src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwCSTab_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_ElecRFCut_VdwCSTab_GeomW3W3_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_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     __m128i          vfitab;
 113     __m128i          ifour       = _mm_set1_epi32(4);
 114     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 115     real             *vftab;
 116     __m128           dummy_mask,cutoff_mask;
 117     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 118     __m128           one     = _mm_set1_ps(1.0);
 119     __m128           two     = _mm_set1_ps(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm_set1_ps(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133     krf              = _mm_set1_ps(fr->ic->k_rf);
 134     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 135     crf              = _mm_set1_ps(fr->ic->c_rf);
 136     nvdwtype         = fr->ntype;
 137     vdwparam         = fr->nbfp;
 138     vdwtype          = mdatoms->typeA;
 139
 140     vftab            = kernel_data->table_vdw->data;
 141     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 146     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 147     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 148     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150     jq0              = _mm_set1_ps(charge[inr+0]);
 151     jq1              = _mm_set1_ps(charge[inr+1]);
 152     jq2              = _mm_set1_ps(charge[inr+2]);
 153     vdwjidx0A        = 2*vdwtype[inr+0];
 154     qq00             = _mm_mul_ps(iq0,jq0);
 155     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 156     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 157     qq01             = _mm_mul_ps(iq0,jq1);
 158     qq02             = _mm_mul_ps(iq0,jq2);
 159     qq10             = _mm_mul_ps(iq1,jq0);
 160     qq11             = _mm_mul_ps(iq1,jq1);
 161     qq12             = _mm_mul_ps(iq1,jq2);
 162     qq20             = _mm_mul_ps(iq2,jq0);
 163     qq21             = _mm_mul_ps(iq2,jq1);
 164     qq22             = _mm_mul_ps(iq2,jq2);
 165
 166     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 167     rcutoff_scalar   = fr->rcoulomb;
 168     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 169     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 170
 171     /* Avoid stupid compiler warnings */
 172     jnrA = jnrB = jnrC = jnrD = 0;
 173     j_coord_offsetA = 0;
 174     j_coord_offsetB = 0;
 175     j_coord_offsetC = 0;
 176     j_coord_offsetD = 0;
 177
 178     outeriter        = 0;
 179     inneriter        = 0;
 180
 181     for(iidx=0;iidx<4*DIM;iidx++)
 182     {
 183         scratch[iidx] = 0.0;
 184     }
 185
 186     /* Start outer loop over neighborlists */
 187     for(iidx=0; iidx<nri; iidx++)
 188     {
 189         /* Load shift vector for this list */
 190         i_shift_offset   = DIM*shiftidx[iidx];
 191
 192         /* Load limits for loop over neighbors */
 193         j_index_start    = jindex[iidx];
 194         j_index_end      = jindex[iidx+1];
 195
 196         /* Get outer coordinate index */
 197         inr              = iinr[iidx];
 198         i_coord_offset   = DIM*inr;
 199
 200         /* Load i particle coords and add shift vector */
 201         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 202                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 203
 204         fix0             = _mm_setzero_ps();
 205         fiy0             = _mm_setzero_ps();
 206         fiz0             = _mm_setzero_ps();
 207         fix1             = _mm_setzero_ps();
 208         fiy1             = _mm_setzero_ps();
 209         fiz1             = _mm_setzero_ps();
 210         fix2             = _mm_setzero_ps();
 211         fiy2             = _mm_setzero_ps();
 212         fiz2             = _mm_setzero_ps();
 213
 214         /* Reset potential sums */
 215         velecsum         = _mm_setzero_ps();
 216         vvdwsum          = _mm_setzero_ps();
 217
 218         /* Start inner kernel loop */
 219         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 220         {
 221
 222             /* Get j neighbor index, and coordinate index */
 223             jnrA             = jjnr[jidx];
 224             jnrB             = jjnr[jidx+1];
 225             jnrC             = jjnr[jidx+2];
 226             jnrD             = jjnr[jidx+3];
 227             j_coord_offsetA  = DIM*jnrA;
 228             j_coord_offsetB  = DIM*jnrB;
 229             j_coord_offsetC  = DIM*jnrC;
 230             j_coord_offsetD  = DIM*jnrD;
 231
 232             /* load j atom coordinates */
 233             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 234                                               x+j_coord_offsetC,x+j_coord_offsetD,
 235                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 236
 237             /* Calculate displacement vector */
 238             dx00             = _mm_sub_ps(ix0,jx0);
 239             dy00             = _mm_sub_ps(iy0,jy0);
 240             dz00             = _mm_sub_ps(iz0,jz0);
 241             dx01             = _mm_sub_ps(ix0,jx1);
 242             dy01             = _mm_sub_ps(iy0,jy1);
 243             dz01             = _mm_sub_ps(iz0,jz1);
 244             dx02             = _mm_sub_ps(ix0,jx2);
 245             dy02             = _mm_sub_ps(iy0,jy2);
 246             dz02             = _mm_sub_ps(iz0,jz2);
 247             dx10             = _mm_sub_ps(ix1,jx0);
 248             dy10             = _mm_sub_ps(iy1,jy0);
 249             dz10             = _mm_sub_ps(iz1,jz0);
 250             dx11             = _mm_sub_ps(ix1,jx1);
 251             dy11             = _mm_sub_ps(iy1,jy1);
 252             dz11             = _mm_sub_ps(iz1,jz1);
 253             dx12             = _mm_sub_ps(ix1,jx2);
 254             dy12             = _mm_sub_ps(iy1,jy2);
 255             dz12             = _mm_sub_ps(iz1,jz2);
 256             dx20             = _mm_sub_ps(ix2,jx0);
 257             dy20             = _mm_sub_ps(iy2,jy0);
 258             dz20             = _mm_sub_ps(iz2,jz0);
 259             dx21             = _mm_sub_ps(ix2,jx1);
 260             dy21             = _mm_sub_ps(iy2,jy1);
 261             dz21             = _mm_sub_ps(iz2,jz1);
 262             dx22             = _mm_sub_ps(ix2,jx2);
 263             dy22             = _mm_sub_ps(iy2,jy2);
 264             dz22             = _mm_sub_ps(iz2,jz2);
 265
 266             /* Calculate squared distance and things based on it */
 267             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 268             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 269             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 270             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 271             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 272             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 273             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 274             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 275             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 276
 277             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 278             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 279             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 280             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 281             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 282             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 283             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 284             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 285             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 286
 287             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 288             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 289             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 290             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 291             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 292             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 293             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 294             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 295             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 296
 297             fjx0             = _mm_setzero_ps();
 298             fjy0             = _mm_setzero_ps();
 299             fjz0             = _mm_setzero_ps();
 300             fjx1             = _mm_setzero_ps();
 301             fjy1             = _mm_setzero_ps();
 302             fjz1             = _mm_setzero_ps();
 303             fjx2             = _mm_setzero_ps();
 304             fjy2             = _mm_setzero_ps();
 305             fjz2             = _mm_setzero_ps();
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             if (gmx_mm_any_lt(rsq00,rcutoff2))
 312             {
 313
 314             r00              = _mm_mul_ps(rsq00,rinv00);
 315
 316             /* Calculate table index by multiplying r with table scale and truncate to integer */
 317             rt               = _mm_mul_ps(r00,vftabscale);
 318             vfitab           = _mm_cvttps_epi32(rt);
 319 #ifdef __XOP__
 320             vfeps            = _mm_frcz_ps(rt);
 321 #else
 322             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 323 #endif
 324             twovfeps         = _mm_add_ps(vfeps,vfeps);
 325             vfitab           = _mm_slli_epi32(vfitab,3);
 326
 327             /* REACTION-FIELD ELECTROSTATICS */
 328             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 329             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 330
 331             /* CUBIC SPLINE TABLE DISPERSION */
 332             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 333             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 334             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 335             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 336             _MM_TRANSPOSE4_PS(Y,F,G,H);
 337             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 338             VV               = _mm_macc_ps(vfeps,Fp,Y);
 339             vvdw6            = _mm_mul_ps(c6_00,VV);
 340             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 341             fvdw6            = _mm_mul_ps(c6_00,FF);
 342
 343             /* CUBIC SPLINE TABLE REPULSION */
 344             vfitab           = _mm_add_epi32(vfitab,ifour);
 345             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 346             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 347             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 348             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 349             _MM_TRANSPOSE4_PS(Y,F,G,H);
 350             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 351             VV               = _mm_macc_ps(vfeps,Fp,Y);
 352             vvdw12           = _mm_mul_ps(c12_00,VV);
 353             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 354             fvdw12           = _mm_mul_ps(c12_00,FF);
 355             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 356             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 357
 358             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velec            = _mm_and_ps(velec,cutoff_mask);
 362             velecsum         = _mm_add_ps(velecsum,velec);
 363             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 364             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 365
 366             fscal            = _mm_add_ps(felec,fvdw);
 367
 368             fscal            = _mm_and_ps(fscal,cutoff_mask);
 369
 370              /* Update vectorial force */
 371             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 372             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 373             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 374
 375             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 376             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 377             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 378
 379             }
 380
 381             /**************************
 382              * CALCULATE INTERACTIONS *
 383              **************************/
 384
 385             if (gmx_mm_any_lt(rsq01,rcutoff2))
 386             {
 387
 388             /* REACTION-FIELD ELECTROSTATICS */
 389             velec            = _mm_mul_ps(qq01,_mm_sub_ps(_mm_macc_ps(krf,rsq01,rinv01),crf));
 390             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
 391
 392             cutoff_mask      = _mm_cmplt_ps(rsq01,rcutoff2);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velec            = _mm_and_ps(velec,cutoff_mask);
 396             velecsum         = _mm_add_ps(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             fscal            = _mm_and_ps(fscal,cutoff_mask);
 401
 402              /* Update vectorial force */
 403             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 404             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 405             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 406
 407             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 408             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 409             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 410
 411             }
 412
 413             /**************************
 414              * CALCULATE INTERACTIONS *
 415              **************************/
 416
 417             if (gmx_mm_any_lt(rsq02,rcutoff2))
 418             {
 419
 420             /* REACTION-FIELD ELECTROSTATICS */
 421             velec            = _mm_mul_ps(qq02,_mm_sub_ps(_mm_macc_ps(krf,rsq02,rinv02),crf));
 422             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
 423
 424             cutoff_mask      = _mm_cmplt_ps(rsq02,rcutoff2);
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velec            = _mm_and_ps(velec,cutoff_mask);
 428             velecsum         = _mm_add_ps(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             fscal            = _mm_and_ps(fscal,cutoff_mask);
 433
 434              /* Update vectorial force */
 435             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 436             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 437             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 438
 439             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 440             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 441             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 442
 443             }
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             if (gmx_mm_any_lt(rsq10,rcutoff2))
 450             {
 451
 452             /* REACTION-FIELD ELECTROSTATICS */
 453             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 454             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 455
 456             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velec            = _mm_and_ps(velec,cutoff_mask);
 460             velecsum         = _mm_add_ps(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             fscal            = _mm_and_ps(fscal,cutoff_mask);
 465
 466              /* Update vectorial force */
 467             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 468             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 469             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 470
 471             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 472             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 473             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 474
 475             }
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             if (gmx_mm_any_lt(rsq11,rcutoff2))
 482             {
 483
 484             /* REACTION-FIELD ELECTROSTATICS */
 485             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 486             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 487
 488             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             velec            = _mm_and_ps(velec,cutoff_mask);
 492             velecsum         = _mm_add_ps(velecsum,velec);
 493
 494             fscal            = felec;
 495
 496             fscal            = _mm_and_ps(fscal,cutoff_mask);
 497
 498              /* Update vectorial force */
 499             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 500             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 501             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 502
 503             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 504             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 505             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 506
 507             }
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             if (gmx_mm_any_lt(rsq12,rcutoff2))
 514             {
 515
 516             /* REACTION-FIELD ELECTROSTATICS */
 517             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 518             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 519
 520             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velec            = _mm_and_ps(velec,cutoff_mask);
 524             velecsum         = _mm_add_ps(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm_and_ps(fscal,cutoff_mask);
 529
 530              /* Update vectorial force */
 531             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 532             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 533             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 534
 535             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 536             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 537             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 538
 539             }
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             if (gmx_mm_any_lt(rsq20,rcutoff2))
 546             {
 547
 548             /* REACTION-FIELD ELECTROSTATICS */
 549             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 550             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 551
 552             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velec            = _mm_and_ps(velec,cutoff_mask);
 556             velecsum         = _mm_add_ps(velecsum,velec);
 557
 558             fscal            = felec;
 559
 560             fscal            = _mm_and_ps(fscal,cutoff_mask);
 561
 562              /* Update vectorial force */
 563             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 564             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 565             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 566
 567             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 568             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 569             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 570
 571             }
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             if (gmx_mm_any_lt(rsq21,rcutoff2))
 578             {
 579
 580             /* REACTION-FIELD ELECTROSTATICS */
 581             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 582             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 583
 584             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm_and_ps(velec,cutoff_mask);
 588             velecsum         = _mm_add_ps(velecsum,velec);
 589
 590             fscal            = felec;
 591
 592             fscal            = _mm_and_ps(fscal,cutoff_mask);
 593
 594              /* Update vectorial force */
 595             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 596             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 597             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 598
 599             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 600             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 601             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 602
 603             }
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             if (gmx_mm_any_lt(rsq22,rcutoff2))
 610             {
 611
 612             /* REACTION-FIELD ELECTROSTATICS */
 613             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 614             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 615
 616             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velec            = _mm_and_ps(velec,cutoff_mask);
 620             velecsum         = _mm_add_ps(velecsum,velec);
 621
 622             fscal            = felec;
 623
 624             fscal            = _mm_and_ps(fscal,cutoff_mask);
 625
 626              /* Update vectorial force */
 627             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 628             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 629             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 630
 631             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 632             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 633             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 634
 635             }
 636
 637             fjptrA             = f+j_coord_offsetA;
 638             fjptrB             = f+j_coord_offsetB;
 639             fjptrC             = f+j_coord_offsetC;
 640             fjptrD             = f+j_coord_offsetD;
 641
 642             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 643                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 644
 645             /* Inner loop uses 387 flops */
 646         }
 647
 648         if(jidx<j_index_end)
 649         {
 650
 651             /* Get j neighbor index, and coordinate index */
 652             jnrlistA         = jjnr[jidx];
 653             jnrlistB         = jjnr[jidx+1];
 654             jnrlistC         = jjnr[jidx+2];
 655             jnrlistD         = jjnr[jidx+3];
 656             /* Sign of each element will be negative for non-real atoms.
 657              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 658              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 659              */
 660             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 661             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 662             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 663             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 664             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 665             j_coord_offsetA  = DIM*jnrA;
 666             j_coord_offsetB  = DIM*jnrB;
 667             j_coord_offsetC  = DIM*jnrC;
 668             j_coord_offsetD  = DIM*jnrD;
 669
 670             /* load j atom coordinates */
 671             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 672                                               x+j_coord_offsetC,x+j_coord_offsetD,
 673                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 674
 675             /* Calculate displacement vector */
 676             dx00             = _mm_sub_ps(ix0,jx0);
 677             dy00             = _mm_sub_ps(iy0,jy0);
 678             dz00             = _mm_sub_ps(iz0,jz0);
 679             dx01             = _mm_sub_ps(ix0,jx1);
 680             dy01             = _mm_sub_ps(iy0,jy1);
 681             dz01             = _mm_sub_ps(iz0,jz1);
 682             dx02             = _mm_sub_ps(ix0,jx2);
 683             dy02             = _mm_sub_ps(iy0,jy2);
 684             dz02             = _mm_sub_ps(iz0,jz2);
 685             dx10             = _mm_sub_ps(ix1,jx0);
 686             dy10             = _mm_sub_ps(iy1,jy0);
 687             dz10             = _mm_sub_ps(iz1,jz0);
 688             dx11             = _mm_sub_ps(ix1,jx1);
 689             dy11             = _mm_sub_ps(iy1,jy1);
 690             dz11             = _mm_sub_ps(iz1,jz1);
 691             dx12             = _mm_sub_ps(ix1,jx2);
 692             dy12             = _mm_sub_ps(iy1,jy2);
 693             dz12             = _mm_sub_ps(iz1,jz2);
 694             dx20             = _mm_sub_ps(ix2,jx0);
 695             dy20             = _mm_sub_ps(iy2,jy0);
 696             dz20             = _mm_sub_ps(iz2,jz0);
 697             dx21             = _mm_sub_ps(ix2,jx1);
 698             dy21             = _mm_sub_ps(iy2,jy1);
 699             dz21             = _mm_sub_ps(iz2,jz1);
 700             dx22             = _mm_sub_ps(ix2,jx2);
 701             dy22             = _mm_sub_ps(iy2,jy2);
 702             dz22             = _mm_sub_ps(iz2,jz2);
 703
 704             /* Calculate squared distance and things based on it */
 705             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 706             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 707             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 708             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 709             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 710             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 711             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 712             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 713             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 714
 715             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 716             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 717             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 718             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 719             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 720             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 721             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 722             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 723             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 724
 725             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 726             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 727             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 728             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 729             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 730             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 731             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 732             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 733             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 734
 735             fjx0             = _mm_setzero_ps();
 736             fjy0             = _mm_setzero_ps();
 737             fjz0             = _mm_setzero_ps();
 738             fjx1             = _mm_setzero_ps();
 739             fjy1             = _mm_setzero_ps();
 740             fjz1             = _mm_setzero_ps();
 741             fjx2             = _mm_setzero_ps();
 742             fjy2             = _mm_setzero_ps();
 743             fjz2             = _mm_setzero_ps();
 744
 745             /**************************
 746              * CALCULATE INTERACTIONS *
 747              **************************/
 748
 749             if (gmx_mm_any_lt(rsq00,rcutoff2))
 750             {
 751
 752             r00              = _mm_mul_ps(rsq00,rinv00);
 753             r00              = _mm_andnot_ps(dummy_mask,r00);
 754
 755             /* Calculate table index by multiplying r with table scale and truncate to integer */
 756             rt               = _mm_mul_ps(r00,vftabscale);
 757             vfitab           = _mm_cvttps_epi32(rt);
 758 #ifdef __XOP__
 759             vfeps            = _mm_frcz_ps(rt);
 760 #else
 761             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 762 #endif
 763             twovfeps         = _mm_add_ps(vfeps,vfeps);
 764             vfitab           = _mm_slli_epi32(vfitab,3);
 765
 766             /* REACTION-FIELD ELECTROSTATICS */
 767             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 768             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 769
 770             /* CUBIC SPLINE TABLE DISPERSION */
 771             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 772             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 773             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 774             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 775             _MM_TRANSPOSE4_PS(Y,F,G,H);
 776             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 777             VV               = _mm_macc_ps(vfeps,Fp,Y);
 778             vvdw6            = _mm_mul_ps(c6_00,VV);
 779             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 780             fvdw6            = _mm_mul_ps(c6_00,FF);
 781
 782             /* CUBIC SPLINE TABLE REPULSION */
 783             vfitab           = _mm_add_epi32(vfitab,ifour);
 784             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 785             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 786             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 787             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 788             _MM_TRANSPOSE4_PS(Y,F,G,H);
 789             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 790             VV               = _mm_macc_ps(vfeps,Fp,Y);
 791             vvdw12           = _mm_mul_ps(c12_00,VV);
 792             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 793             fvdw12           = _mm_mul_ps(c12_00,FF);
 794             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 795             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 796
 797             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 798
 799             /* Update potential sum for this i atom from the interaction with this j atom. */
 800             velec            = _mm_and_ps(velec,cutoff_mask);
 801             velec            = _mm_andnot_ps(dummy_mask,velec);
 802             velecsum         = _mm_add_ps(velecsum,velec);
 803             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 804             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 805             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 806
 807             fscal            = _mm_add_ps(felec,fvdw);
 808
 809             fscal            = _mm_and_ps(fscal,cutoff_mask);
 810
 811             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 812
 813              /* Update vectorial force */
 814             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 815             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 816             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 817
 818             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 819             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 820             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 821
 822             }
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             if (gmx_mm_any_lt(rsq01,rcutoff2))
 829             {
 830
 831             /* REACTION-FIELD ELECTROSTATICS */
 832             velec            = _mm_mul_ps(qq01,_mm_sub_ps(_mm_macc_ps(krf,rsq01,rinv01),crf));
 833             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
 834
 835             cutoff_mask      = _mm_cmplt_ps(rsq01,rcutoff2);
 836
 837             /* Update potential sum for this i atom from the interaction with this j atom. */
 838             velec            = _mm_and_ps(velec,cutoff_mask);
 839             velec            = _mm_andnot_ps(dummy_mask,velec);
 840             velecsum         = _mm_add_ps(velecsum,velec);
 841
 842             fscal            = felec;
 843
 844             fscal            = _mm_and_ps(fscal,cutoff_mask);
 845
 846             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 847
 848              /* Update vectorial force */
 849             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 850             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 851             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 852
 853             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 854             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 855             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 856
 857             }
 858
 859             /**************************
 860              * CALCULATE INTERACTIONS *
 861              **************************/
 862
 863             if (gmx_mm_any_lt(rsq02,rcutoff2))
 864             {
 865
 866             /* REACTION-FIELD ELECTROSTATICS */
 867             velec            = _mm_mul_ps(qq02,_mm_sub_ps(_mm_macc_ps(krf,rsq02,rinv02),crf));
 868             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
 869
 870             cutoff_mask      = _mm_cmplt_ps(rsq02,rcutoff2);
 871
 872             /* Update potential sum for this i atom from the interaction with this j atom. */
 873             velec            = _mm_and_ps(velec,cutoff_mask);
 874             velec            = _mm_andnot_ps(dummy_mask,velec);
 875             velecsum         = _mm_add_ps(velecsum,velec);
 876
 877             fscal            = felec;
 878
 879             fscal            = _mm_and_ps(fscal,cutoff_mask);
 880
 881             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 882
 883              /* Update vectorial force */
 884             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 885             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 886             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 887
 888             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 889             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 890             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 891
 892             }
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             if (gmx_mm_any_lt(rsq10,rcutoff2))
 899             {
 900
 901             /* REACTION-FIELD ELECTROSTATICS */
 902             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 903             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 904
 905             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 906
 907             /* Update potential sum for this i atom from the interaction with this j atom. */
 908             velec            = _mm_and_ps(velec,cutoff_mask);
 909             velec            = _mm_andnot_ps(dummy_mask,velec);
 910             velecsum         = _mm_add_ps(velecsum,velec);
 911
 912             fscal            = felec;
 913
 914             fscal            = _mm_and_ps(fscal,cutoff_mask);
 915
 916             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 917
 918              /* Update vectorial force */
 919             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 920             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 921             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 922
 923             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 924             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 925             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 926
 927             }
 928
 929             /**************************
 930              * CALCULATE INTERACTIONS *
 931              **************************/
 932
 933             if (gmx_mm_any_lt(rsq11,rcutoff2))
 934             {
 935
 936             /* REACTION-FIELD ELECTROSTATICS */
 937             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 938             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 939
 940             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
 941
 942             /* Update potential sum for this i atom from the interaction with this j atom. */
 943             velec            = _mm_and_ps(velec,cutoff_mask);
 944             velec            = _mm_andnot_ps(dummy_mask,velec);
 945             velecsum         = _mm_add_ps(velecsum,velec);
 946
 947             fscal            = felec;
 948
 949             fscal            = _mm_and_ps(fscal,cutoff_mask);
 950
 951             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 952
 953              /* Update vectorial force */
 954             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 955             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 956             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 957
 958             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 959             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 960             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 961
 962             }
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             if (gmx_mm_any_lt(rsq12,rcutoff2))
 969             {
 970
 971             /* REACTION-FIELD ELECTROSTATICS */
 972             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 973             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 974
 975             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
 976
 977             /* Update potential sum for this i atom from the interaction with this j atom. */
 978             velec            = _mm_and_ps(velec,cutoff_mask);
 979             velec            = _mm_andnot_ps(dummy_mask,velec);
 980             velecsum         = _mm_add_ps(velecsum,velec);
 981
 982             fscal            = felec;
 983
 984             fscal            = _mm_and_ps(fscal,cutoff_mask);
 985
 986             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 987
 988              /* Update vectorial force */
 989             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 990             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 991             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 992
 993             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 994             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 995             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 996
 997             }
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             if (gmx_mm_any_lt(rsq20,rcutoff2))
1004             {
1005
1006             /* REACTION-FIELD ELECTROSTATICS */
1007             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
1008             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1009
1010             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1011
1012             /* Update potential sum for this i atom from the interaction with this j atom. */
1013             velec            = _mm_and_ps(velec,cutoff_mask);
1014             velec            = _mm_andnot_ps(dummy_mask,velec);
1015             velecsum         = _mm_add_ps(velecsum,velec);
1016
1017             fscal            = felec;
1018
1019             fscal            = _mm_and_ps(fscal,cutoff_mask);
1020
1021             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1022
1023              /* Update vectorial force */
1024             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1025             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1026             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1027
1028             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1029             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1030             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1031
1032             }
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             if (gmx_mm_any_lt(rsq21,rcutoff2))
1039             {
1040
1041             /* REACTION-FIELD ELECTROSTATICS */
1042             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
1043             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1044
1045             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1046
1047             /* Update potential sum for this i atom from the interaction with this j atom. */
1048             velec            = _mm_and_ps(velec,cutoff_mask);
1049             velec            = _mm_andnot_ps(dummy_mask,velec);
1050             velecsum         = _mm_add_ps(velecsum,velec);
1051
1052             fscal            = felec;
1053
1054             fscal            = _mm_and_ps(fscal,cutoff_mask);
1055
1056             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1057
1058              /* Update vectorial force */
1059             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1060             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1061             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1062
1063             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1064             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1065             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1066
1067             }
1068
1069             /**************************
1070              * CALCULATE INTERACTIONS *
1071              **************************/
1072
1073             if (gmx_mm_any_lt(rsq22,rcutoff2))
1074             {
1075
1076             /* REACTION-FIELD ELECTROSTATICS */
1077             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
1078             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1079
1080             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
1081
1082             /* Update potential sum for this i atom from the interaction with this j atom. */
1083             velec            = _mm_and_ps(velec,cutoff_mask);
1084             velec            = _mm_andnot_ps(dummy_mask,velec);
1085             velecsum         = _mm_add_ps(velecsum,velec);
1086
1087             fscal            = felec;
1088
1089             fscal            = _mm_and_ps(fscal,cutoff_mask);
1090
1091             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1092
1093              /* Update vectorial force */
1094             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1095             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1096             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1097
1098             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1099             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1100             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1101
1102             }
1103
1104             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1105             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1106             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1107             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1108
1109             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1110                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1111
1112             /* Inner loop uses 388 flops */
1113         }
1114
1115         /* End of innermost loop */
1116
1117         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1118                                               f+i_coord_offset,fshift+i_shift_offset);
1119
1120         ggid                        = gid[iidx];
1121         /* Update potential energies */
1122         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1123         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1124
1125         /* Increment number of inner iterations */
1126         inneriter                  += j_index_end - j_index_start;
1127
1128         /* Outer loop uses 20 flops */
1129     }
1130
1131     /* Increment number of outer iterations */
1132     outeriter        += nri;
1133
1134     /* Update outer/inner flops */
1135
1136     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*388);
1137 }
1138 /*
1139  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_F_avx_128_fma_single
1140  * Electrostatics interaction: ReactionField
1141  * VdW interaction:            CubicSplineTable
1142  * Geometry:                   Water3-Water3
1143  * Calculate force/pot:        Force
1144  */
1145 void
1146 nb_kernel_ElecRFCut_VdwCSTab_GeomW3W3_F_avx_128_fma_single
1147                     (t_nblist                    * gmx_restrict       nlist,
1148                      rvec                        * gmx_restrict          xx,
1149                      rvec                        * gmx_restrict          ff,
1150                      t_forcerec                  * gmx_restrict          fr,
1151                      t_mdatoms                   * gmx_restrict     mdatoms,
1152                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1153                      t_nrnb                      * gmx_restrict        nrnb)
1154 {
1155     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1156      * just 0 for non-waters.
1157      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1158      * jnr indices corresponding to data put in the four positions in the SIMD register.
1159      */
1160     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1161     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1162     int              jnrA,jnrB,jnrC,jnrD;
1163     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1164     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1165     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1166     real             rcutoff_scalar;
1167     real             *shiftvec,*fshift,*x,*f;
1168     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1169     real             scratch[4*DIM];
1170     __m128           fscal,rcutoff,rcutoff2,jidxall;
1171     int              vdwioffset0;
1172     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1173     int              vdwioffset1;
1174     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1175     int              vdwioffset2;
1176     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1177     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1178     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1179     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1180     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1181     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1182     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1183     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1184     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1185     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1186     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1187     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1188     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1189     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1190     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1191     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1192     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1193     real             *charge;
1194     int              nvdwtype;
1195     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1196     int              *vdwtype;
1197     real             *vdwparam;
1198     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1199     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1200     __m128i          vfitab;
1201     __m128i          ifour       = _mm_set1_epi32(4);
1202     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
1203     real             *vftab;
1204     __m128           dummy_mask,cutoff_mask;
1205     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1206     __m128           one     = _mm_set1_ps(1.0);
1207     __m128           two     = _mm_set1_ps(2.0);
1208     x                = xx[0];
1209     f                = ff[0];
1210
1211     nri              = nlist->nri;
1212     iinr             = nlist->iinr;
1213     jindex           = nlist->jindex;
1214     jjnr             = nlist->jjnr;
1215     shiftidx         = nlist->shift;
1216     gid              = nlist->gid;
1217     shiftvec         = fr->shift_vec[0];
1218     fshift           = fr->fshift[0];
1219     facel            = _mm_set1_ps(fr->epsfac);
1220     charge           = mdatoms->chargeA;
1221     krf              = _mm_set1_ps(fr->ic->k_rf);
1222     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
1223     crf              = _mm_set1_ps(fr->ic->c_rf);
1224     nvdwtype         = fr->ntype;
1225     vdwparam         = fr->nbfp;
1226     vdwtype          = mdatoms->typeA;
1227
1228     vftab            = kernel_data->table_vdw->data;
1229     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
1230
1231     /* Setup water-specific parameters */
1232     inr              = nlist->iinr[0];
1233     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1234     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1235     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1236     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1237
1238     jq0              = _mm_set1_ps(charge[inr+0]);
1239     jq1              = _mm_set1_ps(charge[inr+1]);
1240     jq2              = _mm_set1_ps(charge[inr+2]);
1241     vdwjidx0A        = 2*vdwtype[inr+0];
1242     qq00             = _mm_mul_ps(iq0,jq0);
1243     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1244     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1245     qq01             = _mm_mul_ps(iq0,jq1);
1246     qq02             = _mm_mul_ps(iq0,jq2);
1247     qq10             = _mm_mul_ps(iq1,jq0);
1248     qq11             = _mm_mul_ps(iq1,jq1);
1249     qq12             = _mm_mul_ps(iq1,jq2);
1250     qq20             = _mm_mul_ps(iq2,jq0);
1251     qq21             = _mm_mul_ps(iq2,jq1);
1252     qq22             = _mm_mul_ps(iq2,jq2);
1253
1254     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1255     rcutoff_scalar   = fr->rcoulomb;
1256     rcutoff          = _mm_set1_ps(rcutoff_scalar);
1257     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
1258
1259     /* Avoid stupid compiler warnings */
1260     jnrA = jnrB = jnrC = jnrD = 0;
1261     j_coord_offsetA = 0;
1262     j_coord_offsetB = 0;
1263     j_coord_offsetC = 0;
1264     j_coord_offsetD = 0;
1265
1266     outeriter        = 0;
1267     inneriter        = 0;
1268
1269     for(iidx=0;iidx<4*DIM;iidx++)
1270     {
1271         scratch[iidx] = 0.0;
1272     }
1273
1274     /* Start outer loop over neighborlists */
1275     for(iidx=0; iidx<nri; iidx++)
1276     {
1277         /* Load shift vector for this list */
1278         i_shift_offset   = DIM*shiftidx[iidx];
1279
1280         /* Load limits for loop over neighbors */
1281         j_index_start    = jindex[iidx];
1282         j_index_end      = jindex[iidx+1];
1283
1284         /* Get outer coordinate index */
1285         inr              = iinr[iidx];
1286         i_coord_offset   = DIM*inr;
1287
1288         /* Load i particle coords and add shift vector */
1289         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1290                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1291
1292         fix0             = _mm_setzero_ps();
1293         fiy0             = _mm_setzero_ps();
1294         fiz0             = _mm_setzero_ps();
1295         fix1             = _mm_setzero_ps();
1296         fiy1             = _mm_setzero_ps();
1297         fiz1             = _mm_setzero_ps();
1298         fix2             = _mm_setzero_ps();
1299         fiy2             = _mm_setzero_ps();
1300         fiz2             = _mm_setzero_ps();
1301
1302         /* Start inner kernel loop */
1303         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1304         {
1305
1306             /* Get j neighbor index, and coordinate index */
1307             jnrA             = jjnr[jidx];
1308             jnrB             = jjnr[jidx+1];
1309             jnrC             = jjnr[jidx+2];
1310             jnrD             = jjnr[jidx+3];
1311             j_coord_offsetA  = DIM*jnrA;
1312             j_coord_offsetB  = DIM*jnrB;
1313             j_coord_offsetC  = DIM*jnrC;
1314             j_coord_offsetD  = DIM*jnrD;
1315
1316             /* load j atom coordinates */
1317             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1318                                               x+j_coord_offsetC,x+j_coord_offsetD,
1319                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1320
1321             /* Calculate displacement vector */
1322             dx00             = _mm_sub_ps(ix0,jx0);
1323             dy00             = _mm_sub_ps(iy0,jy0);
1324             dz00             = _mm_sub_ps(iz0,jz0);
1325             dx01             = _mm_sub_ps(ix0,jx1);
1326             dy01             = _mm_sub_ps(iy0,jy1);
1327             dz01             = _mm_sub_ps(iz0,jz1);
1328             dx02             = _mm_sub_ps(ix0,jx2);
1329             dy02             = _mm_sub_ps(iy0,jy2);
1330             dz02             = _mm_sub_ps(iz0,jz2);
1331             dx10             = _mm_sub_ps(ix1,jx0);
1332             dy10             = _mm_sub_ps(iy1,jy0);
1333             dz10             = _mm_sub_ps(iz1,jz0);
1334             dx11             = _mm_sub_ps(ix1,jx1);
1335             dy11             = _mm_sub_ps(iy1,jy1);
1336             dz11             = _mm_sub_ps(iz1,jz1);
1337             dx12             = _mm_sub_ps(ix1,jx2);
1338             dy12             = _mm_sub_ps(iy1,jy2);
1339             dz12             = _mm_sub_ps(iz1,jz2);
1340             dx20             = _mm_sub_ps(ix2,jx0);
1341             dy20             = _mm_sub_ps(iy2,jy0);
1342             dz20             = _mm_sub_ps(iz2,jz0);
1343             dx21             = _mm_sub_ps(ix2,jx1);
1344             dy21             = _mm_sub_ps(iy2,jy1);
1345             dz21             = _mm_sub_ps(iz2,jz1);
1346             dx22             = _mm_sub_ps(ix2,jx2);
1347             dy22             = _mm_sub_ps(iy2,jy2);
1348             dz22             = _mm_sub_ps(iz2,jz2);
1349
1350             /* Calculate squared distance and things based on it */
1351             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1352             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1353             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1354             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1355             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1356             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1357             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1358             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1359             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1360
1361             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1362             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1363             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1364             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1365             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1366             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1367             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1368             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1369             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1370
1371             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1372             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1373             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1374             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1375             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1376             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1377             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1378             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1379             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1380
1381             fjx0             = _mm_setzero_ps();
1382             fjy0             = _mm_setzero_ps();
1383             fjz0             = _mm_setzero_ps();
1384             fjx1             = _mm_setzero_ps();
1385             fjy1             = _mm_setzero_ps();
1386             fjz1             = _mm_setzero_ps();
1387             fjx2             = _mm_setzero_ps();
1388             fjy2             = _mm_setzero_ps();
1389             fjz2             = _mm_setzero_ps();
1390
1391             /**************************
1392              * CALCULATE INTERACTIONS *
1393              **************************/
1394
1395             if (gmx_mm_any_lt(rsq00,rcutoff2))
1396             {
1397
1398             r00              = _mm_mul_ps(rsq00,rinv00);
1399
1400             /* Calculate table index by multiplying r with table scale and truncate to integer */
1401             rt               = _mm_mul_ps(r00,vftabscale);
1402             vfitab           = _mm_cvttps_epi32(rt);
1403 #ifdef __XOP__
1404             vfeps            = _mm_frcz_ps(rt);
1405 #else
1406             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1407 #endif
1408             twovfeps         = _mm_add_ps(vfeps,vfeps);
1409             vfitab           = _mm_slli_epi32(vfitab,3);
1410
1411             /* REACTION-FIELD ELECTROSTATICS */
1412             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1413
1414             /* CUBIC SPLINE TABLE DISPERSION */
1415             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1416             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1417             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1418             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1419             _MM_TRANSPOSE4_PS(Y,F,G,H);
1420             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1421             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1422             fvdw6            = _mm_mul_ps(c6_00,FF);
1423
1424             /* CUBIC SPLINE TABLE REPULSION */
1425             vfitab           = _mm_add_epi32(vfitab,ifour);
1426             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1427             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1428             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1429             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1430             _MM_TRANSPOSE4_PS(Y,F,G,H);
1431             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1432             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1433             fvdw12           = _mm_mul_ps(c12_00,FF);
1434             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1435
1436             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1437
1438             fscal            = _mm_add_ps(felec,fvdw);
1439
1440             fscal            = _mm_and_ps(fscal,cutoff_mask);
1441
1442              /* Update vectorial force */
1443             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1444             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1445             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1446
1447             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1448             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1449             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1450
1451             }
1452
1453             /**************************
1454              * CALCULATE INTERACTIONS *
1455              **************************/
1456
1457             if (gmx_mm_any_lt(rsq01,rcutoff2))
1458             {
1459
1460             /* REACTION-FIELD ELECTROSTATICS */
1461             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
1462
1463             cutoff_mask      = _mm_cmplt_ps(rsq01,rcutoff2);
1464
1465             fscal            = felec;
1466
1467             fscal            = _mm_and_ps(fscal,cutoff_mask);
1468
1469              /* Update vectorial force */
1470             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1471             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1472             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1473
1474             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1475             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1476             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1477
1478             }
1479
1480             /**************************
1481              * CALCULATE INTERACTIONS *
1482              **************************/
1483
1484             if (gmx_mm_any_lt(rsq02,rcutoff2))
1485             {
1486
1487             /* REACTION-FIELD ELECTROSTATICS */
1488             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
1489
1490             cutoff_mask      = _mm_cmplt_ps(rsq02,rcutoff2);
1491
1492             fscal            = felec;
1493
1494             fscal            = _mm_and_ps(fscal,cutoff_mask);
1495
1496              /* Update vectorial force */
1497             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1498             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1499             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1500
1501             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1502             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1503             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1504
1505             }
1506
1507             /**************************
1508              * CALCULATE INTERACTIONS *
1509              **************************/
1510
1511             if (gmx_mm_any_lt(rsq10,rcutoff2))
1512             {
1513
1514             /* REACTION-FIELD ELECTROSTATICS */
1515             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1516
1517             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1518
1519             fscal            = felec;
1520
1521             fscal            = _mm_and_ps(fscal,cutoff_mask);
1522
1523              /* Update vectorial force */
1524             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1525             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1526             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1527
1528             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1529             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1530             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1531
1532             }
1533
1534             /**************************
1535              * CALCULATE INTERACTIONS *
1536              **************************/
1537
1538             if (gmx_mm_any_lt(rsq11,rcutoff2))
1539             {
1540
1541             /* REACTION-FIELD ELECTROSTATICS */
1542             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1543
1544             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1545
1546             fscal            = felec;
1547
1548             fscal            = _mm_and_ps(fscal,cutoff_mask);
1549
1550              /* Update vectorial force */
1551             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1552             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1553             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1554
1555             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1556             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1557             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1558
1559             }
1560
1561             /**************************
1562              * CALCULATE INTERACTIONS *
1563              **************************/
1564
1565             if (gmx_mm_any_lt(rsq12,rcutoff2))
1566             {
1567
1568             /* REACTION-FIELD ELECTROSTATICS */
1569             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1570
1571             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1572
1573             fscal            = felec;
1574
1575             fscal            = _mm_and_ps(fscal,cutoff_mask);
1576
1577              /* Update vectorial force */
1578             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1579             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1580             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1581
1582             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1583             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1584             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1585
1586             }
1587
1588             /**************************
1589              * CALCULATE INTERACTIONS *
1590              **************************/
1591
1592             if (gmx_mm_any_lt(rsq20,rcutoff2))
1593             {
1594
1595             /* REACTION-FIELD ELECTROSTATICS */
1596             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1597
1598             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1599
1600             fscal            = felec;
1601
1602             fscal            = _mm_and_ps(fscal,cutoff_mask);
1603
1604              /* Update vectorial force */
1605             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1606             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1607             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1608
1609             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1610             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1611             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1612
1613             }
1614
1615             /**************************
1616              * CALCULATE INTERACTIONS *
1617              **************************/
1618
1619             if (gmx_mm_any_lt(rsq21,rcutoff2))
1620             {
1621
1622             /* REACTION-FIELD ELECTROSTATICS */
1623             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1624
1625             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1626
1627             fscal            = felec;
1628
1629             fscal            = _mm_and_ps(fscal,cutoff_mask);
1630
1631              /* Update vectorial force */
1632             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1633             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1634             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1635
1636             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1637             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1638             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1639
1640             }
1641
1642             /**************************
1643              * CALCULATE INTERACTIONS *
1644              **************************/
1645
1646             if (gmx_mm_any_lt(rsq22,rcutoff2))
1647             {
1648
1649             /* REACTION-FIELD ELECTROSTATICS */
1650             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1651
1652             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
1653
1654             fscal            = felec;
1655
1656             fscal            = _mm_and_ps(fscal,cutoff_mask);
1657
1658              /* Update vectorial force */
1659             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1660             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1661             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1662
1663             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1664             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1665             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1666
1667             }
1668
1669             fjptrA             = f+j_coord_offsetA;
1670             fjptrB             = f+j_coord_offsetB;
1671             fjptrC             = f+j_coord_offsetC;
1672             fjptrD             = f+j_coord_offsetD;
1673
1674             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1675                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1676
1677             /* Inner loop uses 324 flops */
1678         }
1679
1680         if(jidx<j_index_end)
1681         {
1682
1683             /* Get j neighbor index, and coordinate index */
1684             jnrlistA         = jjnr[jidx];
1685             jnrlistB         = jjnr[jidx+1];
1686             jnrlistC         = jjnr[jidx+2];
1687             jnrlistD         = jjnr[jidx+3];
1688             /* Sign of each element will be negative for non-real atoms.
1689              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1690              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1691              */
1692             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1693             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1694             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1695             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1696             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1697             j_coord_offsetA  = DIM*jnrA;
1698             j_coord_offsetB  = DIM*jnrB;
1699             j_coord_offsetC  = DIM*jnrC;
1700             j_coord_offsetD  = DIM*jnrD;
1701
1702             /* load j atom coordinates */
1703             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1704                                               x+j_coord_offsetC,x+j_coord_offsetD,
1705                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1706
1707             /* Calculate displacement vector */
1708             dx00             = _mm_sub_ps(ix0,jx0);
1709             dy00             = _mm_sub_ps(iy0,jy0);
1710             dz00             = _mm_sub_ps(iz0,jz0);
1711             dx01             = _mm_sub_ps(ix0,jx1);
1712             dy01             = _mm_sub_ps(iy0,jy1);
1713             dz01             = _mm_sub_ps(iz0,jz1);
1714             dx02             = _mm_sub_ps(ix0,jx2);
1715             dy02             = _mm_sub_ps(iy0,jy2);
1716             dz02             = _mm_sub_ps(iz0,jz2);
1717             dx10             = _mm_sub_ps(ix1,jx0);
1718             dy10             = _mm_sub_ps(iy1,jy0);
1719             dz10             = _mm_sub_ps(iz1,jz0);
1720             dx11             = _mm_sub_ps(ix1,jx1);
1721             dy11             = _mm_sub_ps(iy1,jy1);
1722             dz11             = _mm_sub_ps(iz1,jz1);
1723             dx12             = _mm_sub_ps(ix1,jx2);
1724             dy12             = _mm_sub_ps(iy1,jy2);
1725             dz12             = _mm_sub_ps(iz1,jz2);
1726             dx20             = _mm_sub_ps(ix2,jx0);
1727             dy20             = _mm_sub_ps(iy2,jy0);
1728             dz20             = _mm_sub_ps(iz2,jz0);
1729             dx21             = _mm_sub_ps(ix2,jx1);
1730             dy21             = _mm_sub_ps(iy2,jy1);
1731             dz21             = _mm_sub_ps(iz2,jz1);
1732             dx22             = _mm_sub_ps(ix2,jx2);
1733             dy22             = _mm_sub_ps(iy2,jy2);
1734             dz22             = _mm_sub_ps(iz2,jz2);
1735
1736             /* Calculate squared distance and things based on it */
1737             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1738             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1739             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1740             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1741             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1742             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1743             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1744             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1745             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1746
1747             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1748             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1749             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1750             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1751             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1752             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1753             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1754             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1755             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1756
1757             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1758             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1759             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1760             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1761             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1762             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1763             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1764             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1765             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1766
1767             fjx0             = _mm_setzero_ps();
1768             fjy0             = _mm_setzero_ps();
1769             fjz0             = _mm_setzero_ps();
1770             fjx1             = _mm_setzero_ps();
1771             fjy1             = _mm_setzero_ps();
1772             fjz1             = _mm_setzero_ps();
1773             fjx2             = _mm_setzero_ps();
1774             fjy2             = _mm_setzero_ps();
1775             fjz2             = _mm_setzero_ps();
1776
1777             /**************************
1778              * CALCULATE INTERACTIONS *
1779              **************************/
1780
1781             if (gmx_mm_any_lt(rsq00,rcutoff2))
1782             {
1783
1784             r00              = _mm_mul_ps(rsq00,rinv00);
1785             r00              = _mm_andnot_ps(dummy_mask,r00);
1786
1787             /* Calculate table index by multiplying r with table scale and truncate to integer */
1788             rt               = _mm_mul_ps(r00,vftabscale);
1789             vfitab           = _mm_cvttps_epi32(rt);
1790 #ifdef __XOP__
1791             vfeps            = _mm_frcz_ps(rt);
1792 #else
1793             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1794 #endif
1795             twovfeps         = _mm_add_ps(vfeps,vfeps);
1796             vfitab           = _mm_slli_epi32(vfitab,3);
1797
1798             /* REACTION-FIELD ELECTROSTATICS */
1799             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1800
1801             /* CUBIC SPLINE TABLE DISPERSION */
1802             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1803             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1804             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1805             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1806             _MM_TRANSPOSE4_PS(Y,F,G,H);
1807             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1808             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1809             fvdw6            = _mm_mul_ps(c6_00,FF);
1810
1811             /* CUBIC SPLINE TABLE REPULSION */
1812             vfitab           = _mm_add_epi32(vfitab,ifour);
1813             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1814             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1815             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1816             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1817             _MM_TRANSPOSE4_PS(Y,F,G,H);
1818             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1819             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1820             fvdw12           = _mm_mul_ps(c12_00,FF);
1821             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1822
1823             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1824
1825             fscal            = _mm_add_ps(felec,fvdw);
1826
1827             fscal            = _mm_and_ps(fscal,cutoff_mask);
1828
1829             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1830
1831              /* Update vectorial force */
1832             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1833             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1834             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1835
1836             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1837             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1838             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1839
1840             }
1841
1842             /**************************
1843              * CALCULATE INTERACTIONS *
1844              **************************/
1845
1846             if (gmx_mm_any_lt(rsq01,rcutoff2))
1847             {
1848
1849             /* REACTION-FIELD ELECTROSTATICS */
1850             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
1851
1852             cutoff_mask      = _mm_cmplt_ps(rsq01,rcutoff2);
1853
1854             fscal            = felec;
1855
1856             fscal            = _mm_and_ps(fscal,cutoff_mask);
1857
1858             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1859
1860              /* Update vectorial force */
1861             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1862             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1863             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1864
1865             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1866             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1867             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1868
1869             }
1870
1871             /**************************
1872              * CALCULATE INTERACTIONS *
1873              **************************/
1874
1875             if (gmx_mm_any_lt(rsq02,rcutoff2))
1876             {
1877
1878             /* REACTION-FIELD ELECTROSTATICS */
1879             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
1880
1881             cutoff_mask      = _mm_cmplt_ps(rsq02,rcutoff2);
1882
1883             fscal            = felec;
1884
1885             fscal            = _mm_and_ps(fscal,cutoff_mask);
1886
1887             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1888
1889              /* Update vectorial force */
1890             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1891             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1892             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1893
1894             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1895             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1896             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1897
1898             }
1899
1900             /**************************
1901              * CALCULATE INTERACTIONS *
1902              **************************/
1903
1904             if (gmx_mm_any_lt(rsq10,rcutoff2))
1905             {
1906
1907             /* REACTION-FIELD ELECTROSTATICS */
1908             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1909
1910             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1911
1912             fscal            = felec;
1913
1914             fscal            = _mm_and_ps(fscal,cutoff_mask);
1915
1916             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1917
1918              /* Update vectorial force */
1919             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1920             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1921             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1922
1923             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1924             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1925             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1926
1927             }
1928
1929             /**************************
1930              * CALCULATE INTERACTIONS *
1931              **************************/
1932
1933             if (gmx_mm_any_lt(rsq11,rcutoff2))
1934             {
1935
1936             /* REACTION-FIELD ELECTROSTATICS */
1937             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1938
1939             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1940
1941             fscal            = felec;
1942
1943             fscal            = _mm_and_ps(fscal,cutoff_mask);
1944
1945             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1946
1947              /* Update vectorial force */
1948             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1949             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1950             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1951
1952             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1953             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1954             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1955
1956             }
1957
1958             /**************************
1959              * CALCULATE INTERACTIONS *
1960              **************************/
1961
1962             if (gmx_mm_any_lt(rsq12,rcutoff2))
1963             {
1964
1965             /* REACTION-FIELD ELECTROSTATICS */
1966             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1967
1968             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1969
1970             fscal            = felec;
1971
1972             fscal            = _mm_and_ps(fscal,cutoff_mask);
1973
1974             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1975
1976              /* Update vectorial force */
1977             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1978             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1979             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1980
1981             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1982             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1983             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1984
1985             }
1986
1987             /**************************
1988              * CALCULATE INTERACTIONS *
1989              **************************/
1990
1991             if (gmx_mm_any_lt(rsq20,rcutoff2))
1992             {
1993
1994             /* REACTION-FIELD ELECTROSTATICS */
1995             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1996
1997             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1998
1999             fscal            = felec;
2000
2001             fscal            = _mm_and_ps(fscal,cutoff_mask);
2002
2003             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2004
2005              /* Update vectorial force */
2006             fix2             = _mm_macc_ps(dx20,fscal,fix2);
2007             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
2008             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
2009
2010             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
2011             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
2012             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
2013
2014             }
2015
2016             /**************************
2017              * CALCULATE INTERACTIONS *
2018              **************************/
2019
2020             if (gmx_mm_any_lt(rsq21,rcutoff2))
2021             {
2022
2023             /* REACTION-FIELD ELECTROSTATICS */
2024             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
2025
2026             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2027
2028             fscal            = felec;
2029
2030             fscal            = _mm_and_ps(fscal,cutoff_mask);
2031
2032             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2033
2034              /* Update vectorial force */
2035             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2036             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2037             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2038
2039             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2040             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2041             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2042
2043             }
2044
2045             /**************************
2046              * CALCULATE INTERACTIONS *
2047              **************************/
2048
2049             if (gmx_mm_any_lt(rsq22,rcutoff2))
2050             {
2051
2052             /* REACTION-FIELD ELECTROSTATICS */
2053             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
2054
2055             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2056
2057             fscal            = felec;
2058
2059             fscal            = _mm_and_ps(fscal,cutoff_mask);
2060
2061             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2062
2063              /* Update vectorial force */
2064             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2065             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2066             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2067
2068             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2069             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2070             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2071
2072             }
2073
2074             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2075             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2076             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2077             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2078
2079             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2080                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2081
2082             /* Inner loop uses 325 flops */
2083         }
2084
2085         /* End of innermost loop */
2086
2087         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2088                                               f+i_coord_offset,fshift+i_shift_offset);
2089
2090         /* Increment number of inner iterations */
2091         inneriter                  += j_index_end - j_index_start;
2092
2093         /* Outer loop uses 18 flops */
2094     }
2095
2096     /* Increment number of outer iterations */
2097     outeriter        += nri;
2098
2099     /* Update outer/inner flops */
2100
2101     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*325);
2102 }