscal            = felec;
 700
 701             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 702
 703              /* Update vectorial force */
 704             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 705             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 706             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 707
 708             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 709             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 710             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 711
 712             /**************************
 713              * CALCULATE INTERACTIONS *
 714              **************************/
 715
 716             /* REACTION-FIELD ELECTROSTATICS */
 717             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 718             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 719
 720             /* Update potential sum for this i atom from the interaction with this j atom. */
 721             velec            = _mm_andnot_ps(dummy_mask,velec);
 722             velecsum         = _mm_add_ps(velecsum,velec);
 723
 724             fscal            = felec;
 725
 726             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 727
 728              /* Update vectorial force */
 729             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 730             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 731             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 732
 733             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 734             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 735             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 736
 737             /**************************
 738              * CALCULATE INTERACTIONS *
 739              **************************/
 740
 741             /* REACTION-FIELD ELECTROSTATICS */
 742             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 743             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 744
 745             /* Update potential sum for this i atom from the interaction with this j atom. */
 746             velec            = _mm_andnot_ps(dummy_mask,velec);
 747             velecsum         = _mm_add_ps(velecsum,velec);
 748
 749             fscal            = felec;
 750
 751             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 752
 753              /* Update vectorial force */
 754             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 755             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 756             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 757
 758             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 759             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 760             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 761
 762             /**************************
 763              * CALCULATE INTERACTIONS *
 764              **************************/
 765
 766             /* REACTION-FIELD ELECTROSTATICS */
 767             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 768             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 769
 770             /* Update potential sum for this i atom from the interaction with this j atom. */
 771             velec            = _mm_andnot_ps(dummy_mask,velec);
 772             velecsum         = _mm_add_ps(velecsum,velec);
 773
 774             fscal            = felec;
 775
 776             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 777
 778              /* Update vectorial force */
 779             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 780             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 781             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 782
 783             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 784             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 785             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 786
 787             /**************************
 788              * CALCULATE INTERACTIONS *
 789              **************************/
 790
 791             /* REACTION-FIELD ELECTROSTATICS */
 792             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 793             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 794
 795             /* Update potential sum for this i atom from the interaction with this j atom. */
 796             velec            = _mm_andnot_ps(dummy_mask,velec);
 797             velecsum         = _mm_add_ps(velecsum,velec);
 798
 799             fscal            = felec;
 800
 801             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 802
 803              /* Update vectorial force */
 804             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 805             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 806             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 807
 808             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 809             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 810             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 811
 812             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 813             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 814             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 815             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 816
 817             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 818                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 819
 820             /* Inner loop uses 315 flops */
 821         }
 822
 823         /* End of innermost loop */
 824
 825         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 826                                               f+i_coord_offset,fshift+i_shift_offset);
 827
 828         ggid                        = gid[iidx];
 829         /* Update potential energies */
 830         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 831
 832         /* Increment number of inner iterations */
 833         inneriter                  += j_index_end - j_index_start;
 834
 835         /* Outer loop uses 19 flops */
 836     }
 837
 838     /* Increment number of outer iterations */
 839     outeriter        += nri;
 840
 841     /* Update outer/inner flops */
 842
 843     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*315);
 844 }
 845 /*
 846  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3W3_F_avx_128_fma_single
 847  * Electrostatics interaction: ReactionField
 848  * VdW interaction:            None
 849  * Geometry:                   Water3-Water3
 850  * Calculate force/pot:        Force
 851  */
 852 void
 853 nb_kernel_ElecRF_VdwNone_GeomW3W3_F_avx_128_fma_single
 854                     (t_nblist                    * gmx_restrict       nlist,
 855                      rvec                        * gmx_restrict          xx,
 856                      rvec                        * gmx_restrict          ff,
 857                      t_forcerec                  * gmx_restrict          fr,
 858                      t_mdatoms                   * gmx_restrict     mdatoms,
 859                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 860                      t_nrnb                      * gmx_restrict        nrnb)
 861 {
 862     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 863      * just 0 for non-waters.
 864      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 865      * jnr indices corresponding to data put in the four positions in the SIMD register.
 866      */
 867     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 868     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 869     int              jnrA,jnrB,jnrC,jnrD;
 870     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 871     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 872     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 873     real             rcutoff_scalar;
 874     real             *shiftvec,*fshift,*x,*f;
 875     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 876     real             scratch[4*DIM];
 877     __m128           fscal,rcutoff,rcutoff2,jidxall;
 878     int              vdwioffset0;
 879     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 880     int              vdwioffset1;
 881     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 882     int              vdwioffset2;
 883     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 884     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 885     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 886     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 887     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 888     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 889     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 890     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 891     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 892     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 893     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 894     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 895     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 896     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 897     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 898     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 899     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 900     real             *charge;
 901     __m128           dummy_mask,cutoff_mask;
 902     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 903     __m128           one     = _mm_set1_ps(1.0);
 904     __m128           two     = _mm_set1_ps(2.0);
 905     x                = xx[0];
 906     f                = ff[0];
 907
 908     nri              = nlist->nri;
 909     iinr             = nlist->iinr;
 910     jindex           = nlist->jindex;
 911     jjnr             = nlist->jjnr;
 912     shiftidx         = nlist->shift;
 913     gid              = nlist->gid;
 914     shiftvec         = fr->shift_vec[0];
 915     fshift           = fr->fshift[0];
 916     facel            = _mm_set1_ps(fr->epsfac);
 917     charge           = mdatoms->chargeA;
 918     krf              = _mm_set1_ps(fr->ic->k_rf);
 919     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 920     crf              = _mm_set1_ps(fr->ic->c_rf);
 921
 922     /* Setup water-specific parameters */
 923     inr              = nlist->iinr[0];
 924     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 925     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 926     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 927
 928     jq0              = _mm_set1_ps(charge[inr+0]);
 929     jq1              = _mm_set1_ps(charge[inr+1]);
 930     jq2              = _mm_set1_ps(charge[inr+2]);
 931     qq00             = _mm_mul_ps(iq0,jq0);
 932     qq01             = _mm_mul_ps(iq0,jq1);
 933     qq02             = _mm_mul_ps(iq0,jq2);
 934     qq10             = _mm_mul_ps(iq1,jq0);
 935     qq11             = _mm_mul_ps(iq1,jq1);
 936     qq12             = _mm_mul_ps(iq1,jq2);
 937     qq20             = _mm_mul_ps(iq2,jq0);
 938     qq21             = _mm_mul_ps(iq2,jq1);
 939     qq22             = _mm_mul_ps(iq2,jq2);
 940
 941     /* Avoid stupid compiler warnings */
 942     jnrA = jnrB = jnrC = jnrD = 0;
 943     j_coord_offsetA = 0;
 944     j_coord_offsetB = 0;
 945     j_coord_offsetC = 0;
 946     j_coord_offsetD = 0;
 947
 948     outeriter        = 0;
 949     inneriter        = 0;
 950
 951     for(iidx=0;iidx<4*DIM;iidx++)
 952     {
 953         scratch[iidx] = 0.0;
 954     }
 955
 956     /* Start outer loop over neighborlists */
 957     for(iidx=0; iidx<nri; iidx++)
 958     {
 959         /* Load shift vector for this list */
 960         i_shift_offset   = DIM*shiftidx[iidx];
 961
 962         /* Load limits for loop over neighbors */
 963         j_index_start    = jindex[iidx];
 964         j_index_end      = jindex[iidx+1];
 965
 966         /* Get outer coordinate index */
 967         inr              = iinr[iidx];
 968         i_coord_offset   = DIM*inr;
 969
 970         /* Load i particle coords and add shift vector */
 971         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 972                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 973
 974         fix0             = _mm_setzero_ps();
 975         fiy0             = _mm_setzero_ps();
 976         fiz0             = _mm_setzero_ps();
 977         fix1             = _mm_setzero_ps();
 978         fiy1             = _mm_setzero_ps();
 979         fiz1             = _mm_setzero_ps();
 980         fix2             = _mm_setzero_ps();
 981         fiy2             = _mm_setzero_ps();
 982         fiz2             = _mm_setzero_ps();
 983
 984         /* Start inner kernel loop */
 985         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 986         {
 987
 988             /* Get j neighbor index, and coordinate index */
 989             jnrA             = jjnr[jidx];
 990             jnrB             = jjnr[jidx+1];
 991             jnrC             = jjnr[jidx+2];
 992             jnrD             = jjnr[jidx+3];
 993             j_coord_offsetA  = DIM*jnrA;
 994             j_coord_offsetB  = DIM*jnrB;
 995             j_coord_offsetC  = DIM*jnrC;
 996             j_coord_offsetD  = DIM*jnrD;
 997
 998             /* load j atom coordinates */
 999             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1000                                               x+j_coord_offsetC,x+j_coord_offsetD,
1001                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1002
1003             /* Calculate displacement vector */
1004             dx00             = _mm_sub_ps(ix0,jx0);
1005             dy00             = _mm_sub_ps(iy0,jy0);
1006             dz00             = _mm_sub_ps(iz0,jz0);
1007             dx01             = _mm_sub_ps(ix0,jx1);
1008             dy01             = _mm_sub_ps(iy0,jy1);
1009             dz01             = _mm_sub_ps(iz0,jz1);
1010             dx02             = _mm_sub_ps(ix0,jx2);
1011             dy02             = _mm_sub_ps(iy0,jy2);
1012             dz02             = _mm_sub_ps(iz0,jz2);
1013             dx10             = _mm_sub_ps(ix1,jx0);
1014             dy10             = _mm_sub_ps(iy1,jy0);
1015             dz10             = _mm_sub_ps(iz1,jz0);
1016             dx11             = _mm_sub_ps(ix1,jx1);
1017             dy11             = _mm_sub_ps(iy1,jy1);
1018             dz11             = _mm_sub_ps(iz1,jz1);
1019             dx12             = _mm_sub_ps(ix1,jx2);
1020             dy12             = _mm_sub_ps(iy1,jy2);
1021             dz12             = _mm_sub_ps(iz1,jz2);
1022             dx20             = _mm_sub_ps(ix2,jx0);
1023             dy20             = _mm_sub_ps(iy2,jy0);
1024             dz20             = _mm_sub_ps(iz2,jz0);
1025             dx21             = _mm_sub_ps(ix2,jx1);
1026             dy21             = _mm_sub_ps(iy2,jy1);
1027             dz21             = _mm_sub_ps(iz2,jz1);
1028             dx22             = _mm_sub_ps(ix2,jx2);
1029             dy22             = _mm_sub_ps(iy2,jy2);
1030             dz22             = _mm_sub_ps(iz2,jz2);
1031
1032             /* Calculate squared distance and things based on it */
1033             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1034             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1035             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1036             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1037             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1038             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1039             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1040             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1041             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1042
1043             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1044             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1045             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1046             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1047             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1048             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1049             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1050             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1051             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1052
1053             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1054             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1055             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1056             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1057             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1058             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1059             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1060             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1061             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1062
1063             fjx0             = _mm_setzero_ps();
1064             fjy0             = _mm_setzero_ps();
1065             fjz0             = _mm_setzero_ps();
1066             fjx1             = _mm_setzero_ps();
1067             fjy1             = _mm_setzero_ps();
1068             fjz1             = _mm_setzero_ps();
1069             fjx2             = _mm_setzero_ps();
1070             fjy2             = _mm_setzero_ps();
1071             fjz2             = _mm_setzero_ps();
1072
1073             /**************************
1074              * CALCULATE INTERACTIONS *
1075              **************************/
1076
1077             /* REACTION-FIELD ELECTROSTATICS */
1078             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1079
1080             fscal            = felec;
1081
1082              /* Update vectorial force */
1083             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1084             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1085             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1086
1087             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1088             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1089             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1090
1091             /**************************
1092              * CALCULATE INTERACTIONS *
1093              **************************/
1094
1095             /* REACTION-FIELD ELECTROSTATICS */
1096             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
1097
1098             fscal            = felec;
1099
1100              /* Update vectorial force */
1101             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1102             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1103             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1104
1105             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1106             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1107             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             /* REACTION-FIELD ELECTROSTATICS */
1114             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
1115
1116             fscal            = felec;
1117
1118              /* Update vectorial force */
1119             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1120             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1121             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1122
1123             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1124             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1125             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1126
1127             /**************************
1128              * CALCULATE INTERACTIONS *
1129              **************************/
1130
1131             /* REACTION-FIELD ELECTROSTATICS */
1132             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1133
1134             fscal            = felec;
1135
1136              /* Update vectorial force */
1137             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1138             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1139             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1140
1141             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1142             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1143             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1144
1145             /**************************
1146              * CALCULATE INTERACTIONS *
1147              **************************/
1148
1149             /* REACTION-FIELD ELECTROSTATICS */
1150             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1151
1152             fscal            = felec;
1153
1154              /* Update vectorial force */
1155             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1156             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1157             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1158
1159             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1160             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1161             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1162
1163             /**************************
1164              * CALCULATE INTERACTIONS *
1165              **************************/
1166
1167             /* REACTION-FIELD ELECTROSTATICS */
1168             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1169
1170             fscal            = felec;
1171
1172              /* Update vectorial force */
1173             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1174             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1175             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1176
1177             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1178             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1179             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1180
1181             /**************************
1182              * CALCULATE INTERACTIONS *
1183              **************************/
1184
1185             /* REACTION-FIELD ELECTROSTATICS */
1186             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1187
1188             fscal            = felec;
1189
1190              /* Update vectorial force */
1191             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1192             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1193             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1194
1195             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1196             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1197             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             /* REACTION-FIELD ELECTROSTATICS */
1204             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1205
1206             fscal            = felec;
1207
1208              /* Update vectorial force */
1209             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1210             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1211             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1212
1213             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1214             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1215             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1216
1217             /**************************
1218              * CALCULATE INTERACTIONS *
1219              **************************/
1220
1221             /* REACTION-FIELD ELECTROSTATICS */
1222             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1223
1224             fscal            = felec;
1225
1226              /* Update vectorial force */
1227             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1228             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1229             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1230
1231             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1232             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1233             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1234
1235             fjptrA             = f+j_coord_offsetA;
1236             fjptrB             = f+j_coord_offsetB;
1237             fjptrC             = f+j_coord_offsetC;
1238             fjptrD             = f+j_coord_offsetD;
1239
1240             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1241                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1242
1243             /* Inner loop uses 270 flops */
1244         }
1245
1246         if(jidx<j_index_end)
1247         {
1248
1249             /* Get j neighbor index, and coordinate index */
1250             jnrlistA         = jjnr[jidx];
1251             jnrlistB         = jjnr[jidx+1];
1252             jnrlistC         = jjnr[jidx+2];
1253             jnrlistD         = jjnr[jidx+3];
1254             /* Sign of each element will be negative for non-real atoms.
1255              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1256              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1257              */
1258             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1259             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1260             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1261             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1262             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1263             j_coord_offsetA  = DIM*jnrA;
1264             j_coord_offsetB  = DIM*jnrB;
1265             j_coord_offsetC  = DIM*jnrC;
1266             j_coord_offsetD  = DIM*jnrD;
1267
1268             /* load j atom coordinates */
1269             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1270                                               x+j_coord_offsetC,x+j_coord_offsetD,
1271                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1272
1273             /* Calculate displacement vector */
1274             dx00             = _mm_sub_ps(ix0,jx0);
1275             dy00             = _mm_sub_ps(iy0,jy0);
1276             dz00             = _mm_sub_ps(iz0,jz0);
1277             dx01             = _mm_sub_ps(ix0,jx1);
1278             dy01             = _mm_sub_ps(iy0,jy1);
1279             dz01             = _mm_sub_ps(iz0,jz1);
1280             dx02             = _mm_sub_ps(ix0,jx2);
1281             dy02             = _mm_sub_ps(iy0,jy2);
1282             dz02             = _mm_sub_ps(iz0,jz2);
1283             dx10             = _mm_sub_ps(ix1,jx0);
1284             dy10             = _mm_sub_ps(iy1,jy0);
1285             dz10             = _mm_sub_ps(iz1,jz0);
1286             dx11             = _mm_sub_ps(ix1,jx1);
1287             dy11             = _mm_sub_ps(iy1,jy1);
1288             dz11             = _mm_sub_ps(iz1,jz1);
1289             dx12             = _mm_sub_ps(ix1,jx2);
1290             dy12             = _mm_sub_ps(iy1,jy2);
1291             dz12             = _mm_sub_ps(iz1,jz2);
1292             dx20             = _mm_sub_ps(ix2,jx0);
1293             dy20             = _mm_sub_ps(iy2,jy0);
1294             dz20             = _mm_sub_ps(iz2,jz0);
1295             dx21             = _mm_sub_ps(ix2,jx1);
1296             dy21             = _mm_sub_ps(iy2,jy1);
1297             dz21             = _mm_sub_ps(iz2,jz1);
1298             dx22             = _mm_sub_ps(ix2,jx2);
1299             dy22             = _mm_sub_ps(iy2,jy2);
1300             dz22             = _mm_sub_ps(iz2,jz2);
1301
1302             /* Calculate squared distance and things based on it */
1303             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1304             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1305             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1306             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1307             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1308             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1309             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1310             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1311             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1312
1313             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1314             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1315             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1316             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1317             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1318             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1319             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1320             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1321             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1322
1323             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1324             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1325             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1326             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1327             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1328             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1329             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1330             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1331             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1332
1333             fjx0             = _mm_setzero_ps();
1334             fjy0             = _mm_setzero_ps();
1335             fjz0             = _mm_setzero_ps();
1336             fjx1             = _mm_setzero_ps();
1337             fjy1             = _mm_setzero_ps();
1338             fjz1             = _mm_setzero_ps();
1339             fjx2             = _mm_setzero_ps();
1340             fjy2             = _mm_setzero_ps();
1341             fjz2             = _mm_setzero_ps();
1342
1343             /**************************
1344              * CALCULATE INTERACTIONS *
1345              **************************/
1346
1347             /* REACTION-FIELD ELECTROSTATICS */
1348             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1349
1350             fscal            = felec;
1351
1352             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1353
1354              /* Update vectorial force */
1355             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1356             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1357             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1358
1359             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1360             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1361             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1362
1363             /**************************
1364              * CALCULATE INTERACTIONS *
1365              **************************/
1366
1367             /* REACTION-FIELD ELECTROSTATICS */
1368             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
1369
1370             fscal            = felec;
1371
1372             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1373
1374              /* Update vectorial force */
1375             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1376             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1377             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1378
1379             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1380             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1381             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1382
1383             /**************************
1384              * CALCULATE INTERACTIONS *
1385              **************************/
1386
1387             /* REACTION-FIELD ELECTROSTATICS */
1388             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
1389
1390             fscal            = felec;
1391
1392             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1393
1394              /* Update vectorial force */
1395             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1396             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1397             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1398
1399             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1400             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1401             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1402
1403             /**************************
1404              * CALCULATE INTERACTIONS *
1405              **************************/
1406
1407             /* REACTION-FIELD ELECTROSTATICS */
1408             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1409
1410             fscal            = felec;
1411
1412             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1413
1414              /* Update vectorial force */
1415             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1416             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1417             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1418
1419             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1420             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1421             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1422
1423             /**************************
1424              * CALCULATE INTERACTIONS *
1425              **************************/
1426
1427             /* REACTION-FIELD ELECTROSTATICS */
1428             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1429
1430             fscal            = felec;
1431
1432             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1433
1434              /* Update vectorial force */
1435             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1436             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1437             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1438
1439             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1440             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1441             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1442
1443             /**************************
1444              * CALCULATE INTERACTIONS *
1445              **************************/
1446
1447             /* REACTION-FIELD ELECTROSTATICS */
1448             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1449
1450             fscal            = felec;
1451
1452             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1453
1454              /* Update vectorial force */
1455             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1456             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1457             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1458
1459             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1460             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1461             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1462
1463             /**************************
1464              * CALCULATE INTERACTIONS *
1465              **************************/
1466
1467             /* REACTION-FIELD ELECTROSTATICS */
1468             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1469
1470             fscal            = felec;
1471
1472             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1473
1474              /* Update vectorial force */
1475             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1476             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1477             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1478
1479             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1480             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1481             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1482
1483             /**************************
1484              * CALCULATE INTERACTIONS *
1485              **************************/
1486
1487             /* REACTION-FIELD ELECTROSTATICS */
1488             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1489
1490             fscal            = felec;
1491
1492             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1493
1494              /* Update vectorial force */
1495             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1496             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1497             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1498
1499             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1500             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1501             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1502
1503             /**************************
1504              * CALCULATE INTERACTIONS *
1505              **************************/
1506
1507             /* REACTION-FIELD ELECTROSTATICS */
1508             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1509
1510             fscal            = felec;
1511
1512             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1513
1514              /* Update vectorial force */
1515             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1516             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1517             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1518
1519             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1520             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1521             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1522
1523             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1524             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1525             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1526             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1527
1528             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1529                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1530
1531             /* Inner loop uses 270 flops */
1532         }
1533
1534         /* End of innermost loop */
1535
1536         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1537                                               f+i_coord_offset,fshift+i_shift_offset);
1538
1539         /* Increment number of inner iterations */
1540         inneriter                  += j_index_end - j_index_start;
1541
1542         /* Outer loop uses 18 flops */
1543     }
1544
1545     /* Increment number of outer iterations */
1546     outeriter        += nri;
1547
1548     /* Update outer/inner flops */
1549
1550     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*270);
1551 }