src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwLJ_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3W3_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwLJ_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     int              nvdwtype;
 107     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 108     int              *vdwtype;
 109     real             *vdwparam;
 110     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 111     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 112     __m128           dummy_mask,cutoff_mask;
 113     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 114     __m128           one     = _mm_set1_ps(1.0);
 115     __m128           two     = _mm_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     krf              = _mm_set1_ps(fr->ic->k_rf);
 130     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 131     crf              = _mm_set1_ps(fr->ic->c_rf);
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 139     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 140     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 141     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 142
 143     jq0              = _mm_set1_ps(charge[inr+0]);
 144     jq1              = _mm_set1_ps(charge[inr+1]);
 145     jq2              = _mm_set1_ps(charge[inr+2]);
 146     vdwjidx0A        = 2*vdwtype[inr+0];
 147     qq00             = _mm_mul_ps(iq0,jq0);
 148     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 149     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 150     qq01             = _mm_mul_ps(iq0,jq1);
 151     qq02             = _mm_mul_ps(iq0,jq2);
 152     qq10             = _mm_mul_ps(iq1,jq0);
 153     qq11             = _mm_mul_ps(iq1,jq1);
 154     qq12             = _mm_mul_ps(iq1,jq2);
 155     qq20             = _mm_mul_ps(iq2,jq0);
 156     qq21             = _mm_mul_ps(iq2,jq1);
 157     qq22             = _mm_mul_ps(iq2,jq2);
 158
 159     /* Avoid stupid compiler warnings */
 160     jnrA = jnrB = jnrC = jnrD = 0;
 161     j_coord_offsetA = 0;
 162     j_coord_offsetB = 0;
 163     j_coord_offsetC = 0;
 164     j_coord_offsetD = 0;
 165
 166     outeriter        = 0;
 167     inneriter        = 0;
 168
 169     for(iidx=0;iidx<4*DIM;iidx++)
 170     {
 171         scratch[iidx] = 0.0;
 172     }
 173
 174     /* Start outer loop over neighborlists */
 175     for(iidx=0; iidx<nri; iidx++)
 176     {
 177         /* Load shift vector for this list */
 178         i_shift_offset   = DIM*shiftidx[iidx];
 179
 180         /* Load limits for loop over neighbors */
 181         j_index_start    = jindex[iidx];
 182         j_index_end      = jindex[iidx+1];
 183
 184         /* Get outer coordinate index */
 185         inr              = iinr[iidx];
 186         i_coord_offset   = DIM*inr;
 187
 188         /* Load i particle coords and add shift vector */
 189         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 190                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 191
 192         fix0             = _mm_setzero_ps();
 193         fiy0             = _mm_setzero_ps();
 194         fiz0             = _mm_setzero_ps();
 195         fix1             = _mm_setzero_ps();
 196         fiy1             = _mm_setzero_ps();
 197         fiz1             = _mm_setzero_ps();
 198         fix2             = _mm_setzero_ps();
 199         fiy2             = _mm_setzero_ps();
 200         fiz2             = _mm_setzero_ps();
 201
 202         /* Reset potential sums */
 203         velecsum         = _mm_setzero_ps();
 204         vvdwsum          = _mm_setzero_ps();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             j_coord_offsetA  = DIM*jnrA;
 216             j_coord_offsetB  = DIM*jnrB;
 217             j_coord_offsetC  = DIM*jnrC;
 218             j_coord_offsetD  = DIM*jnrD;
 219
 220             /* load j atom coordinates */
 221             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                               x+j_coord_offsetC,x+j_coord_offsetD,
 223                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 224
 225             /* Calculate displacement vector */
 226             dx00             = _mm_sub_ps(ix0,jx0);
 227             dy00             = _mm_sub_ps(iy0,jy0);
 228             dz00             = _mm_sub_ps(iz0,jz0);
 229             dx01             = _mm_sub_ps(ix0,jx1);
 230             dy01             = _mm_sub_ps(iy0,jy1);
 231             dz01             = _mm_sub_ps(iz0,jz1);
 232             dx02             = _mm_sub_ps(ix0,jx2);
 233             dy02             = _mm_sub_ps(iy0,jy2);
 234             dz02             = _mm_sub_ps(iz0,jz2);
 235             dx10             = _mm_sub_ps(ix1,jx0);
 236             dy10             = _mm_sub_ps(iy1,jy0);
 237             dz10             = _mm_sub_ps(iz1,jz0);
 238             dx11             = _mm_sub_ps(ix1,jx1);
 239             dy11             = _mm_sub_ps(iy1,jy1);
 240             dz11             = _mm_sub_ps(iz1,jz1);
 241             dx12             = _mm_sub_ps(ix1,jx2);
 242             dy12             = _mm_sub_ps(iy1,jy2);
 243             dz12             = _mm_sub_ps(iz1,jz2);
 244             dx20             = _mm_sub_ps(ix2,jx0);
 245             dy20             = _mm_sub_ps(iy2,jy0);
 246             dz20             = _mm_sub_ps(iz2,jz0);
 247             dx21             = _mm_sub_ps(ix2,jx1);
 248             dy21             = _mm_sub_ps(iy2,jy1);
 249             dz21             = _mm_sub_ps(iz2,jz1);
 250             dx22             = _mm_sub_ps(ix2,jx2);
 251             dy22             = _mm_sub_ps(iy2,jy2);
 252             dz22             = _mm_sub_ps(iz2,jz2);
 253
 254             /* Calculate squared distance and things based on it */
 255             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 256             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 257             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 258             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 259             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 260             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 261             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 262             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 263             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 264
 265             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 266             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 267             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 268             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 269             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 270             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 271             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 272             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 273             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 274
 275             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 276             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 277             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 278             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 279             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 280             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 281             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 282             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 283             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 284
 285             fjx0             = _mm_setzero_ps();
 286             fjy0             = _mm_setzero_ps();
 287             fjz0             = _mm_setzero_ps();
 288             fjx1             = _mm_setzero_ps();
 289             fjy1             = _mm_setzero_ps();
 290             fjz1             = _mm_setzero_ps();
 291             fjx2             = _mm_setzero_ps();
 292             fjy2             = _mm_setzero_ps();
 293             fjz2             = _mm_setzero_ps();
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             /* REACTION-FIELD ELECTROSTATICS */
 300             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 301             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 302
 303             /* LENNARD-JONES DISPERSION/REPULSION */
 304
 305             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 306             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 307             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 308             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 309             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velecsum         = _mm_add_ps(velecsum,velec);
 313             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 314
 315             fscal            = _mm_add_ps(felec,fvdw);
 316
 317              /* Update vectorial force */
 318             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 319             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 320             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 321
 322             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 323             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 324             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             /* REACTION-FIELD ELECTROSTATICS */
 331             velec            = _mm_mul_ps(qq01,_mm_sub_ps(_mm_macc_ps(krf,rsq01,rinv01),crf));
 332             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velecsum         = _mm_add_ps(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339              /* Update vectorial force */
 340             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 341             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 342             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 343
 344             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 345             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 346             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             /* REACTION-FIELD ELECTROSTATICS */
 353             velec            = _mm_mul_ps(qq02,_mm_sub_ps(_mm_macc_ps(krf,rsq02,rinv02),crf));
 354             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velecsum         = _mm_add_ps(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361              /* Update vectorial force */
 362             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 363             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 364             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 365
 366             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 367             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 368             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 369
 370             /**************************
 371              * CALCULATE INTERACTIONS *
 372              **************************/
 373
 374             /* REACTION-FIELD ELECTROSTATICS */
 375             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 376             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velecsum         = _mm_add_ps(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383              /* Update vectorial force */
 384             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 385             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 386             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 387
 388             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 389             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 390             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             /* REACTION-FIELD ELECTROSTATICS */
 397             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 398             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velecsum         = _mm_add_ps(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405              /* Update vectorial force */
 406             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 407             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 408             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 409
 410             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 411             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 412             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 413
 414             /**************************
 415              * CALCULATE INTERACTIONS *
 416              **************************/
 417
 418             /* REACTION-FIELD ELECTROSTATICS */
 419             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 420             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velecsum         = _mm_add_ps(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427              /* Update vectorial force */
 428             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 429             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 430             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 431
 432             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 433             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 434             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             /* REACTION-FIELD ELECTROSTATICS */
 441             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 442             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 443
 444             /* Update potential sum for this i atom from the interaction with this j atom. */
 445             velecsum         = _mm_add_ps(velecsum,velec);
 446
 447             fscal            = felec;
 448
 449              /* Update vectorial force */
 450             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 451             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 452             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 453
 454             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 455             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 456             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             /* REACTION-FIELD ELECTROSTATICS */
 463             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 464             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             velecsum         = _mm_add_ps(velecsum,velec);
 468
 469             fscal            = felec;
 470
 471              /* Update vectorial force */
 472             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 473             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 474             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 475
 476             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 477             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 478             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             /* REACTION-FIELD ELECTROSTATICS */
 485             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 486             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velecsum         = _mm_add_ps(velecsum,velec);
 490
 491             fscal            = felec;
 492
 493              /* Update vectorial force */
 494             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 495             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 496             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 497
 498             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 499             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 500             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 501
 502             fjptrA             = f+j_coord_offsetA;
 503             fjptrB             = f+j_coord_offsetB;
 504             fjptrC             = f+j_coord_offsetC;
 505             fjptrD             = f+j_coord_offsetD;
 506
 507             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 508                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 509
 510             /* Inner loop uses 327 flops */
 511         }
 512
 513         if(jidx<j_index_end)
 514         {
 515
 516             /* Get j neighbor index, and coordinate index */
 517             jnrlistA         = jjnr[jidx];
 518             jnrlistB         = jjnr[jidx+1];
 519             jnrlistC         = jjnr[jidx+2];
 520             jnrlistD         = jjnr[jidx+3];
 521             /* Sign of each element will be negative for non-real atoms.
 522              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 523              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 524              */
 525             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 526             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 527             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 528             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 529             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 530             j_coord_offsetA  = DIM*jnrA;
 531             j_coord_offsetB  = DIM*jnrB;
 532             j_coord_offsetC  = DIM*jnrC;
 533             j_coord_offsetD  = DIM*jnrD;
 534
 535             /* load j atom coordinates */
 536             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 537                                               x+j_coord_offsetC,x+j_coord_offsetD,
 538                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 539
 540             /* Calculate displacement vector */
 541             dx00             = _mm_sub_ps(ix0,jx0);
 542             dy00             = _mm_sub_ps(iy0,jy0);
 543             dz00             = _mm_sub_ps(iz0,jz0);
 544             dx01             = _mm_sub_ps(ix0,jx1);
 545             dy01             = _mm_sub_ps(iy0,jy1);
 546             dz01             = _mm_sub_ps(iz0,jz1);
 547             dx02             = _mm_sub_ps(ix0,jx2);
 548             dy02             = _mm_sub_ps(iy0,jy2);
 549             dz02             = _mm_sub_ps(iz0,jz2);
 550             dx10             = _mm_sub_ps(ix1,jx0);
 551             dy10             = _mm_sub_ps(iy1,jy0);
 552             dz10             = _mm_sub_ps(iz1,jz0);
 553             dx11             = _mm_sub_ps(ix1,jx1);
 554             dy11             = _mm_sub_ps(iy1,jy1);
 555             dz11             = _mm_sub_ps(iz1,jz1);
 556             dx12             = _mm_sub_ps(ix1,jx2);
 557             dy12             = _mm_sub_ps(iy1,jy2);
 558             dz12             = _mm_sub_ps(iz1,jz2);
 559             dx20             = _mm_sub_ps(ix2,jx0);
 560             dy20             = _mm_sub_ps(iy2,jy0);
 561             dz20             = _mm_sub_ps(iz2,jz0);
 562             dx21             = _mm_sub_ps(ix2,jx1);
 563             dy21             = _mm_sub_ps(iy2,jy1);
 564             dz21             = _mm_sub_ps(iz2,jz1);
 565             dx22             = _mm_sub_ps(ix2,jx2);
 566             dy22             = _mm_sub_ps(iy2,jy2);
 567             dz22             = _mm_sub_ps(iz2,jz2);
 568
 569             /* Calculate squared distance and things based on it */
 570             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 571             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 572             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 573             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 574             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 575             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 576             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 577             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 578             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 579
 580             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 581             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 582             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 583             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 584             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 585             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 586             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 587             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 588             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 589
 590             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 591             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 592             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 593             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 594             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 595             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 596             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 597             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 598             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 599
 600             fjx0             = _mm_setzero_ps();
 601             fjy0             = _mm_setzero_ps();
 602             fjz0             = _mm_setzero_ps();
 603             fjx1             = _mm_setzero_ps();
 604             fjy1             = _mm_setzero_ps();
 605             fjz1             = _mm_setzero_ps();
 606             fjx2             = _mm_setzero_ps();
 607             fjy2             = _mm_setzero_ps();
 608             fjz2             = _mm_setzero_ps();
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             /* REACTION-FIELD ELECTROSTATICS */
 615             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 616             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 617
 618             /* LENNARD-JONES DISPERSION/REPULSION */
 619
 620             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 621             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 622             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 623             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 624             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm_andnot_ps(dummy_mask,velec);
 628             velecsum         = _mm_add_ps(velecsum,velec);
 629             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 630             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 631
 632             fscal            = _mm_add_ps(felec,fvdw);
 633
 634             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 635
 636              /* Update vectorial force */
 637             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 638             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 639             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 640
 641             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 642             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 643             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 644
 645             /**************************
 646              * CALCULATE INTERACTIONS *
 647              **************************/
 648
 649             /* REACTION-FIELD ELECTROSTATICS */
 650             velec            = _mm_mul_ps(qq01,_mm_sub_ps(_mm_macc_ps(krf,rsq01,rinv01),crf));
 651             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
 652
 653             /* Update potential sum for this i atom from the interaction with this j atom. */
 654             velec            = _mm_andnot_ps(dummy_mask,velec);
 655             velecsum         = _mm_add_ps(velecsum,velec);
 656
 657             fscal            = felec;
 658
 659             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 660
 661              /* Update vectorial force */
 662             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 663             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 664             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 665
 666             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 667             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 668             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             /* REACTION-FIELD ELECTROSTATICS */
 675             velec            = _mm_mul_ps(qq02,_mm_sub_ps(_mm_macc_ps(krf,rsq02,rinv02),crf));
 676             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
 677
 678             /* Update potential sum for this i atom from the interaction with this j atom. */
 679             velec            = _mm_andnot_ps(dummy_mask,velec);
 680             velecsum         = _mm_add_ps(velecsum,velec);
 681
 682             fscal            = felec;
 683
 684             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 685
 686              /* Update vectorial force */
 687             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 688             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 689             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 690
 691             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 692             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 693             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 694
 695             /**************************
 696              * CALCULATE INTERACTIONS *
 697              **************************/
 698
 699             /* REACTION-FIELD ELECTROSTATICS */
 700             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 701             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 702
 703             /* Update potential sum for this i atom from the interaction with this j atom. */
 704             velec            = _mm_andnot_ps(dummy_mask,velec);
 705             velecsum         = _mm_add_ps(velecsum,velec);
 706
 707             fscal            = felec;
 708
 709             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 710
 711              /* Update vectorial force */
 712             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 713             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 714             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 715
 716             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 717             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 718             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             /* REACTION-FIELD ELECTROSTATICS */
 725             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 726             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 727
 728             /* Update potential sum for this i atom from the interaction with this j atom. */
 729             velec            = _mm_andnot_ps(dummy_mask,velec);
 730             velecsum         = _mm_add_ps(velecsum,velec);
 731
 732             fscal            = felec;
 733
 734             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 735
 736              /* Update vectorial force */
 737             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 738             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 739             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 740
 741             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 742             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 743             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 744
 745             /**************************
 746              * CALCULATE INTERACTIONS *
 747              **************************/
 748
 749             /* REACTION-FIELD ELECTROSTATICS */
 750             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 751             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 752
 753             /* Update potential sum for this i atom from the interaction with this j atom. */
 754             velec            = _mm_andnot_ps(dummy_mask,velec);
 755             velecsum         = _mm_add_ps(velecsum,velec);
 756
 757             fscal            = felec;
 758
 759             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 760
 761              /* Update vectorial force */
 762             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 763             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 764             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 765
 766             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 767             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 768             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 769
 770             /**************************
 771              * CALCULATE INTERACTIONS *
 772              **************************/
 773
 774             /* REACTION-FIELD ELECTROSTATICS */
 775             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 776             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 777
 778             /* Update potential sum for this i atom from the interaction with this j atom. */
 779             velec            = _mm_andnot_ps(dummy_mask,velec);
 780             velecsum         = _mm_add_ps(velecsum,velec);
 781
 782             fscal            = felec;
 783
 784             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 785
 786              /* Update vectorial force */
 787             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 788             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 789             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 790
 791             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 792             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 793             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 794
 795             /**************************
 796              * CALCULATE INTERACTIONS *
 797              **************************/
 798
 799             /* REACTION-FIELD ELECTROSTATICS */
 800             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 801             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 802
 803             /* Update potential sum for this i atom from the interaction with this j atom. */
 804             velec            = _mm_andnot_ps(dummy_mask,velec);
 805             velecsum         = _mm_add_ps(velecsum,velec);
 806
 807             fscal            = felec;
 808
 809             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 810
 811              /* Update vectorial force */
 812             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 813             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 814             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 815
 816             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 817             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 818             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 819
 820             /**************************
 821              * CALCULATE INTERACTIONS *
 822              **************************/
 823
 824             /* REACTION-FIELD ELECTROSTATICS */
 825             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 826             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 827
 828             /* Update potential sum for this i atom from the interaction with this j atom. */
 829             velec            = _mm_andnot_ps(dummy_mask,velec);
 830             velecsum         = _mm_add_ps(velecsum,velec);
 831
 832             fscal            = felec;
 833
 834             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 835
 836              /* Update vectorial force */
 837             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 838             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 839             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 840
 841             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 842             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 843             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 844
 845             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 846             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 847             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 848             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 849
 850             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 851                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 852
 853             /* Inner loop uses 327 flops */
 854         }
 855
 856         /* End of innermost loop */
 857
 858         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 859                                               f+i_coord_offset,fshift+i_shift_offset);
 860
 861         ggid                        = gid[iidx];
 862         /* Update potential energies */
 863         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 864         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 865
 866         /* Increment number of inner iterations */
 867         inneriter                  += j_index_end - j_index_start;
 868
 869         /* Outer loop uses 20 flops */
 870     }
 871
 872     /* Increment number of outer iterations */
 873     outeriter        += nri;
 874
 875     /* Update outer/inner flops */
 876
 877     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*327);
 878 }
 879 /*
 880  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3W3_F_avx_128_fma_single
 881  * Electrostatics interaction: ReactionField
 882  * VdW interaction:            LennardJones
 883  * Geometry:                   Water3-Water3
 884  * Calculate force/pot:        Force
 885  */
 886 void
 887 nb_kernel_ElecRF_VdwLJ_GeomW3W3_F_avx_128_fma_single
 888                     (t_nblist                    * gmx_restrict       nlist,
 889                      rvec                        * gmx_restrict          xx,
 890                      rvec                        * gmx_restrict          ff,
 891                      t_forcerec                  * gmx_restrict          fr,
 892                      t_mdatoms                   * gmx_restrict     mdatoms,
 893                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 894                      t_nrnb                      * gmx_restrict        nrnb)
 895 {
 896     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 897      * just 0 for non-waters.
 898      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 899      * jnr indices corresponding to data put in the four positions in the SIMD register.
 900      */
 901     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 902     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 903     int              jnrA,jnrB,jnrC,jnrD;
 904     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 905     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 906     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 907     real             rcutoff_scalar;
 908     real             *shiftvec,*fshift,*x,*f;
 909     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 910     real             scratch[4*DIM];
 911     __m128           fscal,rcutoff,rcutoff2,jidxall;
 912     int              vdwioffset0;
 913     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 914     int              vdwioffset1;
 915     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 916     int              vdwioffset2;
 917     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 918     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 919     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 920     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 921     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 922     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 923     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 924     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 925     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 926     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 927     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 928     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 929     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 930     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 931     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 932     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 933     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 934     real             *charge;
 935     int              nvdwtype;
 936     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 937     int              *vdwtype;
 938     real             *vdwparam;
 939     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 940     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 941     __m128           dummy_mask,cutoff_mask;
 942     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 943     __m128           one     = _mm_set1_ps(1.0);
 944     __m128           two     = _mm_set1_ps(2.0);
 945     x                = xx[0];
 946     f                = ff[0];
 947
 948     nri              = nlist->nri;
 949     iinr             = nlist->iinr;
 950     jindex           = nlist->jindex;
 951     jjnr             = nlist->jjnr;
 952     shiftidx         = nlist->shift;
 953     gid              = nlist->gid;
 954     shiftvec         = fr->shift_vec[0];
 955     fshift           = fr->fshift[0];
 956     facel            = _mm_set1_ps(fr->epsfac);
 957     charge           = mdatoms->chargeA;
 958     krf              = _mm_set1_ps(fr->ic->k_rf);
 959     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 960     crf              = _mm_set1_ps(fr->ic->c_rf);
 961     nvdwtype         = fr->ntype;
 962     vdwparam         = fr->nbfp;
 963     vdwtype          = mdatoms->typeA;
 964
 965     /* Setup water-specific parameters */
 966     inr              = nlist->iinr[0];
 967     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 968     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 969     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 970     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 971
 972     jq0              = _mm_set1_ps(charge[inr+0]);
 973     jq1              = _mm_set1_ps(charge[inr+1]);
 974     jq2              = _mm_set1_ps(charge[inr+2]);
 975     vdwjidx0A        = 2*vdwtype[inr+0];
 976     qq00             = _mm_mul_ps(iq0,jq0);
 977     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 978     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 979     qq01             = _mm_mul_ps(iq0,jq1);
 980     qq02             = _mm_mul_ps(iq0,jq2);
 981     qq10             = _mm_mul_ps(iq1,jq0);
 982     qq11             = _mm_mul_ps(iq1,jq1);
 983     qq12             = _mm_mul_ps(iq1,jq2);
 984     qq20             = _mm_mul_ps(iq2,jq0);
 985     qq21             = _mm_mul_ps(iq2,jq1);
 986     qq22             = _mm_mul_ps(iq2,jq2);
 987
 988     /* Avoid stupid compiler warnings */
 989     jnrA = jnrB = jnrC = jnrD = 0;
 990     j_coord_offsetA = 0;
 991     j_coord_offsetB = 0;
 992     j_coord_offsetC = 0;
 993     j_coord_offsetD = 0;
 994
 995     outeriter        = 0;
 996     inneriter        = 0;
 997
 998     for(iidx=0;iidx<4*DIM;iidx++)
 999     {
1000         scratch[iidx] = 0.0;
1001     }
1002
1003     /* Start outer loop over neighborlists */
1004     for(iidx=0; iidx<nri; iidx++)
1005     {
1006         /* Load shift vector for this list */
1007         i_shift_offset   = DIM*shiftidx[iidx];
1008
1009         /* Load limits for loop over neighbors */
1010         j_index_start    = jindex[iidx];
1011         j_index_end      = jindex[iidx+1];
1012
1013         /* Get outer coordinate index */
1014         inr              = iinr[iidx];
1015         i_coord_offset   = DIM*inr;
1016
1017         /* Load i particle coords and add shift vector */
1018         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1019                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1020
1021         fix0             = _mm_setzero_ps();
1022         fiy0             = _mm_setzero_ps();
1023         fiz0             = _mm_setzero_ps();
1024         fix1             = _mm_setzero_ps();
1025         fiy1             = _mm_setzero_ps();
1026         fiz1             = _mm_setzero_ps();
1027         fix2             = _mm_setzero_ps();
1028         fiy2             = _mm_setzero_ps();
1029         fiz2             = _mm_setzero_ps();
1030
1031         /* Start inner kernel loop */
1032         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1033         {
1034
1035             /* Get j neighbor index, and coordinate index */
1036             jnrA             = jjnr[jidx];
1037             jnrB             = jjnr[jidx+1];
1038             jnrC             = jjnr[jidx+2];
1039             jnrD             = jjnr[jidx+3];
1040             j_coord_offsetA  = DIM*jnrA;
1041             j_coord_offsetB  = DIM*jnrB;
1042             j_coord_offsetC  = DIM*jnrC;
1043             j_coord_offsetD  = DIM*jnrD;
1044
1045             /* load j atom coordinates */
1046             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1047                                               x+j_coord_offsetC,x+j_coord_offsetD,
1048                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1049
1050             /* Calculate displacement vector */
1051             dx00             = _mm_sub_ps(ix0,jx0);
1052             dy00             = _mm_sub_ps(iy0,jy0);
1053             dz00             = _mm_sub_ps(iz0,jz0);
1054             dx01             = _mm_sub_ps(ix0,jx1);
1055             dy01             = _mm_sub_ps(iy0,jy1);
1056             dz01             = _mm_sub_ps(iz0,jz1);
1057             dx02             = _mm_sub_ps(ix0,jx2);
1058             dy02             = _mm_sub_ps(iy0,jy2);
1059             dz02             = _mm_sub_ps(iz0,jz2);
1060             dx10             = _mm_sub_ps(ix1,jx0);
1061             dy10             = _mm_sub_ps(iy1,jy0);
1062             dz10             = _mm_sub_ps(iz1,jz0);
1063             dx11             = _mm_sub_ps(ix1,jx1);
1064             dy11             = _mm_sub_ps(iy1,jy1);
1065             dz11             = _mm_sub_ps(iz1,jz1);
1066             dx12             = _mm_sub_ps(ix1,jx2);
1067             dy12             = _mm_sub_ps(iy1,jy2);
1068             dz12             = _mm_sub_ps(iz1,jz2);
1069             dx20             = _mm_sub_ps(ix2,jx0);
1070             dy20             = _mm_sub_ps(iy2,jy0);
1071             dz20             = _mm_sub_ps(iz2,jz0);
1072             dx21             = _mm_sub_ps(ix2,jx1);
1073             dy21             = _mm_sub_ps(iy2,jy1);
1074             dz21             = _mm_sub_ps(iz2,jz1);
1075             dx22             = _mm_sub_ps(ix2,jx2);
1076             dy22             = _mm_sub_ps(iy2,jy2);
1077             dz22             = _mm_sub_ps(iz2,jz2);
1078
1079             /* Calculate squared distance and things based on it */
1080             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1081             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1082             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1083             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1084             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1085             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1086             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1087             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1088             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1089
1090             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1091             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1092             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1093             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1094             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1095             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1096             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1097             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1098             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1099
1100             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1101             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1102             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1103             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1104             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1105             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1106             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1107             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1108             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1109
1110             fjx0             = _mm_setzero_ps();
1111             fjy0             = _mm_setzero_ps();
1112             fjz0             = _mm_setzero_ps();
1113             fjx1             = _mm_setzero_ps();
1114             fjy1             = _mm_setzero_ps();
1115             fjz1             = _mm_setzero_ps();
1116             fjx2             = _mm_setzero_ps();
1117             fjy2             = _mm_setzero_ps();
1118             fjz2             = _mm_setzero_ps();
1119
1120             /**************************
1121              * CALCULATE INTERACTIONS *
1122              **************************/
1123
1124             /* REACTION-FIELD ELECTROSTATICS */
1125             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1126
1127             /* LENNARD-JONES DISPERSION/REPULSION */
1128
1129             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1130             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1131
1132             fscal            = _mm_add_ps(felec,fvdw);
1133
1134              /* Update vectorial force */
1135             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1136             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1137             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1138
1139             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1140             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1141             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1142
1143             /**************************
1144              * CALCULATE INTERACTIONS *
1145              **************************/
1146
1147             /* REACTION-FIELD ELECTROSTATICS */
1148             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
1149
1150             fscal            = felec;
1151
1152              /* Update vectorial force */
1153             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1154             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1155             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1156
1157             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1158             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1159             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1160
1161             /**************************
1162              * CALCULATE INTERACTIONS *
1163              **************************/
1164
1165             /* REACTION-FIELD ELECTROSTATICS */
1166             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
1167
1168             fscal            = felec;
1169
1170              /* Update vectorial force */
1171             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1172             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1173             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1174
1175             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1176             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1177             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1178
1179             /**************************
1180              * CALCULATE INTERACTIONS *
1181              **************************/
1182
1183             /* REACTION-FIELD ELECTROSTATICS */
1184             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1185
1186             fscal            = felec;
1187
1188              /* Update vectorial force */
1189             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1190             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1191             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1192
1193             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1194             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1195             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1196
1197             /**************************
1198              * CALCULATE INTERACTIONS *
1199              **************************/
1200
1201             /* REACTION-FIELD ELECTROSTATICS */
1202             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1203
1204             fscal            = felec;
1205
1206              /* Update vectorial force */
1207             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1208             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1209             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1210
1211             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1212             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1213             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1214
1215             /**************************
1216              * CALCULATE INTERACTIONS *
1217              **************************/
1218
1219             /* REACTION-FIELD ELECTROSTATICS */
1220             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1221
1222             fscal            = felec;
1223
1224              /* Update vectorial force */
1225             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1226             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1227             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1228
1229             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1230             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1231             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             /* REACTION-FIELD ELECTROSTATICS */
1238             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1239
1240             fscal            = felec;
1241
1242              /* Update vectorial force */
1243             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1244             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1245             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1246
1247             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1248             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1249             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1250
1251             /**************************
1252              * CALCULATE INTERACTIONS *
1253              **************************/
1254
1255             /* REACTION-FIELD ELECTROSTATICS */
1256             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1257
1258             fscal            = felec;
1259
1260              /* Update vectorial force */
1261             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1262             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1263             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1264
1265             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1266             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1267             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1268
1269             /**************************
1270              * CALCULATE INTERACTIONS *
1271              **************************/
1272
1273             /* REACTION-FIELD ELECTROSTATICS */
1274             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1275
1276             fscal            = felec;
1277
1278              /* Update vectorial force */
1279             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1280             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1281             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1282
1283             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1284             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1285             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1286
1287             fjptrA             = f+j_coord_offsetA;
1288             fjptrB             = f+j_coord_offsetB;
1289             fjptrC             = f+j_coord_offsetC;
1290             fjptrD             = f+j_coord_offsetD;
1291
1292             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1293                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1294
1295             /* Inner loop uses 277 flops */
1296         }
1297
1298         if(jidx<j_index_end)
1299         {
1300
1301             /* Get j neighbor index, and coordinate index */
1302             jnrlistA         = jjnr[jidx];
1303             jnrlistB         = jjnr[jidx+1];
1304             jnrlistC         = jjnr[jidx+2];
1305             jnrlistD         = jjnr[jidx+3];
1306             /* Sign of each element will be negative for non-real atoms.
1307              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1308              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1309              */
1310             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1311             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1312             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1313             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1314             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1315             j_coord_offsetA  = DIM*jnrA;
1316             j_coord_offsetB  = DIM*jnrB;
1317             j_coord_offsetC  = DIM*jnrC;
1318             j_coord_offsetD  = DIM*jnrD;
1319
1320             /* load j atom coordinates */
1321             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1322                                               x+j_coord_offsetC,x+j_coord_offsetD,
1323                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1324
1325             /* Calculate displacement vector */
1326             dx00             = _mm_sub_ps(ix0,jx0);
1327             dy00             = _mm_sub_ps(iy0,jy0);
1328             dz00             = _mm_sub_ps(iz0,jz0);
1329             dx01             = _mm_sub_ps(ix0,jx1);
1330             dy01             = _mm_sub_ps(iy0,jy1);
1331             dz01             = _mm_sub_ps(iz0,jz1);
1332             dx02             = _mm_sub_ps(ix0,jx2);
1333             dy02             = _mm_sub_ps(iy0,jy2);
1334             dz02             = _mm_sub_ps(iz0,jz2);
1335             dx10             = _mm_sub_ps(ix1,jx0);
1336             dy10             = _mm_sub_ps(iy1,jy0);
1337             dz10             = _mm_sub_ps(iz1,jz0);
1338             dx11             = _mm_sub_ps(ix1,jx1);
1339             dy11             = _mm_sub_ps(iy1,jy1);
1340             dz11             = _mm_sub_ps(iz1,jz1);
1341             dx12             = _mm_sub_ps(ix1,jx2);
1342             dy12             = _mm_sub_ps(iy1,jy2);
1343             dz12             = _mm_sub_ps(iz1,jz2);
1344             dx20             = _mm_sub_ps(ix2,jx0);
1345             dy20             = _mm_sub_ps(iy2,jy0);
1346             dz20             = _mm_sub_ps(iz2,jz0);
1347             dx21             = _mm_sub_ps(ix2,jx1);
1348             dy21             = _mm_sub_ps(iy2,jy1);
1349             dz21             = _mm_sub_ps(iz2,jz1);
1350             dx22             = _mm_sub_ps(ix2,jx2);
1351             dy22             = _mm_sub_ps(iy2,jy2);
1352             dz22             = _mm_sub_ps(iz2,jz2);
1353
1354             /* Calculate squared distance and things based on it */
1355             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1356             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1357             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1358             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1359             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1360             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1361             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1362             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1363             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1364
1365             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1366             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1367             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1368             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1369             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1370             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1371             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1372             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1373             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1374
1375             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1376             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1377             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1378             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1379             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1380             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1381             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1382             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1383             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1384
1385             fjx0             = _mm_setzero_ps();
1386             fjy0             = _mm_setzero_ps();
1387             fjz0             = _mm_setzero_ps();
1388             fjx1             = _mm_setzero_ps();
1389             fjy1             = _mm_setzero_ps();
1390             fjz1             = _mm_setzero_ps();
1391             fjx2             = _mm_setzero_ps();
1392             fjy2             = _mm_setzero_ps();
1393             fjz2             = _mm_setzero_ps();
1394
1395             /**************************
1396              * CALCULATE INTERACTIONS *
1397              **************************/
1398
1399             /* REACTION-FIELD ELECTROSTATICS */
1400             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1401
1402             /* LENNARD-JONES DISPERSION/REPULSION */
1403
1404             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1405             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1406
1407             fscal            = _mm_add_ps(felec,fvdw);
1408
1409             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1410
1411              /* Update vectorial force */
1412             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1413             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1414             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1415
1416             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1417             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1418             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1419
1420             /**************************
1421              * CALCULATE INTERACTIONS *
1422              **************************/
1423
1424             /* REACTION-FIELD ELECTROSTATICS */
1425             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
1426
1427             fscal            = felec;
1428
1429             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1430
1431              /* Update vectorial force */
1432             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1433             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1434             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1435
1436             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1437             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1438             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1439
1440             /**************************
1441              * CALCULATE INTERACTIONS *
1442              **************************/
1443
1444             /* REACTION-FIELD ELECTROSTATICS */
1445             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
1446
1447             fscal            = felec;
1448
1449             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1450
1451              /* Update vectorial force */
1452             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1453             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1454             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1455
1456             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1457             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1458             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1459
1460             /**************************
1461              * CALCULATE INTERACTIONS *
1462              **************************/
1463
1464             /* REACTION-FIELD ELECTROSTATICS */
1465             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1466
1467             fscal            = felec;
1468
1469             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1470
1471              /* Update vectorial force */
1472             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1473             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1474             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1475
1476             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1477             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1478             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1479
1480             /**************************
1481              * CALCULATE INTERACTIONS *
1482              **************************/
1483
1484             /* REACTION-FIELD ELECTROSTATICS */
1485             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1486
1487             fscal            = felec;
1488
1489             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1490
1491              /* Update vectorial force */
1492             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1493             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1494             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1495
1496             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1497             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1498             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1499
1500             /**************************
1501              * CALCULATE INTERACTIONS *
1502              **************************/
1503
1504             /* REACTION-FIELD ELECTROSTATICS */
1505             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1506
1507             fscal            = felec;
1508
1509             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1510
1511              /* Update vectorial force */
1512             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1513             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1514             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1515
1516             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1517             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1518             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1519
1520             /**************************
1521              * CALCULATE INTERACTIONS *
1522              **************************/
1523
1524             /* REACTION-FIELD ELECTROSTATICS */
1525             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1526
1527             fscal            = felec;
1528
1529             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1530
1531              /* Update vectorial force */
1532             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1533             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1534             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1535
1536             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1537             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1538             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1539
1540             /**************************
1541              * CALCULATE INTERACTIONS *
1542              **************************/
1543
1544             /* REACTION-FIELD ELECTROSTATICS */
1545             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1546
1547             fscal            = felec;
1548
1549             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1550
1551              /* Update vectorial force */
1552             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1553             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1554             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1555
1556             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1557             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1558             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1559
1560             /**************************
1561              * CALCULATE INTERACTIONS *
1562              **************************/
1563
1564             /* REACTION-FIELD ELECTROSTATICS */
1565             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1566
1567             fscal            = felec;
1568
1569             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1570
1571              /* Update vectorial force */
1572             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1573             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1574             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1575
1576             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1577             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1578             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1579
1580             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1581             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1582             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1583             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1584
1585             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1586                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1587
1588             /* Inner loop uses 277 flops */
1589         }
1590
1591         /* End of innermost loop */
1592
1593         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1594                                               f+i_coord_offset,fshift+i_shift_offset);
1595
1596         /* Increment number of inner iterations */
1597         inneriter                  += j_index_end - j_index_start;
1598
1599         /* Outer loop uses 18 flops */
1600     }
1601
1602     /* Increment number of outer iterations */
1603     outeriter        += nri;
1604
1605     /* Update outer/inner flops */
1606
1607     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*277);
1608 }