src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW4W4_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_VdwLJSw_GeomW4W4_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_GeomW4W4_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  94     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  95     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  96     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  97     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  98     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  99     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 100     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 101     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 102     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 103     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 104     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 105     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 106     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 107     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 108     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 109     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     int              nvdwtype;
 112     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 113     int              *vdwtype;
 114     real             *vdwparam;
 115     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 116     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 117     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 118     real             rswitch_scalar,d_scalar;
 119     __m128           dummy_mask,cutoff_mask;
 120     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 121     __m128           one     = _mm_set1_ps(1.0);
 122     __m128           two     = _mm_set1_ps(2.0);
 123     x                = xx[0];
 124     f                = ff[0];
 125
 126     nri              = nlist->nri;
 127     iinr             = nlist->iinr;
 128     jindex           = nlist->jindex;
 129     jjnr             = nlist->jjnr;
 130     shiftidx         = nlist->shift;
 131     gid              = nlist->gid;
 132     shiftvec         = fr->shift_vec[0];
 133     fshift           = fr->fshift[0];
 134     facel            = _mm_set1_ps(fr->epsfac);
 135     charge           = mdatoms->chargeA;
 136     krf              = _mm_set1_ps(fr->ic->k_rf);
 137     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 138     crf              = _mm_set1_ps(fr->ic->c_rf);
 139     nvdwtype         = fr->ntype;
 140     vdwparam         = fr->nbfp;
 141     vdwtype          = mdatoms->typeA;
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 146     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 147     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 148     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150     jq1              = _mm_set1_ps(charge[inr+1]);
 151     jq2              = _mm_set1_ps(charge[inr+2]);
 152     jq3              = _mm_set1_ps(charge[inr+3]);
 153     vdwjidx0A        = 2*vdwtype[inr+0];
 154     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 155     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 156     qq11             = _mm_mul_ps(iq1,jq1);
 157     qq12             = _mm_mul_ps(iq1,jq2);
 158     qq13             = _mm_mul_ps(iq1,jq3);
 159     qq21             = _mm_mul_ps(iq2,jq1);
 160     qq22             = _mm_mul_ps(iq2,jq2);
 161     qq23             = _mm_mul_ps(iq2,jq3);
 162     qq31             = _mm_mul_ps(iq3,jq1);
 163     qq32             = _mm_mul_ps(iq3,jq2);
 164     qq33             = _mm_mul_ps(iq3,jq3);
 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     rswitch_scalar   = fr->rvdw_switch;
 172     rswitch          = _mm_set1_ps(rswitch_scalar);
 173     /* Setup switch parameters */
 174     d_scalar         = rcutoff_scalar-rswitch_scalar;
 175     d                = _mm_set1_ps(d_scalar);
 176     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 177     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 178     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 179     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 180     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 181     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 182
 183     /* Avoid stupid compiler warnings */
 184     jnrA = jnrB = jnrC = jnrD = 0;
 185     j_coord_offsetA = 0;
 186     j_coord_offsetB = 0;
 187     j_coord_offsetC = 0;
 188     j_coord_offsetD = 0;
 189
 190     outeriter        = 0;
 191     inneriter        = 0;
 192
 193     for(iidx=0;iidx<4*DIM;iidx++)
 194     {
 195         scratch[iidx] = 0.0;
 196     }
 197
 198     /* Start outer loop over neighborlists */
 199     for(iidx=0; iidx<nri; iidx++)
 200     {
 201         /* Load shift vector for this list */
 202         i_shift_offset   = DIM*shiftidx[iidx];
 203
 204         /* Load limits for loop over neighbors */
 205         j_index_start    = jindex[iidx];
 206         j_index_end      = jindex[iidx+1];
 207
 208         /* Get outer coordinate index */
 209         inr              = iinr[iidx];
 210         i_coord_offset   = DIM*inr;
 211
 212         /* Load i particle coords and add shift vector */
 213         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 214                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 215
 216         fix0             = _mm_setzero_ps();
 217         fiy0             = _mm_setzero_ps();
 218         fiz0             = _mm_setzero_ps();
 219         fix1             = _mm_setzero_ps();
 220         fiy1             = _mm_setzero_ps();
 221         fiz1             = _mm_setzero_ps();
 222         fix2             = _mm_setzero_ps();
 223         fiy2             = _mm_setzero_ps();
 224         fiz2             = _mm_setzero_ps();
 225         fix3             = _mm_setzero_ps();
 226         fiy3             = _mm_setzero_ps();
 227         fiz3             = _mm_setzero_ps();
 228
 229         /* Reset potential sums */
 230         velecsum         = _mm_setzero_ps();
 231         vvdwsum          = _mm_setzero_ps();
 232
 233         /* Start inner kernel loop */
 234         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 235         {
 236
 237             /* Get j neighbor index, and coordinate index */
 238             jnrA             = jjnr[jidx];
 239             jnrB             = jjnr[jidx+1];
 240             jnrC             = jjnr[jidx+2];
 241             jnrD             = jjnr[jidx+3];
 242             j_coord_offsetA  = DIM*jnrA;
 243             j_coord_offsetB  = DIM*jnrB;
 244             j_coord_offsetC  = DIM*jnrC;
 245             j_coord_offsetD  = DIM*jnrD;
 246
 247             /* load j atom coordinates */
 248             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 249                                               x+j_coord_offsetC,x+j_coord_offsetD,
 250                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 251                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 252
 253             /* Calculate displacement vector */
 254             dx00             = _mm_sub_ps(ix0,jx0);
 255             dy00             = _mm_sub_ps(iy0,jy0);
 256             dz00             = _mm_sub_ps(iz0,jz0);
 257             dx11             = _mm_sub_ps(ix1,jx1);
 258             dy11             = _mm_sub_ps(iy1,jy1);
 259             dz11             = _mm_sub_ps(iz1,jz1);
 260             dx12             = _mm_sub_ps(ix1,jx2);
 261             dy12             = _mm_sub_ps(iy1,jy2);
 262             dz12             = _mm_sub_ps(iz1,jz2);
 263             dx13             = _mm_sub_ps(ix1,jx3);
 264             dy13             = _mm_sub_ps(iy1,jy3);
 265             dz13             = _mm_sub_ps(iz1,jz3);
 266             dx21             = _mm_sub_ps(ix2,jx1);
 267             dy21             = _mm_sub_ps(iy2,jy1);
 268             dz21             = _mm_sub_ps(iz2,jz1);
 269             dx22             = _mm_sub_ps(ix2,jx2);
 270             dy22             = _mm_sub_ps(iy2,jy2);
 271             dz22             = _mm_sub_ps(iz2,jz2);
 272             dx23             = _mm_sub_ps(ix2,jx3);
 273             dy23             = _mm_sub_ps(iy2,jy3);
 274             dz23             = _mm_sub_ps(iz2,jz3);
 275             dx31             = _mm_sub_ps(ix3,jx1);
 276             dy31             = _mm_sub_ps(iy3,jy1);
 277             dz31             = _mm_sub_ps(iz3,jz1);
 278             dx32             = _mm_sub_ps(ix3,jx2);
 279             dy32             = _mm_sub_ps(iy3,jy2);
 280             dz32             = _mm_sub_ps(iz3,jz2);
 281             dx33             = _mm_sub_ps(ix3,jx3);
 282             dy33             = _mm_sub_ps(iy3,jy3);
 283             dz33             = _mm_sub_ps(iz3,jz3);
 284
 285             /* Calculate squared distance and things based on it */
 286             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 287             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 288             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 289             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 290             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 291             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 292             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 293             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 294             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 295             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 296
 297             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 298             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 299             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 300             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 301             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 302             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 303             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 304             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 305             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 306             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 307
 308             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 309             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 310             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 311             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 312             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 313             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 314             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 315             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 316             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 317             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 318
 319             fjx0             = _mm_setzero_ps();
 320             fjy0             = _mm_setzero_ps();
 321             fjz0             = _mm_setzero_ps();
 322             fjx1             = _mm_setzero_ps();
 323             fjy1             = _mm_setzero_ps();
 324             fjz1             = _mm_setzero_ps();
 325             fjx2             = _mm_setzero_ps();
 326             fjy2             = _mm_setzero_ps();
 327             fjz2             = _mm_setzero_ps();
 328             fjx3             = _mm_setzero_ps();
 329             fjy3             = _mm_setzero_ps();
 330             fjz3             = _mm_setzero_ps();
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             if (gmx_mm_any_lt(rsq00,rcutoff2))
 337             {
 338
 339             r00              = _mm_mul_ps(rsq00,rinv00);
 340
 341             /* LENNARD-JONES DISPERSION/REPULSION */
 342
 343             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 344             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 345             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 346             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 347             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 348
 349             d                = _mm_sub_ps(r00,rswitch);
 350             d                = _mm_max_ps(d,_mm_setzero_ps());
 351             d2               = _mm_mul_ps(d,d);
 352             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 353
 354             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 355
 356             /* Evaluate switch function */
 357             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 358             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 359             vvdw             = _mm_mul_ps(vvdw,sw);
 360             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 364             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 365
 366             fscal            = 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(rsq11,rcutoff2))
 386             {
 387
 388             /* REACTION-FIELD ELECTROSTATICS */
 389             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 390             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 391
 392             cutoff_mask      = _mm_cmplt_ps(rsq11,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             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 404             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 405             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 406
 407             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 408             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 409             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 410
 411             }
 412
 413             /**************************
 414              * CALCULATE INTERACTIONS *
 415              **************************/
 416
 417             if (gmx_mm_any_lt(rsq12,rcutoff2))
 418             {
 419
 420             /* REACTION-FIELD ELECTROSTATICS */
 421             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 422             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 423
 424             cutoff_mask      = _mm_cmplt_ps(rsq12,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             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 436             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 437             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 438
 439             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 440             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 441             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 442
 443             }
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             if (gmx_mm_any_lt(rsq13,rcutoff2))
 450             {
 451
 452             /* REACTION-FIELD ELECTROSTATICS */
 453             velec            = _mm_mul_ps(qq13,_mm_sub_ps(_mm_macc_ps(krf,rsq13,rinv13),crf));
 454             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
 455
 456             cutoff_mask      = _mm_cmplt_ps(rsq13,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(dx13,fscal,fix1);
 468             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 469             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 470
 471             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 472             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 473             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 474
 475             }
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             if (gmx_mm_any_lt(rsq21,rcutoff2))
 482             {
 483
 484             /* REACTION-FIELD ELECTROSTATICS */
 485             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 486             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 487
 488             cutoff_mask      = _mm_cmplt_ps(rsq21,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             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 500             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 501             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 502
 503             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 504             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 505             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 506
 507             }
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             if (gmx_mm_any_lt(rsq22,rcutoff2))
 514             {
 515
 516             /* REACTION-FIELD ELECTROSTATICS */
 517             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 518             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 519
 520             cutoff_mask      = _mm_cmplt_ps(rsq22,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             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 532             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 533             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 534
 535             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 536             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 537             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 538
 539             }
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             if (gmx_mm_any_lt(rsq23,rcutoff2))
 546             {
 547
 548             /* REACTION-FIELD ELECTROSTATICS */
 549             velec            = _mm_mul_ps(qq23,_mm_sub_ps(_mm_macc_ps(krf,rsq23,rinv23),crf));
 550             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
 551
 552             cutoff_mask      = _mm_cmplt_ps(rsq23,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(dx23,fscal,fix2);
 564             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 565             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 566
 567             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 568             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 569             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 570
 571             }
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             if (gmx_mm_any_lt(rsq31,rcutoff2))
 578             {
 579
 580             /* REACTION-FIELD ELECTROSTATICS */
 581             velec            = _mm_mul_ps(qq31,_mm_sub_ps(_mm_macc_ps(krf,rsq31,rinv31),crf));
 582             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
 583
 584             cutoff_mask      = _mm_cmplt_ps(rsq31,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             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 596             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 597             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 598
 599             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 600             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 601             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 602
 603             }
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             if (gmx_mm_any_lt(rsq32,rcutoff2))
 610             {
 611
 612             /* REACTION-FIELD ELECTROSTATICS */
 613             velec            = _mm_mul_ps(qq32,_mm_sub_ps(_mm_macc_ps(krf,rsq32,rinv32),crf));
 614             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
 615
 616             cutoff_mask      = _mm_cmplt_ps(rsq32,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             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 628             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 629             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 630
 631             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 632             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 633             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 634
 635             }
 636
 637             /**************************
 638              * CALCULATE INTERACTIONS *
 639              **************************/
 640
 641             if (gmx_mm_any_lt(rsq33,rcutoff2))
 642             {
 643
 644             /* REACTION-FIELD ELECTROSTATICS */
 645             velec            = _mm_mul_ps(qq33,_mm_sub_ps(_mm_macc_ps(krf,rsq33,rinv33),crf));
 646             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
 647
 648             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm_and_ps(velec,cutoff_mask);
 652             velecsum         = _mm_add_ps(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             fscal            = _mm_and_ps(fscal,cutoff_mask);
 657
 658              /* Update vectorial force */
 659             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 660             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 661             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 662
 663             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 664             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 665             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 666
 667             }
 668
 669             fjptrA             = f+j_coord_offsetA;
 670             fjptrB             = f+j_coord_offsetB;
 671             fjptrC             = f+j_coord_offsetC;
 672             fjptrD             = f+j_coord_offsetD;
 673
 674             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 675                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 676                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 677
 678             /* Inner loop uses 416 flops */
 679         }
 680
 681         if(jidx<j_index_end)
 682         {
 683
 684             /* Get j neighbor index, and coordinate index */
 685             jnrlistA         = jjnr[jidx];
 686             jnrlistB         = jjnr[jidx+1];
 687             jnrlistC         = jjnr[jidx+2];
 688             jnrlistD         = jjnr[jidx+3];
 689             /* Sign of each element will be negative for non-real atoms.
 690              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 691              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 692              */
 693             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 694             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 695             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 696             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 697             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 698             j_coord_offsetA  = DIM*jnrA;
 699             j_coord_offsetB  = DIM*jnrB;
 700             j_coord_offsetC  = DIM*jnrC;
 701             j_coord_offsetD  = DIM*jnrD;
 702
 703             /* load j atom coordinates */
 704             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 705                                               x+j_coord_offsetC,x+j_coord_offsetD,
 706                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 707                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 708
 709             /* Calculate displacement vector */
 710             dx00             = _mm_sub_ps(ix0,jx0);
 711             dy00             = _mm_sub_ps(iy0,jy0);
 712             dz00             = _mm_sub_ps(iz0,jz0);
 713             dx11             = _mm_sub_ps(ix1,jx1);
 714             dy11             = _mm_sub_ps(iy1,jy1);
 715             dz11             = _mm_sub_ps(iz1,jz1);
 716             dx12             = _mm_sub_ps(ix1,jx2);
 717             dy12             = _mm_sub_ps(iy1,jy2);
 718             dz12             = _mm_sub_ps(iz1,jz2);
 719             dx13             = _mm_sub_ps(ix1,jx3);
 720             dy13             = _mm_sub_ps(iy1,jy3);
 721             dz13             = _mm_sub_ps(iz1,jz3);
 722             dx21             = _mm_sub_ps(ix2,jx1);
 723             dy21             = _mm_sub_ps(iy2,jy1);
 724             dz21             = _mm_sub_ps(iz2,jz1);
 725             dx22             = _mm_sub_ps(ix2,jx2);
 726             dy22             = _mm_sub_ps(iy2,jy2);
 727             dz22             = _mm_sub_ps(iz2,jz2);
 728             dx23             = _mm_sub_ps(ix2,jx3);
 729             dy23             = _mm_sub_ps(iy2,jy3);
 730             dz23             = _mm_sub_ps(iz2,jz3);
 731             dx31             = _mm_sub_ps(ix3,jx1);
 732             dy31             = _mm_sub_ps(iy3,jy1);
 733             dz31             = _mm_sub_ps(iz3,jz1);
 734             dx32             = _mm_sub_ps(ix3,jx2);
 735             dy32             = _mm_sub_ps(iy3,jy2);
 736             dz32             = _mm_sub_ps(iz3,jz2);
 737             dx33             = _mm_sub_ps(ix3,jx3);
 738             dy33             = _mm_sub_ps(iy3,jy3);
 739             dz33             = _mm_sub_ps(iz3,jz3);
 740
 741             /* Calculate squared distance and things based on it */
 742             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 743             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 744             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 745             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 746             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 747             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 748             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 749             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 750             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 751             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 752
 753             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 754             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 755             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 756             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 757             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 758             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 759             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 760             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 761             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 762             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 763
 764             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 765             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 766             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 767             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 768             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 769             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 770             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 771             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 772             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 773             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 774
 775             fjx0             = _mm_setzero_ps();
 776             fjy0             = _mm_setzero_ps();
 777             fjz0             = _mm_setzero_ps();
 778             fjx1             = _mm_setzero_ps();
 779             fjy1             = _mm_setzero_ps();
 780             fjz1             = _mm_setzero_ps();
 781             fjx2             = _mm_setzero_ps();
 782             fjy2             = _mm_setzero_ps();
 783             fjz2             = _mm_setzero_ps();
 784             fjx3             = _mm_setzero_ps();
 785             fjy3             = _mm_setzero_ps();
 786             fjz3             = _mm_setzero_ps();
 787
 788             /**************************
 789              * CALCULATE INTERACTIONS *
 790              **************************/
 791
 792             if (gmx_mm_any_lt(rsq00,rcutoff2))
 793             {
 794
 795             r00              = _mm_mul_ps(rsq00,rinv00);
 796             r00              = _mm_andnot_ps(dummy_mask,r00);
 797
 798             /* LENNARD-JONES DISPERSION/REPULSION */
 799
 800             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 801             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 802             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 803             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 804             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 805
 806             d                = _mm_sub_ps(r00,rswitch);
 807             d                = _mm_max_ps(d,_mm_setzero_ps());
 808             d2               = _mm_mul_ps(d,d);
 809             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 810
 811             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 812
 813             /* Evaluate switch function */
 814             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 815             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 816             vvdw             = _mm_mul_ps(vvdw,sw);
 817             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 818
 819             /* Update potential sum for this i atom from the interaction with this j atom. */
 820             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 821             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 822             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 823
 824             fscal            = fvdw;
 825
 826             fscal            = _mm_and_ps(fscal,cutoff_mask);
 827
 828             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 829
 830              /* Update vectorial force */
 831             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 832             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 833             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 834
 835             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 836             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 837             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 838
 839             }
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             if (gmx_mm_any_lt(rsq11,rcutoff2))
 846             {
 847
 848             /* REACTION-FIELD ELECTROSTATICS */
 849             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 850             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 851
 852             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
 853
 854             /* Update potential sum for this i atom from the interaction with this j atom. */
 855             velec            = _mm_and_ps(velec,cutoff_mask);
 856             velec            = _mm_andnot_ps(dummy_mask,velec);
 857             velecsum         = _mm_add_ps(velecsum,velec);
 858
 859             fscal            = felec;
 860
 861             fscal            = _mm_and_ps(fscal,cutoff_mask);
 862
 863             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 864
 865              /* Update vectorial force */
 866             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 867             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 868             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 869
 870             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 871             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 872             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 873
 874             }
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             if (gmx_mm_any_lt(rsq12,rcutoff2))
 881             {
 882
 883             /* REACTION-FIELD ELECTROSTATICS */
 884             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 885             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 886
 887             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
 888
 889             /* Update potential sum for this i atom from the interaction with this j atom. */
 890             velec            = _mm_and_ps(velec,cutoff_mask);
 891             velec            = _mm_andnot_ps(dummy_mask,velec);
 892             velecsum         = _mm_add_ps(velecsum,velec);
 893
 894             fscal            = felec;
 895
 896             fscal            = _mm_and_ps(fscal,cutoff_mask);
 897
 898             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 899
 900              /* Update vectorial force */
 901             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 902             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 903             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 904
 905             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 906             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 907             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 908
 909             }
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             if (gmx_mm_any_lt(rsq13,rcutoff2))
 916             {
 917
 918             /* REACTION-FIELD ELECTROSTATICS */
 919             velec            = _mm_mul_ps(qq13,_mm_sub_ps(_mm_macc_ps(krf,rsq13,rinv13),crf));
 920             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
 921
 922             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
 923
 924             /* Update potential sum for this i atom from the interaction with this j atom. */
 925             velec            = _mm_and_ps(velec,cutoff_mask);
 926             velec            = _mm_andnot_ps(dummy_mask,velec);
 927             velecsum         = _mm_add_ps(velecsum,velec);
 928
 929             fscal            = felec;
 930
 931             fscal            = _mm_and_ps(fscal,cutoff_mask);
 932
 933             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 934
 935              /* Update vectorial force */
 936             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 937             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 938             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 939
 940             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 941             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 942             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 943
 944             }
 945
 946             /**************************
 947              * CALCULATE INTERACTIONS *
 948              **************************/
 949
 950             if (gmx_mm_any_lt(rsq21,rcutoff2))
 951             {
 952
 953             /* REACTION-FIELD ELECTROSTATICS */
 954             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 955             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 956
 957             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
 958
 959             /* Update potential sum for this i atom from the interaction with this j atom. */
 960             velec            = _mm_and_ps(velec,cutoff_mask);
 961             velec            = _mm_andnot_ps(dummy_mask,velec);
 962             velecsum         = _mm_add_ps(velecsum,velec);
 963
 964             fscal            = felec;
 965
 966             fscal            = _mm_and_ps(fscal,cutoff_mask);
 967
 968             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 969
 970              /* Update vectorial force */
 971             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 972             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 973             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 974
 975             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 976             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 977             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 978
 979             }
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             if (gmx_mm_any_lt(rsq22,rcutoff2))
 986             {
 987
 988             /* REACTION-FIELD ELECTROSTATICS */
 989             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 990             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 991
 992             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
 993
 994             /* Update potential sum for this i atom from the interaction with this j atom. */
 995             velec            = _mm_and_ps(velec,cutoff_mask);
 996             velec            = _mm_andnot_ps(dummy_mask,velec);
 997             velecsum         = _mm_add_ps(velecsum,velec);
 998
 999             fscal            = felec;
1000
1001             fscal            = _mm_and_ps(fscal,cutoff_mask);
1002
1003             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1004
1005              /* Update vectorial force */
1006             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1007             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1008             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1009
1010             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1011             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1012             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1013
1014             }
1015
1016             /**************************
1017              * CALCULATE INTERACTIONS *
1018              **************************/
1019
1020             if (gmx_mm_any_lt(rsq23,rcutoff2))
1021             {
1022
1023             /* REACTION-FIELD ELECTROSTATICS */
1024             velec            = _mm_mul_ps(qq23,_mm_sub_ps(_mm_macc_ps(krf,rsq23,rinv23),crf));
1025             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
1026
1027             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
1028
1029             /* Update potential sum for this i atom from the interaction with this j atom. */
1030             velec            = _mm_and_ps(velec,cutoff_mask);
1031             velec            = _mm_andnot_ps(dummy_mask,velec);
1032             velecsum         = _mm_add_ps(velecsum,velec);
1033
1034             fscal            = felec;
1035
1036             fscal            = _mm_and_ps(fscal,cutoff_mask);
1037
1038             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1039
1040              /* Update vectorial force */
1041             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1042             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1043             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1044
1045             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1046             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1047             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1048
1049             }
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             if (gmx_mm_any_lt(rsq31,rcutoff2))
1056             {
1057
1058             /* REACTION-FIELD ELECTROSTATICS */
1059             velec            = _mm_mul_ps(qq31,_mm_sub_ps(_mm_macc_ps(krf,rsq31,rinv31),crf));
1060             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
1061
1062             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
1063
1064             /* Update potential sum for this i atom from the interaction with this j atom. */
1065             velec            = _mm_and_ps(velec,cutoff_mask);
1066             velec            = _mm_andnot_ps(dummy_mask,velec);
1067             velecsum         = _mm_add_ps(velecsum,velec);
1068
1069             fscal            = felec;
1070
1071             fscal            = _mm_and_ps(fscal,cutoff_mask);
1072
1073             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1074
1075              /* Update vectorial force */
1076             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1077             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1078             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1079
1080             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1081             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1082             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1083
1084             }
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             if (gmx_mm_any_lt(rsq32,rcutoff2))
1091             {
1092
1093             /* REACTION-FIELD ELECTROSTATICS */
1094             velec            = _mm_mul_ps(qq32,_mm_sub_ps(_mm_macc_ps(krf,rsq32,rinv32),crf));
1095             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
1096
1097             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
1098
1099             /* Update potential sum for this i atom from the interaction with this j atom. */
1100             velec            = _mm_and_ps(velec,cutoff_mask);
1101             velec            = _mm_andnot_ps(dummy_mask,velec);
1102             velecsum         = _mm_add_ps(velecsum,velec);
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm_and_ps(fscal,cutoff_mask);
1107
1108             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1109
1110              /* Update vectorial force */
1111             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1112             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1113             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1114
1115             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1116             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1117             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1118
1119             }
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             if (gmx_mm_any_lt(rsq33,rcutoff2))
1126             {
1127
1128             /* REACTION-FIELD ELECTROSTATICS */
1129             velec            = _mm_mul_ps(qq33,_mm_sub_ps(_mm_macc_ps(krf,rsq33,rinv33),crf));
1130             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
1131
1132             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
1133
1134             /* Update potential sum for this i atom from the interaction with this j atom. */
1135             velec            = _mm_and_ps(velec,cutoff_mask);
1136             velec            = _mm_andnot_ps(dummy_mask,velec);
1137             velecsum         = _mm_add_ps(velecsum,velec);
1138
1139             fscal            = felec;
1140
1141             fscal            = _mm_and_ps(fscal,cutoff_mask);
1142
1143             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1144
1145              /* Update vectorial force */
1146             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1147             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1148             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1149
1150             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1151             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1152             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1153
1154             }
1155
1156             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1157             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1158             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1159             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1160
1161             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1162                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1163                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1164
1165             /* Inner loop uses 417 flops */
1166         }
1167
1168         /* End of innermost loop */
1169
1170         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1171                                               f+i_coord_offset,fshift+i_shift_offset);
1172
1173         ggid                        = gid[iidx];
1174         /* Update potential energies */
1175         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1176         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1177
1178         /* Increment number of inner iterations */
1179         inneriter                  += j_index_end - j_index_start;
1180
1181         /* Outer loop uses 26 flops */
1182     }
1183
1184     /* Increment number of outer iterations */
1185     outeriter        += nri;
1186
1187     /* Update outer/inner flops */
1188
1189     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*417);
1190 }
1191 /*
1192  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4W4_F_avx_128_fma_single
1193  * Electrostatics interaction: ReactionField
1194  * VdW interaction:            LennardJones
1195  * Geometry:                   Water4-Water4
1196  * Calculate force/pot:        Force
1197  */
1198 void
1199 nb_kernel_ElecRFCut_VdwLJSw_GeomW4W4_F_avx_128_fma_single
1200                     (t_nblist                    * gmx_restrict       nlist,
1201                      rvec                        * gmx_restrict          xx,
1202                      rvec                        * gmx_restrict          ff,
1203                      t_forcerec                  * gmx_restrict          fr,
1204                      t_mdatoms                   * gmx_restrict     mdatoms,
1205                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1206                      t_nrnb                      * gmx_restrict        nrnb)
1207 {
1208     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1209      * just 0 for non-waters.
1210      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1211      * jnr indices corresponding to data put in the four positions in the SIMD register.
1212      */
1213     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1214     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1215     int              jnrA,jnrB,jnrC,jnrD;
1216     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1217     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1218     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1219     real             rcutoff_scalar;
1220     real             *shiftvec,*fshift,*x,*f;
1221     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1222     real             scratch[4*DIM];
1223     __m128           fscal,rcutoff,rcutoff2,jidxall;
1224     int              vdwioffset0;
1225     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1226     int              vdwioffset1;
1227     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1228     int              vdwioffset2;
1229     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1230     int              vdwioffset3;
1231     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1232     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1233     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1234     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1235     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1236     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1237     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1238     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1239     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1240     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1241     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1242     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1243     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1244     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1245     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1246     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1247     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1248     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1249     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1250     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1251     real             *charge;
1252     int              nvdwtype;
1253     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1254     int              *vdwtype;
1255     real             *vdwparam;
1256     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1257     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1258     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1259     real             rswitch_scalar,d_scalar;
1260     __m128           dummy_mask,cutoff_mask;
1261     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1262     __m128           one     = _mm_set1_ps(1.0);
1263     __m128           two     = _mm_set1_ps(2.0);
1264     x                = xx[0];
1265     f                = ff[0];
1266
1267     nri              = nlist->nri;
1268     iinr             = nlist->iinr;
1269     jindex           = nlist->jindex;
1270     jjnr             = nlist->jjnr;
1271     shiftidx         = nlist->shift;
1272     gid              = nlist->gid;
1273     shiftvec         = fr->shift_vec[0];
1274     fshift           = fr->fshift[0];
1275     facel            = _mm_set1_ps(fr->epsfac);
1276     charge           = mdatoms->chargeA;
1277     krf              = _mm_set1_ps(fr->ic->k_rf);
1278     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
1279     crf              = _mm_set1_ps(fr->ic->c_rf);
1280     nvdwtype         = fr->ntype;
1281     vdwparam         = fr->nbfp;
1282     vdwtype          = mdatoms->typeA;
1283
1284     /* Setup water-specific parameters */
1285     inr              = nlist->iinr[0];
1286     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1287     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1288     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1289     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1290
1291     jq1              = _mm_set1_ps(charge[inr+1]);
1292     jq2              = _mm_set1_ps(charge[inr+2]);
1293     jq3              = _mm_set1_ps(charge[inr+3]);
1294     vdwjidx0A        = 2*vdwtype[inr+0];
1295     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1296     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1297     qq11             = _mm_mul_ps(iq1,jq1);
1298     qq12             = _mm_mul_ps(iq1,jq2);
1299     qq13             = _mm_mul_ps(iq1,jq3);
1300     qq21             = _mm_mul_ps(iq2,jq1);
1301     qq22             = _mm_mul_ps(iq2,jq2);
1302     qq23             = _mm_mul_ps(iq2,jq3);
1303     qq31             = _mm_mul_ps(iq3,jq1);
1304     qq32             = _mm_mul_ps(iq3,jq2);
1305     qq33             = _mm_mul_ps(iq3,jq3);
1306
1307     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1308     rcutoff_scalar   = fr->rcoulomb;
1309     rcutoff          = _mm_set1_ps(rcutoff_scalar);
1310     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
1311
1312     rswitch_scalar   = fr->rvdw_switch;
1313     rswitch          = _mm_set1_ps(rswitch_scalar);
1314     /* Setup switch parameters */
1315     d_scalar         = rcutoff_scalar-rswitch_scalar;
1316     d                = _mm_set1_ps(d_scalar);
1317     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
1318     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1319     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1320     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
1321     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1322     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1323
1324     /* Avoid stupid compiler warnings */
1325     jnrA = jnrB = jnrC = jnrD = 0;
1326     j_coord_offsetA = 0;
1327     j_coord_offsetB = 0;
1328     j_coord_offsetC = 0;
1329     j_coord_offsetD = 0;
1330
1331     outeriter        = 0;
1332     inneriter        = 0;
1333
1334     for(iidx=0;iidx<4*DIM;iidx++)
1335     {
1336         scratch[iidx] = 0.0;
1337     }
1338
1339     /* Start outer loop over neighborlists */
1340     for(iidx=0; iidx<nri; iidx++)
1341     {
1342         /* Load shift vector for this list */
1343         i_shift_offset   = DIM*shiftidx[iidx];
1344
1345         /* Load limits for loop over neighbors */
1346         j_index_start    = jindex[iidx];
1347         j_index_end      = jindex[iidx+1];
1348
1349         /* Get outer coordinate index */
1350         inr              = iinr[iidx];
1351         i_coord_offset   = DIM*inr;
1352
1353         /* Load i particle coords and add shift vector */
1354         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1355                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1356
1357         fix0             = _mm_setzero_ps();
1358         fiy0             = _mm_setzero_ps();
1359         fiz0             = _mm_setzero_ps();
1360         fix1             = _mm_setzero_ps();
1361         fiy1             = _mm_setzero_ps();
1362         fiz1             = _mm_setzero_ps();
1363         fix2             = _mm_setzero_ps();
1364         fiy2             = _mm_setzero_ps();
1365         fiz2             = _mm_setzero_ps();
1366         fix3             = _mm_setzero_ps();
1367         fiy3             = _mm_setzero_ps();
1368         fiz3             = _mm_setzero_ps();
1369
1370         /* Start inner kernel loop */
1371         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1372         {
1373
1374             /* Get j neighbor index, and coordinate index */
1375             jnrA             = jjnr[jidx];
1376             jnrB             = jjnr[jidx+1];
1377             jnrC             = jjnr[jidx+2];
1378             jnrD             = jjnr[jidx+3];
1379             j_coord_offsetA  = DIM*jnrA;
1380             j_coord_offsetB  = DIM*jnrB;
1381             j_coord_offsetC  = DIM*jnrC;
1382             j_coord_offsetD  = DIM*jnrD;
1383
1384             /* load j atom coordinates */
1385             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1386                                               x+j_coord_offsetC,x+j_coord_offsetD,
1387                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1388                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1389
1390             /* Calculate displacement vector */
1391             dx00             = _mm_sub_ps(ix0,jx0);
1392             dy00             = _mm_sub_ps(iy0,jy0);
1393             dz00             = _mm_sub_ps(iz0,jz0);
1394             dx11             = _mm_sub_ps(ix1,jx1);
1395             dy11             = _mm_sub_ps(iy1,jy1);
1396             dz11             = _mm_sub_ps(iz1,jz1);
1397             dx12             = _mm_sub_ps(ix1,jx2);
1398             dy12             = _mm_sub_ps(iy1,jy2);
1399             dz12             = _mm_sub_ps(iz1,jz2);
1400             dx13             = _mm_sub_ps(ix1,jx3);
1401             dy13             = _mm_sub_ps(iy1,jy3);
1402             dz13             = _mm_sub_ps(iz1,jz3);
1403             dx21             = _mm_sub_ps(ix2,jx1);
1404             dy21             = _mm_sub_ps(iy2,jy1);
1405             dz21             = _mm_sub_ps(iz2,jz1);
1406             dx22             = _mm_sub_ps(ix2,jx2);
1407             dy22             = _mm_sub_ps(iy2,jy2);
1408             dz22             = _mm_sub_ps(iz2,jz2);
1409             dx23             = _mm_sub_ps(ix2,jx3);
1410             dy23             = _mm_sub_ps(iy2,jy3);
1411             dz23             = _mm_sub_ps(iz2,jz3);
1412             dx31             = _mm_sub_ps(ix3,jx1);
1413             dy31             = _mm_sub_ps(iy3,jy1);
1414             dz31             = _mm_sub_ps(iz3,jz1);
1415             dx32             = _mm_sub_ps(ix3,jx2);
1416             dy32             = _mm_sub_ps(iy3,jy2);
1417             dz32             = _mm_sub_ps(iz3,jz2);
1418             dx33             = _mm_sub_ps(ix3,jx3);
1419             dy33             = _mm_sub_ps(iy3,jy3);
1420             dz33             = _mm_sub_ps(iz3,jz3);
1421
1422             /* Calculate squared distance and things based on it */
1423             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1424             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1425             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1426             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1427             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1428             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1429             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1430             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1431             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1432             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1433
1434             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1435             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1436             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1437             rinv13           = gmx_mm_invsqrt_ps(rsq13);
1438             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1439             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1440             rinv23           = gmx_mm_invsqrt_ps(rsq23);
1441             rinv31           = gmx_mm_invsqrt_ps(rsq31);
1442             rinv32           = gmx_mm_invsqrt_ps(rsq32);
1443             rinv33           = gmx_mm_invsqrt_ps(rsq33);
1444
1445             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1446             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1447             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1448             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1449             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1450             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1451             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1452             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1453             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1454             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1455
1456             fjx0             = _mm_setzero_ps();
1457             fjy0             = _mm_setzero_ps();
1458             fjz0             = _mm_setzero_ps();
1459             fjx1             = _mm_setzero_ps();
1460             fjy1             = _mm_setzero_ps();
1461             fjz1             = _mm_setzero_ps();
1462             fjx2             = _mm_setzero_ps();
1463             fjy2             = _mm_setzero_ps();
1464             fjz2             = _mm_setzero_ps();
1465             fjx3             = _mm_setzero_ps();
1466             fjy3             = _mm_setzero_ps();
1467             fjz3             = _mm_setzero_ps();
1468
1469             /**************************
1470              * CALCULATE INTERACTIONS *
1471              **************************/
1472
1473             if (gmx_mm_any_lt(rsq00,rcutoff2))
1474             {
1475
1476             r00              = _mm_mul_ps(rsq00,rinv00);
1477
1478             /* LENNARD-JONES DISPERSION/REPULSION */
1479
1480             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1481             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1482             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1483             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1484             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1485
1486             d                = _mm_sub_ps(r00,rswitch);
1487             d                = _mm_max_ps(d,_mm_setzero_ps());
1488             d2               = _mm_mul_ps(d,d);
1489             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1490
1491             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1492
1493             /* Evaluate switch function */
1494             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1495             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1496             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1497
1498             fscal            = fvdw;
1499
1500             fscal            = _mm_and_ps(fscal,cutoff_mask);
1501
1502              /* Update vectorial force */
1503             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1504             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1505             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1506
1507             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1508             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1509             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1510
1511             }
1512
1513             /**************************
1514              * CALCULATE INTERACTIONS *
1515              **************************/
1516
1517             if (gmx_mm_any_lt(rsq11,rcutoff2))
1518             {
1519
1520             /* REACTION-FIELD ELECTROSTATICS */
1521             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1522
1523             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1524
1525             fscal            = felec;
1526
1527             fscal            = _mm_and_ps(fscal,cutoff_mask);
1528
1529              /* Update vectorial force */
1530             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1531             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1532             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1533
1534             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1535             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1536             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1537
1538             }
1539
1540             /**************************
1541              * CALCULATE INTERACTIONS *
1542              **************************/
1543
1544             if (gmx_mm_any_lt(rsq12,rcutoff2))
1545             {
1546
1547             /* REACTION-FIELD ELECTROSTATICS */
1548             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1549
1550             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1551
1552             fscal            = felec;
1553
1554             fscal            = _mm_and_ps(fscal,cutoff_mask);
1555
1556              /* Update vectorial force */
1557             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1558             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1559             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1560
1561             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1562             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1563             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1564
1565             }
1566
1567             /**************************
1568              * CALCULATE INTERACTIONS *
1569              **************************/
1570
1571             if (gmx_mm_any_lt(rsq13,rcutoff2))
1572             {
1573
1574             /* REACTION-FIELD ELECTROSTATICS */
1575             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
1576
1577             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
1578
1579             fscal            = felec;
1580
1581             fscal            = _mm_and_ps(fscal,cutoff_mask);
1582
1583              /* Update vectorial force */
1584             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1585             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1586             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1587
1588             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1589             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1590             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1591
1592             }
1593
1594             /**************************
1595              * CALCULATE INTERACTIONS *
1596              **************************/
1597
1598             if (gmx_mm_any_lt(rsq21,rcutoff2))
1599             {
1600
1601             /* REACTION-FIELD ELECTROSTATICS */
1602             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1603
1604             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1605
1606             fscal            = felec;
1607
1608             fscal            = _mm_and_ps(fscal,cutoff_mask);
1609
1610              /* Update vectorial force */
1611             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1612             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1613             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1614
1615             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1616             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1617             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1618
1619             }
1620
1621             /**************************
1622              * CALCULATE INTERACTIONS *
1623              **************************/
1624
1625             if (gmx_mm_any_lt(rsq22,rcutoff2))
1626             {
1627
1628             /* REACTION-FIELD ELECTROSTATICS */
1629             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1630
1631             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
1632
1633             fscal            = felec;
1634
1635             fscal            = _mm_and_ps(fscal,cutoff_mask);
1636
1637              /* Update vectorial force */
1638             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1639             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1640             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1641
1642             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1643             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1644             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1645
1646             }
1647
1648             /**************************
1649              * CALCULATE INTERACTIONS *
1650              **************************/
1651
1652             if (gmx_mm_any_lt(rsq23,rcutoff2))
1653             {
1654
1655             /* REACTION-FIELD ELECTROSTATICS */
1656             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
1657
1658             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
1659
1660             fscal            = felec;
1661
1662             fscal            = _mm_and_ps(fscal,cutoff_mask);
1663
1664              /* Update vectorial force */
1665             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1666             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1667             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1668
1669             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1670             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1671             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1672
1673             }
1674
1675             /**************************
1676              * CALCULATE INTERACTIONS *
1677              **************************/
1678
1679             if (gmx_mm_any_lt(rsq31,rcutoff2))
1680             {
1681
1682             /* REACTION-FIELD ELECTROSTATICS */
1683             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
1684
1685             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
1686
1687             fscal            = felec;
1688
1689             fscal            = _mm_and_ps(fscal,cutoff_mask);
1690
1691              /* Update vectorial force */
1692             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1693             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1694             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1695
1696             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1697             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1698             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1699
1700             }
1701
1702             /**************************
1703              * CALCULATE INTERACTIONS *
1704              **************************/
1705
1706             if (gmx_mm_any_lt(rsq32,rcutoff2))
1707             {
1708
1709             /* REACTION-FIELD ELECTROSTATICS */
1710             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
1711
1712             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
1713
1714             fscal            = felec;
1715
1716             fscal            = _mm_and_ps(fscal,cutoff_mask);
1717
1718              /* Update vectorial force */
1719             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1720             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1721             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1722
1723             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1724             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1725             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1726
1727             }
1728
1729             /**************************
1730              * CALCULATE INTERACTIONS *
1731              **************************/
1732
1733             if (gmx_mm_any_lt(rsq33,rcutoff2))
1734             {
1735
1736             /* REACTION-FIELD ELECTROSTATICS */
1737             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
1738
1739             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
1740
1741             fscal            = felec;
1742
1743             fscal            = _mm_and_ps(fscal,cutoff_mask);
1744
1745              /* Update vectorial force */
1746             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1747             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1748             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1749
1750             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1751             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1752             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1753
1754             }
1755
1756             fjptrA             = f+j_coord_offsetA;
1757             fjptrB             = f+j_coord_offsetB;
1758             fjptrC             = f+j_coord_offsetC;
1759             fjptrD             = f+j_coord_offsetD;
1760
1761             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1762                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1763                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1764
1765             /* Inner loop uses 359 flops */
1766         }
1767
1768         if(jidx<j_index_end)
1769         {
1770
1771             /* Get j neighbor index, and coordinate index */
1772             jnrlistA         = jjnr[jidx];
1773             jnrlistB         = jjnr[jidx+1];
1774             jnrlistC         = jjnr[jidx+2];
1775             jnrlistD         = jjnr[jidx+3];
1776             /* Sign of each element will be negative for non-real atoms.
1777              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1778              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1779              */
1780             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1781             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1782             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1783             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1784             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1785             j_coord_offsetA  = DIM*jnrA;
1786             j_coord_offsetB  = DIM*jnrB;
1787             j_coord_offsetC  = DIM*jnrC;
1788             j_coord_offsetD  = DIM*jnrD;
1789
1790             /* load j atom coordinates */
1791             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1792                                               x+j_coord_offsetC,x+j_coord_offsetD,
1793                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1794                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1795
1796             /* Calculate displacement vector */
1797             dx00             = _mm_sub_ps(ix0,jx0);
1798             dy00             = _mm_sub_ps(iy0,jy0);
1799             dz00             = _mm_sub_ps(iz0,jz0);
1800             dx11             = _mm_sub_ps(ix1,jx1);
1801             dy11             = _mm_sub_ps(iy1,jy1);
1802             dz11             = _mm_sub_ps(iz1,jz1);
1803             dx12             = _mm_sub_ps(ix1,jx2);
1804             dy12             = _mm_sub_ps(iy1,jy2);
1805             dz12             = _mm_sub_ps(iz1,jz2);
1806             dx13             = _mm_sub_ps(ix1,jx3);
1807             dy13             = _mm_sub_ps(iy1,jy3);
1808             dz13             = _mm_sub_ps(iz1,jz3);
1809             dx21             = _mm_sub_ps(ix2,jx1);
1810             dy21             = _mm_sub_ps(iy2,jy1);
1811             dz21             = _mm_sub_ps(iz2,jz1);
1812             dx22             = _mm_sub_ps(ix2,jx2);
1813             dy22             = _mm_sub_ps(iy2,jy2);
1814             dz22             = _mm_sub_ps(iz2,jz2);
1815             dx23             = _mm_sub_ps(ix2,jx3);
1816             dy23             = _mm_sub_ps(iy2,jy3);
1817             dz23             = _mm_sub_ps(iz2,jz3);
1818             dx31             = _mm_sub_ps(ix3,jx1);
1819             dy31             = _mm_sub_ps(iy3,jy1);
1820             dz31             = _mm_sub_ps(iz3,jz1);
1821             dx32             = _mm_sub_ps(ix3,jx2);
1822             dy32             = _mm_sub_ps(iy3,jy2);
1823             dz32             = _mm_sub_ps(iz3,jz2);
1824             dx33             = _mm_sub_ps(ix3,jx3);
1825             dy33             = _mm_sub_ps(iy3,jy3);
1826             dz33             = _mm_sub_ps(iz3,jz3);
1827
1828             /* Calculate squared distance and things based on it */
1829             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1830             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1831             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1832             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1833             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1834             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1835             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1836             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1837             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1838             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1839
1840             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1841             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1842             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1843             rinv13           = gmx_mm_invsqrt_ps(rsq13);
1844             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1845             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1846             rinv23           = gmx_mm_invsqrt_ps(rsq23);
1847             rinv31           = gmx_mm_invsqrt_ps(rsq31);
1848             rinv32           = gmx_mm_invsqrt_ps(rsq32);
1849             rinv33           = gmx_mm_invsqrt_ps(rsq33);
1850
1851             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1852             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1853             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1854             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1855             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1856             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1857             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1858             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1859             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1860             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1861
1862             fjx0             = _mm_setzero_ps();
1863             fjy0             = _mm_setzero_ps();
1864             fjz0             = _mm_setzero_ps();
1865             fjx1             = _mm_setzero_ps();
1866             fjy1             = _mm_setzero_ps();
1867             fjz1             = _mm_setzero_ps();
1868             fjx2             = _mm_setzero_ps();
1869             fjy2             = _mm_setzero_ps();
1870             fjz2             = _mm_setzero_ps();
1871             fjx3             = _mm_setzero_ps();
1872             fjy3             = _mm_setzero_ps();
1873             fjz3             = _mm_setzero_ps();
1874
1875             /**************************
1876              * CALCULATE INTERACTIONS *
1877              **************************/
1878
1879             if (gmx_mm_any_lt(rsq00,rcutoff2))
1880             {
1881
1882             r00              = _mm_mul_ps(rsq00,rinv00);
1883             r00              = _mm_andnot_ps(dummy_mask,r00);
1884
1885             /* LENNARD-JONES DISPERSION/REPULSION */
1886
1887             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1888             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1889             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1890             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1891             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1892
1893             d                = _mm_sub_ps(r00,rswitch);
1894             d                = _mm_max_ps(d,_mm_setzero_ps());
1895             d2               = _mm_mul_ps(d,d);
1896             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1897
1898             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1899
1900             /* Evaluate switch function */
1901             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1902             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1903             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1904
1905             fscal            = fvdw;
1906
1907             fscal            = _mm_and_ps(fscal,cutoff_mask);
1908
1909             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1910
1911              /* Update vectorial force */
1912             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1913             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1914             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1915
1916             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1917             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1918             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1919
1920             }
1921
1922             /**************************
1923              * CALCULATE INTERACTIONS *
1924              **************************/
1925
1926             if (gmx_mm_any_lt(rsq11,rcutoff2))
1927             {
1928
1929             /* REACTION-FIELD ELECTROSTATICS */
1930             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1931
1932             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1933
1934             fscal            = felec;
1935
1936             fscal            = _mm_and_ps(fscal,cutoff_mask);
1937
1938             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1939
1940              /* Update vectorial force */
1941             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1942             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1943             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1944
1945             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1946             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1947             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1948
1949             }
1950
1951             /**************************
1952              * CALCULATE INTERACTIONS *
1953              **************************/
1954
1955             if (gmx_mm_any_lt(rsq12,rcutoff2))
1956             {
1957
1958             /* REACTION-FIELD ELECTROSTATICS */
1959             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1960
1961             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1962
1963             fscal            = felec;
1964
1965             fscal            = _mm_and_ps(fscal,cutoff_mask);
1966
1967             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1968
1969              /* Update vectorial force */
1970             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1971             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1972             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1973
1974             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1975             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1976             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1977
1978             }
1979
1980             /**************************
1981              * CALCULATE INTERACTIONS *
1982              **************************/
1983
1984             if (gmx_mm_any_lt(rsq13,rcutoff2))
1985             {
1986
1987             /* REACTION-FIELD ELECTROSTATICS */
1988             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
1989
1990             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
1991
1992             fscal            = felec;
1993
1994             fscal            = _mm_and_ps(fscal,cutoff_mask);
1995
1996             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1997
1998              /* Update vectorial force */
1999             fix1             = _mm_macc_ps(dx13,fscal,fix1);
2000             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
2001             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
2002
2003             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
2004             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
2005             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
2006
2007             }
2008
2009             /**************************
2010              * CALCULATE INTERACTIONS *
2011              **************************/
2012
2013             if (gmx_mm_any_lt(rsq21,rcutoff2))
2014             {
2015
2016             /* REACTION-FIELD ELECTROSTATICS */
2017             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
2018
2019             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2020
2021             fscal            = felec;
2022
2023             fscal            = _mm_and_ps(fscal,cutoff_mask);
2024
2025             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2026
2027              /* Update vectorial force */
2028             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2029             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2030             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2031
2032             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2033             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2034             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2035
2036             }
2037
2038             /**************************
2039              * CALCULATE INTERACTIONS *
2040              **************************/
2041
2042             if (gmx_mm_any_lt(rsq22,rcutoff2))
2043             {
2044
2045             /* REACTION-FIELD ELECTROSTATICS */
2046             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
2047
2048             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2049
2050             fscal            = felec;
2051
2052             fscal            = _mm_and_ps(fscal,cutoff_mask);
2053
2054             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2055
2056              /* Update vectorial force */
2057             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2058             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2059             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2060
2061             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2062             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2063             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2064
2065             }
2066
2067             /**************************
2068              * CALCULATE INTERACTIONS *
2069              **************************/
2070
2071             if (gmx_mm_any_lt(rsq23,rcutoff2))
2072             {
2073
2074             /* REACTION-FIELD ELECTROSTATICS */
2075             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
2076
2077             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
2078
2079             fscal            = felec;
2080
2081             fscal            = _mm_and_ps(fscal,cutoff_mask);
2082
2083             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2084
2085              /* Update vectorial force */
2086             fix2             = _mm_macc_ps(dx23,fscal,fix2);
2087             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
2088             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
2089
2090             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
2091             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
2092             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
2093
2094             }
2095
2096             /**************************
2097              * CALCULATE INTERACTIONS *
2098              **************************/
2099
2100             if (gmx_mm_any_lt(rsq31,rcutoff2))
2101             {
2102
2103             /* REACTION-FIELD ELECTROSTATICS */
2104             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
2105
2106             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
2107
2108             fscal            = felec;
2109
2110             fscal            = _mm_and_ps(fscal,cutoff_mask);
2111
2112             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2113
2114              /* Update vectorial force */
2115             fix3             = _mm_macc_ps(dx31,fscal,fix3);
2116             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
2117             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
2118
2119             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
2120             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
2121             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
2122
2123             }
2124
2125             /**************************
2126              * CALCULATE INTERACTIONS *
2127              **************************/
2128
2129             if (gmx_mm_any_lt(rsq32,rcutoff2))
2130             {
2131
2132             /* REACTION-FIELD ELECTROSTATICS */
2133             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
2134
2135             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
2136
2137             fscal            = felec;
2138
2139             fscal            = _mm_and_ps(fscal,cutoff_mask);
2140
2141             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2142
2143              /* Update vectorial force */
2144             fix3             = _mm_macc_ps(dx32,fscal,fix3);
2145             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
2146             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
2147
2148             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
2149             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
2150             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
2151
2152             }
2153
2154             /**************************
2155              * CALCULATE INTERACTIONS *
2156              **************************/
2157
2158             if (gmx_mm_any_lt(rsq33,rcutoff2))
2159             {
2160
2161             /* REACTION-FIELD ELECTROSTATICS */
2162             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
2163
2164             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
2165
2166             fscal            = felec;
2167
2168             fscal            = _mm_and_ps(fscal,cutoff_mask);
2169
2170             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2171
2172              /* Update vectorial force */
2173             fix3             = _mm_macc_ps(dx33,fscal,fix3);
2174             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
2175             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
2176
2177             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
2178             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
2179             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
2180
2181             }
2182
2183             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2184             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2185             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2186             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2187
2188             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2189                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2190                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2191
2192             /* Inner loop uses 360 flops */
2193         }
2194
2195         /* End of innermost loop */
2196
2197         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2198                                               f+i_coord_offset,fshift+i_shift_offset);
2199
2200         /* Increment number of inner iterations */
2201         inneriter                  += j_index_end - j_index_start;
2202
2203         /* Outer loop uses 24 flops */
2204     }
2205
2206     /* Increment number of outer iterations */
2207     outeriter        += nri;
2208
2209     /* Update outer/inner flops */
2210
2211     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*360);
2212 }