src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_VdwCSTab_GeomW3W3_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3W3_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW3W3_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     int              vdwjidx1A,vdwjidx1B;
  91     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  92     int              vdwjidx2A,vdwjidx2B;
  93     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  94     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  96     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  97     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 104     real             *charge;
 105     int              nvdwtype;
 106     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 107     int              *vdwtype;
 108     real             *vdwparam;
 109     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 110     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 111     __m128i          vfitab;
 112     __m128i          ifour       = _mm_set1_epi32(4);
 113     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 114     real             *vftab;
 115     __m128d          dummy_mask,cutoff_mask;
 116     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 117     __m128d          one     = _mm_set1_pd(1.0);
 118     __m128d          two     = _mm_set1_pd(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_pd(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     krf              = _mm_set1_pd(fr->ic->k_rf);
 133     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 134     crf              = _mm_set1_pd(fr->ic->c_rf);
 135     nvdwtype         = fr->ntype;
 136     vdwparam         = fr->nbfp;
 137     vdwtype          = mdatoms->typeA;
 138
 139     vftab            = kernel_data->table_vdw->data;
 140     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 145     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 146     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 147     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149     jq0              = _mm_set1_pd(charge[inr+0]);
 150     jq1              = _mm_set1_pd(charge[inr+1]);
 151     jq2              = _mm_set1_pd(charge[inr+2]);
 152     vdwjidx0A        = 2*vdwtype[inr+0];
 153     qq00             = _mm_mul_pd(iq0,jq0);
 154     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 155     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 156     qq01             = _mm_mul_pd(iq0,jq1);
 157     qq02             = _mm_mul_pd(iq0,jq2);
 158     qq10             = _mm_mul_pd(iq1,jq0);
 159     qq11             = _mm_mul_pd(iq1,jq1);
 160     qq12             = _mm_mul_pd(iq1,jq2);
 161     qq20             = _mm_mul_pd(iq2,jq0);
 162     qq21             = _mm_mul_pd(iq2,jq1);
 163     qq22             = _mm_mul_pd(iq2,jq2);
 164
 165     /* Avoid stupid compiler warnings */
 166     jnrA = jnrB = 0;
 167     j_coord_offsetA = 0;
 168     j_coord_offsetB = 0;
 169
 170     outeriter        = 0;
 171     inneriter        = 0;
 172
 173     /* Start outer loop over neighborlists */
 174     for(iidx=0; iidx<nri; iidx++)
 175     {
 176         /* Load shift vector for this list */
 177         i_shift_offset   = DIM*shiftidx[iidx];
 178
 179         /* Load limits for loop over neighbors */
 180         j_index_start    = jindex[iidx];
 181         j_index_end      = jindex[iidx+1];
 182
 183         /* Get outer coordinate index */
 184         inr              = iinr[iidx];
 185         i_coord_offset   = DIM*inr;
 186
 187         /* Load i particle coords and add shift vector */
 188         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 189                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 190
 191         fix0             = _mm_setzero_pd();
 192         fiy0             = _mm_setzero_pd();
 193         fiz0             = _mm_setzero_pd();
 194         fix1             = _mm_setzero_pd();
 195         fiy1             = _mm_setzero_pd();
 196         fiz1             = _mm_setzero_pd();
 197         fix2             = _mm_setzero_pd();
 198         fiy2             = _mm_setzero_pd();
 199         fiz2             = _mm_setzero_pd();
 200
 201         /* Reset potential sums */
 202         velecsum         = _mm_setzero_pd();
 203         vvdwsum          = _mm_setzero_pd();
 204
 205         /* Start inner kernel loop */
 206         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 207         {
 208
 209             /* Get j neighbor index, and coordinate index */
 210             jnrA             = jjnr[jidx];
 211             jnrB             = jjnr[jidx+1];
 212             j_coord_offsetA  = DIM*jnrA;
 213             j_coord_offsetB  = DIM*jnrB;
 214
 215             /* load j atom coordinates */
 216             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 217                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm_sub_pd(ix0,jx0);
 221             dy00             = _mm_sub_pd(iy0,jy0);
 222             dz00             = _mm_sub_pd(iz0,jz0);
 223             dx01             = _mm_sub_pd(ix0,jx1);
 224             dy01             = _mm_sub_pd(iy0,jy1);
 225             dz01             = _mm_sub_pd(iz0,jz1);
 226             dx02             = _mm_sub_pd(ix0,jx2);
 227             dy02             = _mm_sub_pd(iy0,jy2);
 228             dz02             = _mm_sub_pd(iz0,jz2);
 229             dx10             = _mm_sub_pd(ix1,jx0);
 230             dy10             = _mm_sub_pd(iy1,jy0);
 231             dz10             = _mm_sub_pd(iz1,jz0);
 232             dx11             = _mm_sub_pd(ix1,jx1);
 233             dy11             = _mm_sub_pd(iy1,jy1);
 234             dz11             = _mm_sub_pd(iz1,jz1);
 235             dx12             = _mm_sub_pd(ix1,jx2);
 236             dy12             = _mm_sub_pd(iy1,jy2);
 237             dz12             = _mm_sub_pd(iz1,jz2);
 238             dx20             = _mm_sub_pd(ix2,jx0);
 239             dy20             = _mm_sub_pd(iy2,jy0);
 240             dz20             = _mm_sub_pd(iz2,jz0);
 241             dx21             = _mm_sub_pd(ix2,jx1);
 242             dy21             = _mm_sub_pd(iy2,jy1);
 243             dz21             = _mm_sub_pd(iz2,jz1);
 244             dx22             = _mm_sub_pd(ix2,jx2);
 245             dy22             = _mm_sub_pd(iy2,jy2);
 246             dz22             = _mm_sub_pd(iz2,jz2);
 247
 248             /* Calculate squared distance and things based on it */
 249             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 250             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 251             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 252             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 253             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 254             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 255             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 256             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 257             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 258
 259             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 260             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 261             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 262             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 263             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 264             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 265             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 266             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 267             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 268
 269             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 270             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 271             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 272             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 273             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 274             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 275             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 276             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 277             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 278
 279             fjx0             = _mm_setzero_pd();
 280             fjy0             = _mm_setzero_pd();
 281             fjz0             = _mm_setzero_pd();
 282             fjx1             = _mm_setzero_pd();
 283             fjy1             = _mm_setzero_pd();
 284             fjz1             = _mm_setzero_pd();
 285             fjx2             = _mm_setzero_pd();
 286             fjy2             = _mm_setzero_pd();
 287             fjz2             = _mm_setzero_pd();
 288
 289             /**************************
 290              * CALCULATE INTERACTIONS *
 291              **************************/
 292
 293             r00              = _mm_mul_pd(rsq00,rinv00);
 294
 295             /* Calculate table index by multiplying r with table scale and truncate to integer */
 296             rt               = _mm_mul_pd(r00,vftabscale);
 297             vfitab           = _mm_cvttpd_epi32(rt);
 298 #ifdef __XOP__
 299             vfeps            = _mm_frcz_pd(rt);
 300 #else
 301             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 302 #endif
 303             twovfeps         = _mm_add_pd(vfeps,vfeps);
 304             vfitab           = _mm_slli_epi32(vfitab,3);
 305
 306             /* REACTION-FIELD ELECTROSTATICS */
 307             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 308             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 309
 310             /* CUBIC SPLINE TABLE DISPERSION */
 311             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 312             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 313             GMX_MM_TRANSPOSE2_PD(Y,F);
 314             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 315             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 316             GMX_MM_TRANSPOSE2_PD(G,H);
 317             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 318             VV               = _mm_macc_pd(vfeps,Fp,Y);
 319             vvdw6            = _mm_mul_pd(c6_00,VV);
 320             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 321             fvdw6            = _mm_mul_pd(c6_00,FF);
 322
 323             /* CUBIC SPLINE TABLE REPULSION */
 324             vfitab           = _mm_add_epi32(vfitab,ifour);
 325             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 326             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 327             GMX_MM_TRANSPOSE2_PD(Y,F);
 328             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 329             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 330             GMX_MM_TRANSPOSE2_PD(G,H);
 331             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 332             VV               = _mm_macc_pd(vfeps,Fp,Y);
 333             vvdw12           = _mm_mul_pd(c12_00,VV);
 334             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 335             fvdw12           = _mm_mul_pd(c12_00,FF);
 336             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 337             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velecsum         = _mm_add_pd(velecsum,velec);
 341             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 342
 343             fscal            = _mm_add_pd(felec,fvdw);
 344
 345             /* Update vectorial force */
 346             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 347             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 348             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 349
 350             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 351             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 352             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             /* REACTION-FIELD ELECTROSTATICS */
 359             velec            = _mm_mul_pd(qq01,_mm_sub_pd(_mm_macc_pd(krf,rsq01,rinv01),crf));
 360             felec            = _mm_mul_pd(qq01,_mm_msub_pd(rinv01,rinvsq01,krf2));
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velecsum         = _mm_add_pd(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             /* Update vectorial force */
 368             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 369             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 370             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 371
 372             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 373             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 374             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             /* REACTION-FIELD ELECTROSTATICS */
 381             velec            = _mm_mul_pd(qq02,_mm_sub_pd(_mm_macc_pd(krf,rsq02,rinv02),crf));
 382             felec            = _mm_mul_pd(qq02,_mm_msub_pd(rinv02,rinvsq02,krf2));
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velecsum         = _mm_add_pd(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             /* Update vectorial force */
 390             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 391             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 392             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 393
 394             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 395             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 396             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             /* REACTION-FIELD ELECTROSTATICS */
 403             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 404             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velecsum         = _mm_add_pd(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             /* Update vectorial force */
 412             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 413             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 414             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 415
 416             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 417             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 418             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             /* REACTION-FIELD ELECTROSTATICS */
 425             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_macc_pd(krf,rsq11,rinv11),crf));
 426             felec            = _mm_mul_pd(qq11,_mm_msub_pd(rinv11,rinvsq11,krf2));
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velecsum         = _mm_add_pd(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             /* Update vectorial force */
 434             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 435             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 436             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 437
 438             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 439             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 440             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 441
 442             /**************************
 443              * CALCULATE INTERACTIONS *
 444              **************************/
 445
 446             /* REACTION-FIELD ELECTROSTATICS */
 447             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_macc_pd(krf,rsq12,rinv12),crf));
 448             felec            = _mm_mul_pd(qq12,_mm_msub_pd(rinv12,rinvsq12,krf2));
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velecsum         = _mm_add_pd(velecsum,velec);
 452
 453             fscal            = felec;
 454
 455             /* Update vectorial force */
 456             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 457             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 458             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 459
 460             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 461             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 462             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             /* REACTION-FIELD ELECTROSTATICS */
 469             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 470             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             velecsum         = _mm_add_pd(velecsum,velec);
 474
 475             fscal            = felec;
 476
 477             /* Update vectorial force */
 478             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 479             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 480             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 481
 482             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 483             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 484             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             /* REACTION-FIELD ELECTROSTATICS */
 491             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_macc_pd(krf,rsq21,rinv21),crf));
 492             felec            = _mm_mul_pd(qq21,_mm_msub_pd(rinv21,rinvsq21,krf2));
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             velecsum         = _mm_add_pd(velecsum,velec);
 496
 497             fscal            = felec;
 498
 499             /* Update vectorial force */
 500             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 501             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 502             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 503
 504             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 505             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 506             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             /* REACTION-FIELD ELECTROSTATICS */
 513             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_macc_pd(krf,rsq22,rinv22),crf));
 514             felec            = _mm_mul_pd(qq22,_mm_msub_pd(rinv22,rinvsq22,krf2));
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velecsum         = _mm_add_pd(velecsum,velec);
 518
 519             fscal            = felec;
 520
 521             /* Update vectorial force */
 522             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 523             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 524             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 525
 526             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 527             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 528             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 529
 530             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 531
 532             /* Inner loop uses 350 flops */
 533         }
 534
 535         if(jidx<j_index_end)
 536         {
 537
 538             jnrA             = jjnr[jidx];
 539             j_coord_offsetA  = DIM*jnrA;
 540
 541             /* load j atom coordinates */
 542             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 543                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 544
 545             /* Calculate displacement vector */
 546             dx00             = _mm_sub_pd(ix0,jx0);
 547             dy00             = _mm_sub_pd(iy0,jy0);
 548             dz00             = _mm_sub_pd(iz0,jz0);
 549             dx01             = _mm_sub_pd(ix0,jx1);
 550             dy01             = _mm_sub_pd(iy0,jy1);
 551             dz01             = _mm_sub_pd(iz0,jz1);
 552             dx02             = _mm_sub_pd(ix0,jx2);
 553             dy02             = _mm_sub_pd(iy0,jy2);
 554             dz02             = _mm_sub_pd(iz0,jz2);
 555             dx10             = _mm_sub_pd(ix1,jx0);
 556             dy10             = _mm_sub_pd(iy1,jy0);
 557             dz10             = _mm_sub_pd(iz1,jz0);
 558             dx11             = _mm_sub_pd(ix1,jx1);
 559             dy11             = _mm_sub_pd(iy1,jy1);
 560             dz11             = _mm_sub_pd(iz1,jz1);
 561             dx12             = _mm_sub_pd(ix1,jx2);
 562             dy12             = _mm_sub_pd(iy1,jy2);
 563             dz12             = _mm_sub_pd(iz1,jz2);
 564             dx20             = _mm_sub_pd(ix2,jx0);
 565             dy20             = _mm_sub_pd(iy2,jy0);
 566             dz20             = _mm_sub_pd(iz2,jz0);
 567             dx21             = _mm_sub_pd(ix2,jx1);
 568             dy21             = _mm_sub_pd(iy2,jy1);
 569             dz21             = _mm_sub_pd(iz2,jz1);
 570             dx22             = _mm_sub_pd(ix2,jx2);
 571             dy22             = _mm_sub_pd(iy2,jy2);
 572             dz22             = _mm_sub_pd(iz2,jz2);
 573
 574             /* Calculate squared distance and things based on it */
 575             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 576             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 577             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 578             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 579             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 580             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 581             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 582             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 583             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 584
 585             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 586             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 587             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 588             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 589             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 590             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 591             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 592             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 593             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 594
 595             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 596             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 597             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 598             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 599             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 600             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 601             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 602             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 603             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 604
 605             fjx0             = _mm_setzero_pd();
 606             fjy0             = _mm_setzero_pd();
 607             fjz0             = _mm_setzero_pd();
 608             fjx1             = _mm_setzero_pd();
 609             fjy1             = _mm_setzero_pd();
 610             fjz1             = _mm_setzero_pd();
 611             fjx2             = _mm_setzero_pd();
 612             fjy2             = _mm_setzero_pd();
 613             fjz2             = _mm_setzero_pd();
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             r00              = _mm_mul_pd(rsq00,rinv00);
 620
 621             /* Calculate table index by multiplying r with table scale and truncate to integer */
 622             rt               = _mm_mul_pd(r00,vftabscale);
 623             vfitab           = _mm_cvttpd_epi32(rt);
 624 #ifdef __XOP__
 625             vfeps            = _mm_frcz_pd(rt);
 626 #else
 627             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 628 #endif
 629             twovfeps         = _mm_add_pd(vfeps,vfeps);
 630             vfitab           = _mm_slli_epi32(vfitab,3);
 631
 632             /* REACTION-FIELD ELECTROSTATICS */
 633             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 634             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 635
 636             /* CUBIC SPLINE TABLE DISPERSION */
 637             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 638             F                = _mm_setzero_pd();
 639             GMX_MM_TRANSPOSE2_PD(Y,F);
 640             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 641             H                = _mm_setzero_pd();
 642             GMX_MM_TRANSPOSE2_PD(G,H);
 643             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 644             VV               = _mm_macc_pd(vfeps,Fp,Y);
 645             vvdw6            = _mm_mul_pd(c6_00,VV);
 646             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 647             fvdw6            = _mm_mul_pd(c6_00,FF);
 648
 649             /* CUBIC SPLINE TABLE REPULSION */
 650             vfitab           = _mm_add_epi32(vfitab,ifour);
 651             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 652             F                = _mm_setzero_pd();
 653             GMX_MM_TRANSPOSE2_PD(Y,F);
 654             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 655             H                = _mm_setzero_pd();
 656             GMX_MM_TRANSPOSE2_PD(G,H);
 657             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 658             VV               = _mm_macc_pd(vfeps,Fp,Y);
 659             vvdw12           = _mm_mul_pd(c12_00,VV);
 660             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 661             fvdw12           = _mm_mul_pd(c12_00,FF);
 662             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 663             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 664
 665             /* Update potential sum for this i atom from the interaction with this j atom. */
 666             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 667             velecsum         = _mm_add_pd(velecsum,velec);
 668             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 669             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 670
 671             fscal            = _mm_add_pd(felec,fvdw);
 672
 673             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 674
 675             /* Update vectorial force */
 676             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 677             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 678             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 679
 680             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 681             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 682             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             /* REACTION-FIELD ELECTROSTATICS */
 689             velec            = _mm_mul_pd(qq01,_mm_sub_pd(_mm_macc_pd(krf,rsq01,rinv01),crf));
 690             felec            = _mm_mul_pd(qq01,_mm_msub_pd(rinv01,rinvsq01,krf2));
 691
 692             /* Update potential sum for this i atom from the interaction with this j atom. */
 693             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 694             velecsum         = _mm_add_pd(velecsum,velec);
 695
 696             fscal            = felec;
 697
 698             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 699
 700             /* Update vectorial force */
 701             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 702             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 703             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 704
 705             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 706             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 707             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 708
 709             /**************************
 710              * CALCULATE INTERACTIONS *
 711              **************************/
 712
 713             /* REACTION-FIELD ELECTROSTATICS */
 714             velec            = _mm_mul_pd(qq02,_mm_sub_pd(_mm_macc_pd(krf,rsq02,rinv02),crf));
 715             felec            = _mm_mul_pd(qq02,_mm_msub_pd(rinv02,rinvsq02,krf2));
 716
 717             /* Update potential sum for this i atom from the interaction with this j atom. */
 718             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 719             velecsum         = _mm_add_pd(velecsum,velec);
 720
 721             fscal            = felec;
 722
 723             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 724
 725             /* Update vectorial force */
 726             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 727             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 728             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 729
 730             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 731             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 732             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 733
 734             /**************************
 735              * CALCULATE INTERACTIONS *
 736              **************************/
 737
 738             /* REACTION-FIELD ELECTROSTATICS */
 739             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 740             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 741
 742             /* Update potential sum for this i atom from the interaction with this j atom. */
 743             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 744             velecsum         = _mm_add_pd(velecsum,velec);
 745
 746             fscal            = felec;
 747
 748             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 749
 750             /* Update vectorial force */
 751             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 752             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 753             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 754
 755             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 756             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 757             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 758
 759             /**************************
 760              * CALCULATE INTERACTIONS *
 761              **************************/
 762
 763             /* REACTION-FIELD ELECTROSTATICS */
 764             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_macc_pd(krf,rsq11,rinv11),crf));
 765             felec            = _mm_mul_pd(qq11,_mm_msub_pd(rinv11,rinvsq11,krf2));
 766
 767             /* Update potential sum for this i atom from the interaction with this j atom. */
 768             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 769             velecsum         = _mm_add_pd(velecsum,velec);
 770
 771             fscal            = felec;
 772
 773             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 774
 775             /* Update vectorial force */
 776             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 777             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 778             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 779
 780             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 781             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 782             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 783
 784             /**************************
 785              * CALCULATE INTERACTIONS *
 786              **************************/
 787
 788             /* REACTION-FIELD ELECTROSTATICS */
 789             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_macc_pd(krf,rsq12,rinv12),crf));
 790             felec            = _mm_mul_pd(qq12,_mm_msub_pd(rinv12,rinvsq12,krf2));
 791
 792             /* Update potential sum for this i atom from the interaction with this j atom. */
 793             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 794             velecsum         = _mm_add_pd(velecsum,velec);
 795
 796             fscal            = felec;
 797
 798             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 799
 800             /* Update vectorial force */
 801             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 802             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 803             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 804
 805             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 806             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 807             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 808
 809             /**************************
 810              * CALCULATE INTERACTIONS *
 811              **************************/
 812
 813             /* REACTION-FIELD ELECTROSTATICS */
 814             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 815             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 816
 817             /* Update potential sum for this i atom from the interaction with this j atom. */
 818             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 819             velecsum         = _mm_add_pd(velecsum,velec);
 820
 821             fscal            = felec;
 822
 823             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 824
 825             /* Update vectorial force */
 826             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 827             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 828             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 829
 830             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 831             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 832             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 833
 834             /**************************
 835              * CALCULATE INTERACTIONS *
 836              **************************/
 837
 838             /* REACTION-FIELD ELECTROSTATICS */
 839             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_macc_pd(krf,rsq21,rinv21),crf));
 840             felec            = _mm_mul_pd(qq21,_mm_msub_pd(rinv21,rinvsq21,krf2));
 841
 842             /* Update potential sum for this i atom from the interaction with this j atom. */
 843             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 844             velecsum         = _mm_add_pd(velecsum,velec);
 845
 846             fscal            = felec;
 847
 848             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 849
 850             /* Update vectorial force */
 851             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 852             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 853             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 854
 855             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 856             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 857             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 858
 859             /**************************
 860              * CALCULATE INTERACTIONS *
 861              **************************/
 862
 863             /* REACTION-FIELD ELECTROSTATICS */
 864             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_macc_pd(krf,rsq22,rinv22),crf));
 865             felec            = _mm_mul_pd(qq22,_mm_msub_pd(rinv22,rinvsq22,krf2));
 866
 867             /* Update potential sum for this i atom from the interaction with this j atom. */
 868             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 869             velecsum         = _mm_add_pd(velecsum,velec);
 870
 871             fscal            = felec;
 872
 873             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 874
 875             /* Update vectorial force */
 876             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 877             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 878             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 879
 880             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 881             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 882             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 883
 884             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 885
 886             /* Inner loop uses 350 flops */
 887         }
 888
 889         /* End of innermost loop */
 890
 891         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 892                                               f+i_coord_offset,fshift+i_shift_offset);
 893
 894         ggid                        = gid[iidx];
 895         /* Update potential energies */
 896         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 897         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 898
 899         /* Increment number of inner iterations */
 900         inneriter                  += j_index_end - j_index_start;
 901
 902         /* Outer loop uses 20 flops */
 903     }
 904
 905     /* Increment number of outer iterations */
 906     outeriter        += nri;
 907
 908     /* Update outer/inner flops */
 909
 910     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*350);
 911 }
 912 /*
 913  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_128_fma_double
 914  * Electrostatics interaction: ReactionField
 915  * VdW interaction:            CubicSplineTable
 916  * Geometry:                   Water3-Water3
 917  * Calculate force/pot:        Force
 918  */
 919 void
 920 nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_avx_128_fma_double
 921                     (t_nblist                    * gmx_restrict       nlist,
 922                      rvec                        * gmx_restrict          xx,
 923                      rvec                        * gmx_restrict          ff,
 924                      t_forcerec                  * gmx_restrict          fr,
 925                      t_mdatoms                   * gmx_restrict     mdatoms,
 926                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 927                      t_nrnb                      * gmx_restrict        nrnb)
 928 {
 929     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 930      * just 0 for non-waters.
 931      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 932      * jnr indices corresponding to data put in the four positions in the SIMD register.
 933      */
 934     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 935     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 936     int              jnrA,jnrB;
 937     int              j_coord_offsetA,j_coord_offsetB;
 938     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 939     real             rcutoff_scalar;
 940     real             *shiftvec,*fshift,*x,*f;
 941     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 942     int              vdwioffset0;
 943     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 944     int              vdwioffset1;
 945     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 946     int              vdwioffset2;
 947     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 948     int              vdwjidx0A,vdwjidx0B;
 949     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 950     int              vdwjidx1A,vdwjidx1B;
 951     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 952     int              vdwjidx2A,vdwjidx2B;
 953     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 954     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 955     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 956     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 957     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 958     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 959     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 960     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 961     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 962     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 963     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 964     real             *charge;
 965     int              nvdwtype;
 966     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 967     int              *vdwtype;
 968     real             *vdwparam;
 969     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 970     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 971     __m128i          vfitab;
 972     __m128i          ifour       = _mm_set1_epi32(4);
 973     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 974     real             *vftab;
 975     __m128d          dummy_mask,cutoff_mask;
 976     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 977     __m128d          one     = _mm_set1_pd(1.0);
 978     __m128d          two     = _mm_set1_pd(2.0);
 979     x                = xx[0];
 980     f                = ff[0];
 981
 982     nri              = nlist->nri;
 983     iinr             = nlist->iinr;
 984     jindex           = nlist->jindex;
 985     jjnr             = nlist->jjnr;
 986     shiftidx         = nlist->shift;
 987     gid              = nlist->gid;
 988     shiftvec         = fr->shift_vec[0];
 989     fshift           = fr->fshift[0];
 990     facel            = _mm_set1_pd(fr->epsfac);
 991     charge           = mdatoms->chargeA;
 992     krf              = _mm_set1_pd(fr->ic->k_rf);
 993     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 994     crf              = _mm_set1_pd(fr->ic->c_rf);
 995     nvdwtype         = fr->ntype;
 996     vdwparam         = fr->nbfp;
 997     vdwtype          = mdatoms->typeA;
 998
 999     vftab            = kernel_data->table_vdw->data;
1000     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
1001
1002     /* Setup water-specific parameters */
1003     inr              = nlist->iinr[0];
1004     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1005     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1006     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1007     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1008
1009     jq0              = _mm_set1_pd(charge[inr+0]);
1010     jq1              = _mm_set1_pd(charge[inr+1]);
1011     jq2              = _mm_set1_pd(charge[inr+2]);
1012     vdwjidx0A        = 2*vdwtype[inr+0];
1013     qq00             = _mm_mul_pd(iq0,jq0);
1014     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1015     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1016     qq01             = _mm_mul_pd(iq0,jq1);
1017     qq02             = _mm_mul_pd(iq0,jq2);
1018     qq10             = _mm_mul_pd(iq1,jq0);
1019     qq11             = _mm_mul_pd(iq1,jq1);
1020     qq12             = _mm_mul_pd(iq1,jq2);
1021     qq20             = _mm_mul_pd(iq2,jq0);
1022     qq21             = _mm_mul_pd(iq2,jq1);
1023     qq22             = _mm_mul_pd(iq2,jq2);
1024
1025     /* Avoid stupid compiler warnings */
1026     jnrA = jnrB = 0;
1027     j_coord_offsetA = 0;
1028     j_coord_offsetB = 0;
1029
1030     outeriter        = 0;
1031     inneriter        = 0;
1032
1033     /* Start outer loop over neighborlists */
1034     for(iidx=0; iidx<nri; iidx++)
1035     {
1036         /* Load shift vector for this list */
1037         i_shift_offset   = DIM*shiftidx[iidx];
1038
1039         /* Load limits for loop over neighbors */
1040         j_index_start    = jindex[iidx];
1041         j_index_end      = jindex[iidx+1];
1042
1043         /* Get outer coordinate index */
1044         inr              = iinr[iidx];
1045         i_coord_offset   = DIM*inr;
1046
1047         /* Load i particle coords and add shift vector */
1048         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1049                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1050
1051         fix0             = _mm_setzero_pd();
1052         fiy0             = _mm_setzero_pd();
1053         fiz0             = _mm_setzero_pd();
1054         fix1             = _mm_setzero_pd();
1055         fiy1             = _mm_setzero_pd();
1056         fiz1             = _mm_setzero_pd();
1057         fix2             = _mm_setzero_pd();
1058         fiy2             = _mm_setzero_pd();
1059         fiz2             = _mm_setzero_pd();
1060
1061         /* Start inner kernel loop */
1062         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1063         {
1064
1065             /* Get j neighbor index, and coordinate index */
1066             jnrA             = jjnr[jidx];
1067             jnrB             = jjnr[jidx+1];
1068             j_coord_offsetA  = DIM*jnrA;
1069             j_coord_offsetB  = DIM*jnrB;
1070
1071             /* load j atom coordinates */
1072             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1073                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1074
1075             /* Calculate displacement vector */
1076             dx00             = _mm_sub_pd(ix0,jx0);
1077             dy00             = _mm_sub_pd(iy0,jy0);
1078             dz00             = _mm_sub_pd(iz0,jz0);
1079             dx01             = _mm_sub_pd(ix0,jx1);
1080             dy01             = _mm_sub_pd(iy0,jy1);
1081             dz01             = _mm_sub_pd(iz0,jz1);
1082             dx02             = _mm_sub_pd(ix0,jx2);
1083             dy02             = _mm_sub_pd(iy0,jy2);
1084             dz02             = _mm_sub_pd(iz0,jz2);
1085             dx10             = _mm_sub_pd(ix1,jx0);
1086             dy10             = _mm_sub_pd(iy1,jy0);
1087             dz10             = _mm_sub_pd(iz1,jz0);
1088             dx11             = _mm_sub_pd(ix1,jx1);
1089             dy11             = _mm_sub_pd(iy1,jy1);
1090             dz11             = _mm_sub_pd(iz1,jz1);
1091             dx12             = _mm_sub_pd(ix1,jx2);
1092             dy12             = _mm_sub_pd(iy1,jy2);
1093             dz12             = _mm_sub_pd(iz1,jz2);
1094             dx20             = _mm_sub_pd(ix2,jx0);
1095             dy20             = _mm_sub_pd(iy2,jy0);
1096             dz20             = _mm_sub_pd(iz2,jz0);
1097             dx21             = _mm_sub_pd(ix2,jx1);
1098             dy21             = _mm_sub_pd(iy2,jy1);
1099             dz21             = _mm_sub_pd(iz2,jz1);
1100             dx22             = _mm_sub_pd(ix2,jx2);
1101             dy22             = _mm_sub_pd(iy2,jy2);
1102             dz22             = _mm_sub_pd(iz2,jz2);
1103
1104             /* Calculate squared distance and things based on it */
1105             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1106             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1107             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1108             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1109             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1110             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1111             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1112             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1113             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1114
1115             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1116             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1117             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1118             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1119             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1120             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1121             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1122             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1123             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1124
1125             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1126             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1127             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1128             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1129             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1130             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1131             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1132             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1133             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1134
1135             fjx0             = _mm_setzero_pd();
1136             fjy0             = _mm_setzero_pd();
1137             fjz0             = _mm_setzero_pd();
1138             fjx1             = _mm_setzero_pd();
1139             fjy1             = _mm_setzero_pd();
1140             fjz1             = _mm_setzero_pd();
1141             fjx2             = _mm_setzero_pd();
1142             fjy2             = _mm_setzero_pd();
1143             fjz2             = _mm_setzero_pd();
1144
1145             /**************************
1146              * CALCULATE INTERACTIONS *
1147              **************************/
1148
1149             r00              = _mm_mul_pd(rsq00,rinv00);
1150
1151             /* Calculate table index by multiplying r with table scale and truncate to integer */
1152             rt               = _mm_mul_pd(r00,vftabscale);
1153             vfitab           = _mm_cvttpd_epi32(rt);
1154 #ifdef __XOP__
1155             vfeps            = _mm_frcz_pd(rt);
1156 #else
1157             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1158 #endif
1159             twovfeps         = _mm_add_pd(vfeps,vfeps);
1160             vfitab           = _mm_slli_epi32(vfitab,3);
1161
1162             /* REACTION-FIELD ELECTROSTATICS */
1163             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
1164
1165             /* CUBIC SPLINE TABLE DISPERSION */
1166             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1167             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1168             GMX_MM_TRANSPOSE2_PD(Y,F);
1169             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1170             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1171             GMX_MM_TRANSPOSE2_PD(G,H);
1172             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1173             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1174             fvdw6            = _mm_mul_pd(c6_00,FF);
1175
1176             /* CUBIC SPLINE TABLE REPULSION */
1177             vfitab           = _mm_add_epi32(vfitab,ifour);
1178             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1179             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1180             GMX_MM_TRANSPOSE2_PD(Y,F);
1181             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1182             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1183             GMX_MM_TRANSPOSE2_PD(G,H);
1184             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1185             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1186             fvdw12           = _mm_mul_pd(c12_00,FF);
1187             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1188
1189             fscal            = _mm_add_pd(felec,fvdw);
1190
1191             /* Update vectorial force */
1192             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1193             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1194             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1195
1196             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1197             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1198             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1199
1200             /**************************
1201              * CALCULATE INTERACTIONS *
1202              **************************/
1203
1204             /* REACTION-FIELD ELECTROSTATICS */
1205             felec            = _mm_mul_pd(qq01,_mm_msub_pd(rinv01,rinvsq01,krf2));
1206
1207             fscal            = felec;
1208
1209             /* Update vectorial force */
1210             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1211             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1212             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1213
1214             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1215             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1216             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             /* REACTION-FIELD ELECTROSTATICS */
1223             felec            = _mm_mul_pd(qq02,_mm_msub_pd(rinv02,rinvsq02,krf2));
1224
1225             fscal            = felec;
1226
1227             /* Update vectorial force */
1228             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1229             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1230             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1231
1232             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1233             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1234             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             /* REACTION-FIELD ELECTROSTATICS */
1241             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1242
1243             fscal            = felec;
1244
1245             /* Update vectorial force */
1246             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1247             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1248             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1249
1250             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1251             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1252             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1253
1254             /**************************
1255              * CALCULATE INTERACTIONS *
1256              **************************/
1257
1258             /* REACTION-FIELD ELECTROSTATICS */
1259             felec            = _mm_mul_pd(qq11,_mm_msub_pd(rinv11,rinvsq11,krf2));
1260
1261             fscal            = felec;
1262
1263             /* Update vectorial force */
1264             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1265             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1266             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1267
1268             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1269             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1270             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1271
1272             /**************************
1273              * CALCULATE INTERACTIONS *
1274              **************************/
1275
1276             /* REACTION-FIELD ELECTROSTATICS */
1277             felec            = _mm_mul_pd(qq12,_mm_msub_pd(rinv12,rinvsq12,krf2));
1278
1279             fscal            = felec;
1280
1281             /* Update vectorial force */
1282             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1283             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1284             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1285
1286             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1287             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1288             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1289
1290             /**************************
1291              * CALCULATE INTERACTIONS *
1292              **************************/
1293
1294             /* REACTION-FIELD ELECTROSTATICS */
1295             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1296
1297             fscal            = felec;
1298
1299             /* Update vectorial force */
1300             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1301             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1302             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1303
1304             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1305             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1306             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1307
1308             /**************************
1309              * CALCULATE INTERACTIONS *
1310              **************************/
1311
1312             /* REACTION-FIELD ELECTROSTATICS */
1313             felec            = _mm_mul_pd(qq21,_mm_msub_pd(rinv21,rinvsq21,krf2));
1314
1315             fscal            = felec;
1316
1317             /* Update vectorial force */
1318             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1319             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1320             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1321
1322             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1323             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1324             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1325
1326             /**************************
1327              * CALCULATE INTERACTIONS *
1328              **************************/
1329
1330             /* REACTION-FIELD ELECTROSTATICS */
1331             felec            = _mm_mul_pd(qq22,_mm_msub_pd(rinv22,rinvsq22,krf2));
1332
1333             fscal            = felec;
1334
1335             /* Update vectorial force */
1336             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1337             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1338             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1339
1340             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1341             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1342             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1343
1344             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1345
1346             /* Inner loop uses 297 flops */
1347         }
1348
1349         if(jidx<j_index_end)
1350         {
1351
1352             jnrA             = jjnr[jidx];
1353             j_coord_offsetA  = DIM*jnrA;
1354
1355             /* load j atom coordinates */
1356             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1357                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1358
1359             /* Calculate displacement vector */
1360             dx00             = _mm_sub_pd(ix0,jx0);
1361             dy00             = _mm_sub_pd(iy0,jy0);
1362             dz00             = _mm_sub_pd(iz0,jz0);
1363             dx01             = _mm_sub_pd(ix0,jx1);
1364             dy01             = _mm_sub_pd(iy0,jy1);
1365             dz01             = _mm_sub_pd(iz0,jz1);
1366             dx02             = _mm_sub_pd(ix0,jx2);
1367             dy02             = _mm_sub_pd(iy0,jy2);
1368             dz02             = _mm_sub_pd(iz0,jz2);
1369             dx10             = _mm_sub_pd(ix1,jx0);
1370             dy10             = _mm_sub_pd(iy1,jy0);
1371             dz10             = _mm_sub_pd(iz1,jz0);
1372             dx11             = _mm_sub_pd(ix1,jx1);
1373             dy11             = _mm_sub_pd(iy1,jy1);
1374             dz11             = _mm_sub_pd(iz1,jz1);
1375             dx12             = _mm_sub_pd(ix1,jx2);
1376             dy12             = _mm_sub_pd(iy1,jy2);
1377             dz12             = _mm_sub_pd(iz1,jz2);
1378             dx20             = _mm_sub_pd(ix2,jx0);
1379             dy20             = _mm_sub_pd(iy2,jy0);
1380             dz20             = _mm_sub_pd(iz2,jz0);
1381             dx21             = _mm_sub_pd(ix2,jx1);
1382             dy21             = _mm_sub_pd(iy2,jy1);
1383             dz21             = _mm_sub_pd(iz2,jz1);
1384             dx22             = _mm_sub_pd(ix2,jx2);
1385             dy22             = _mm_sub_pd(iy2,jy2);
1386             dz22             = _mm_sub_pd(iz2,jz2);
1387
1388             /* Calculate squared distance and things based on it */
1389             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1390             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1391             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1392             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1393             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1394             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1395             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1396             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1397             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1398
1399             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1400             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1401             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1402             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1403             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1404             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1405             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1406             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1407             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1408
1409             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1410             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1411             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1412             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1413             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1414             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1415             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1416             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1417             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1418
1419             fjx0             = _mm_setzero_pd();
1420             fjy0             = _mm_setzero_pd();
1421             fjz0             = _mm_setzero_pd();
1422             fjx1             = _mm_setzero_pd();
1423             fjy1             = _mm_setzero_pd();
1424             fjz1             = _mm_setzero_pd();
1425             fjx2             = _mm_setzero_pd();
1426             fjy2             = _mm_setzero_pd();
1427             fjz2             = _mm_setzero_pd();
1428
1429             /**************************
1430              * CALCULATE INTERACTIONS *
1431              **************************/
1432
1433             r00              = _mm_mul_pd(rsq00,rinv00);
1434
1435             /* Calculate table index by multiplying r with table scale and truncate to integer */
1436             rt               = _mm_mul_pd(r00,vftabscale);
1437             vfitab           = _mm_cvttpd_epi32(rt);
1438 #ifdef __XOP__
1439             vfeps            = _mm_frcz_pd(rt);
1440 #else
1441             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1442 #endif
1443             twovfeps         = _mm_add_pd(vfeps,vfeps);
1444             vfitab           = _mm_slli_epi32(vfitab,3);
1445
1446             /* REACTION-FIELD ELECTROSTATICS */
1447             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
1448
1449             /* CUBIC SPLINE TABLE DISPERSION */
1450             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1451             F                = _mm_setzero_pd();
1452             GMX_MM_TRANSPOSE2_PD(Y,F);
1453             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1454             H                = _mm_setzero_pd();
1455             GMX_MM_TRANSPOSE2_PD(G,H);
1456             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1457             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1458             fvdw6            = _mm_mul_pd(c6_00,FF);
1459
1460             /* CUBIC SPLINE TABLE REPULSION */
1461             vfitab           = _mm_add_epi32(vfitab,ifour);
1462             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1463             F                = _mm_setzero_pd();
1464             GMX_MM_TRANSPOSE2_PD(Y,F);
1465             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1466             H                = _mm_setzero_pd();
1467             GMX_MM_TRANSPOSE2_PD(G,H);
1468             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1469             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1470             fvdw12           = _mm_mul_pd(c12_00,FF);
1471             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1472
1473             fscal            = _mm_add_pd(felec,fvdw);
1474
1475             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1476
1477             /* Update vectorial force */
1478             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1479             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1480             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1481
1482             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1483             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1484             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1485
1486             /**************************
1487              * CALCULATE INTERACTIONS *
1488              **************************/
1489
1490             /* REACTION-FIELD ELECTROSTATICS */
1491             felec            = _mm_mul_pd(qq01,_mm_msub_pd(rinv01,rinvsq01,krf2));
1492
1493             fscal            = felec;
1494
1495             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1496
1497             /* Update vectorial force */
1498             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1499             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1500             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1501
1502             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1503             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1504             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1505
1506             /**************************
1507              * CALCULATE INTERACTIONS *
1508              **************************/
1509
1510             /* REACTION-FIELD ELECTROSTATICS */
1511             felec            = _mm_mul_pd(qq02,_mm_msub_pd(rinv02,rinvsq02,krf2));
1512
1513             fscal            = felec;
1514
1515             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1516
1517             /* Update vectorial force */
1518             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1519             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1520             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1521
1522             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1523             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1524             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1525
1526             /**************************
1527              * CALCULATE INTERACTIONS *
1528              **************************/
1529
1530             /* REACTION-FIELD ELECTROSTATICS */
1531             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1532
1533             fscal            = felec;
1534
1535             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1536
1537             /* Update vectorial force */
1538             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1539             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1540             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1541
1542             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1543             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1544             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1545
1546             /**************************
1547              * CALCULATE INTERACTIONS *
1548              **************************/
1549
1550             /* REACTION-FIELD ELECTROSTATICS */
1551             felec            = _mm_mul_pd(qq11,_mm_msub_pd(rinv11,rinvsq11,krf2));
1552
1553             fscal            = felec;
1554
1555             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1556
1557             /* Update vectorial force */
1558             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1559             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1560             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1561
1562             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1563             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1564             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1565
1566             /**************************
1567              * CALCULATE INTERACTIONS *
1568              **************************/
1569
1570             /* REACTION-FIELD ELECTROSTATICS */
1571             felec            = _mm_mul_pd(qq12,_mm_msub_pd(rinv12,rinvsq12,krf2));
1572
1573             fscal            = felec;
1574
1575             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1576
1577             /* Update vectorial force */
1578             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1579             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1580             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1581
1582             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1583             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1584             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1585
1586             /**************************
1587              * CALCULATE INTERACTIONS *
1588              **************************/
1589
1590             /* REACTION-FIELD ELECTROSTATICS */
1591             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1592
1593             fscal            = felec;
1594
1595             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1596
1597             /* Update vectorial force */
1598             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1599             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1600             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1601
1602             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1603             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1604             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1605
1606             /**************************
1607              * CALCULATE INTERACTIONS *
1608              **************************/
1609
1610             /* REACTION-FIELD ELECTROSTATICS */
1611             felec            = _mm_mul_pd(qq21,_mm_msub_pd(rinv21,rinvsq21,krf2));
1612
1613             fscal            = felec;
1614
1615             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1616
1617             /* Update vectorial force */
1618             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1619             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1620             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1621
1622             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1623             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1624             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1625
1626             /**************************
1627              * CALCULATE INTERACTIONS *
1628              **************************/
1629
1630             /* REACTION-FIELD ELECTROSTATICS */
1631             felec            = _mm_mul_pd(qq22,_mm_msub_pd(rinv22,rinvsq22,krf2));
1632
1633             fscal            = felec;
1634
1635             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1636
1637             /* Update vectorial force */
1638             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1639             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1640             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1641
1642             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1643             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1644             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1645
1646             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1647
1648             /* Inner loop uses 297 flops */
1649         }
1650
1651         /* End of innermost loop */
1652
1653         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1654                                               f+i_coord_offset,fshift+i_shift_offset);
1655
1656         /* Increment number of inner iterations */
1657         inneriter                  += j_index_end - j_index_start;
1658
1659         /* Outer loop uses 18 flops */
1660     }
1661
1662     /* Increment number of outer iterations */
1663     outeriter        += nri;
1664
1665     /* Update outer/inner flops */
1666
1667     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*297);
1668 }