src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_VF_avx_128_fma_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     int              vdwjidx1A,vdwjidx1B;
  75     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  76     int              vdwjidx2A,vdwjidx2B;
  77     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  78     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  79     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  80     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  81     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  83     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  84     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  85     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  86     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  98     real             *vftab;
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_elec_vdw->data;
 121     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 126     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 127     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 128     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 129
 130     jq0              = _mm_set1_pd(charge[inr+0]);
 131     jq1              = _mm_set1_pd(charge[inr+1]);
 132     jq2              = _mm_set1_pd(charge[inr+2]);
 133     vdwjidx0A        = 2*vdwtype[inr+0];
 134     qq00             = _mm_mul_pd(iq0,jq0);
 135     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 136     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 137     qq01             = _mm_mul_pd(iq0,jq1);
 138     qq02             = _mm_mul_pd(iq0,jq2);
 139     qq10             = _mm_mul_pd(iq1,jq0);
 140     qq11             = _mm_mul_pd(iq1,jq1);
 141     qq12             = _mm_mul_pd(iq1,jq2);
 142     qq20             = _mm_mul_pd(iq2,jq0);
 143     qq21             = _mm_mul_pd(iq2,jq1);
 144     qq22             = _mm_mul_pd(iq2,jq2);
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm_setzero_pd();
 173         fiy0             = _mm_setzero_pd();
 174         fiz0             = _mm_setzero_pd();
 175         fix1             = _mm_setzero_pd();
 176         fiy1             = _mm_setzero_pd();
 177         fiz1             = _mm_setzero_pd();
 178         fix2             = _mm_setzero_pd();
 179         fiy2             = _mm_setzero_pd();
 180         fiz2             = _mm_setzero_pd();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_pd();
 184         vvdwsum          = _mm_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_pd(ix0,jx0);
 202             dy00             = _mm_sub_pd(iy0,jy0);
 203             dz00             = _mm_sub_pd(iz0,jz0);
 204             dx01             = _mm_sub_pd(ix0,jx1);
 205             dy01             = _mm_sub_pd(iy0,jy1);
 206             dz01             = _mm_sub_pd(iz0,jz1);
 207             dx02             = _mm_sub_pd(ix0,jx2);
 208             dy02             = _mm_sub_pd(iy0,jy2);
 209             dz02             = _mm_sub_pd(iz0,jz2);
 210             dx10             = _mm_sub_pd(ix1,jx0);
 211             dy10             = _mm_sub_pd(iy1,jy0);
 212             dz10             = _mm_sub_pd(iz1,jz0);
 213             dx11             = _mm_sub_pd(ix1,jx1);
 214             dy11             = _mm_sub_pd(iy1,jy1);
 215             dz11             = _mm_sub_pd(iz1,jz1);
 216             dx12             = _mm_sub_pd(ix1,jx2);
 217             dy12             = _mm_sub_pd(iy1,jy2);
 218             dz12             = _mm_sub_pd(iz1,jz2);
 219             dx20             = _mm_sub_pd(ix2,jx0);
 220             dy20             = _mm_sub_pd(iy2,jy0);
 221             dz20             = _mm_sub_pd(iz2,jz0);
 222             dx21             = _mm_sub_pd(ix2,jx1);
 223             dy21             = _mm_sub_pd(iy2,jy1);
 224             dz21             = _mm_sub_pd(iz2,jz1);
 225             dx22             = _mm_sub_pd(ix2,jx2);
 226             dy22             = _mm_sub_pd(iy2,jy2);
 227             dz22             = _mm_sub_pd(iz2,jz2);
 228
 229             /* Calculate squared distance and things based on it */
 230             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 231             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 232             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 233             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 234             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 235             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 236             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 237             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 238             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 239
 240             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 241             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 242             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 243             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 244             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 245             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 246             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 247             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 248             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 249
 250             fjx0             = _mm_setzero_pd();
 251             fjy0             = _mm_setzero_pd();
 252             fjz0             = _mm_setzero_pd();
 253             fjx1             = _mm_setzero_pd();
 254             fjy1             = _mm_setzero_pd();
 255             fjz1             = _mm_setzero_pd();
 256             fjx2             = _mm_setzero_pd();
 257             fjy2             = _mm_setzero_pd();
 258             fjz2             = _mm_setzero_pd();
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             r00              = _mm_mul_pd(rsq00,rinv00);
 265
 266             /* Calculate table index by multiplying r with table scale and truncate to integer */
 267             rt               = _mm_mul_pd(r00,vftabscale);
 268             vfitab           = _mm_cvttpd_epi32(rt);
 269 #ifdef __XOP__
 270             vfeps            = _mm_frcz_pd(rt);
 271 #else
 272             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 273 #endif
 274             twovfeps         = _mm_add_pd(vfeps,vfeps);
 275             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 276
 277             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 278             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 279             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 280             GMX_MM_TRANSPOSE2_PD(Y,F);
 281             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 282             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 283             GMX_MM_TRANSPOSE2_PD(G,H);
 284             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 285             VV               = _mm_macc_pd(vfeps,Fp,Y);
 286             velec            = _mm_mul_pd(qq00,VV);
 287             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 288             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 289
 290             /* CUBIC SPLINE TABLE DISPERSION */
 291             vfitab           = _mm_add_epi32(vfitab,ifour);
 292             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 293             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 294             GMX_MM_TRANSPOSE2_PD(Y,F);
 295             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 296             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 297             GMX_MM_TRANSPOSE2_PD(G,H);
 298             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 299             VV               = _mm_macc_pd(vfeps,Fp,Y);
 300             vvdw6            = _mm_mul_pd(c6_00,VV);
 301             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 302             fvdw6            = _mm_mul_pd(c6_00,FF);
 303
 304             /* CUBIC SPLINE TABLE REPULSION */
 305             vfitab           = _mm_add_epi32(vfitab,ifour);
 306             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 307             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 308             GMX_MM_TRANSPOSE2_PD(Y,F);
 309             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 310             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 311             GMX_MM_TRANSPOSE2_PD(G,H);
 312             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 313             VV               = _mm_macc_pd(vfeps,Fp,Y);
 314             vvdw12           = _mm_mul_pd(c12_00,VV);
 315             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 316             fvdw12           = _mm_mul_pd(c12_00,FF);
 317             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 318             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velecsum         = _mm_add_pd(velecsum,velec);
 322             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 323
 324             fscal            = _mm_add_pd(felec,fvdw);
 325
 326             /* Update vectorial force */
 327             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 328             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 329             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 330
 331             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 332             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 333             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             r01              = _mm_mul_pd(rsq01,rinv01);
 340
 341             /* Calculate table index by multiplying r with table scale and truncate to integer */
 342             rt               = _mm_mul_pd(r01,vftabscale);
 343             vfitab           = _mm_cvttpd_epi32(rt);
 344 #ifdef __XOP__
 345             vfeps            = _mm_frcz_pd(rt);
 346 #else
 347             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 348 #endif
 349             twovfeps         = _mm_add_pd(vfeps,vfeps);
 350             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 351
 352             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 353             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 354             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 355             GMX_MM_TRANSPOSE2_PD(Y,F);
 356             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 357             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 358             GMX_MM_TRANSPOSE2_PD(G,H);
 359             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 360             VV               = _mm_macc_pd(vfeps,Fp,Y);
 361             velec            = _mm_mul_pd(qq01,VV);
 362             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 363             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velecsum         = _mm_add_pd(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             /* Update vectorial force */
 371             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 372             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 373             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 374
 375             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 376             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 377             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             r02              = _mm_mul_pd(rsq02,rinv02);
 384
 385             /* Calculate table index by multiplying r with table scale and truncate to integer */
 386             rt               = _mm_mul_pd(r02,vftabscale);
 387             vfitab           = _mm_cvttpd_epi32(rt);
 388 #ifdef __XOP__
 389             vfeps            = _mm_frcz_pd(rt);
 390 #else
 391             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 392 #endif
 393             twovfeps         = _mm_add_pd(vfeps,vfeps);
 394             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 395
 396             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 397             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 398             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 399             GMX_MM_TRANSPOSE2_PD(Y,F);
 400             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 401             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 402             GMX_MM_TRANSPOSE2_PD(G,H);
 403             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 404             VV               = _mm_macc_pd(vfeps,Fp,Y);
 405             velec            = _mm_mul_pd(qq02,VV);
 406             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 407             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velecsum         = _mm_add_pd(velecsum,velec);
 411
 412             fscal            = felec;
 413
 414             /* Update vectorial force */
 415             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 416             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 417             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 418
 419             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 420             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 421             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             r10              = _mm_mul_pd(rsq10,rinv10);
 428
 429             /* Calculate table index by multiplying r with table scale and truncate to integer */
 430             rt               = _mm_mul_pd(r10,vftabscale);
 431             vfitab           = _mm_cvttpd_epi32(rt);
 432 #ifdef __XOP__
 433             vfeps            = _mm_frcz_pd(rt);
 434 #else
 435             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 436 #endif
 437             twovfeps         = _mm_add_pd(vfeps,vfeps);
 438             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 439
 440             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 441             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 442             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 443             GMX_MM_TRANSPOSE2_PD(Y,F);
 444             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 445             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 446             GMX_MM_TRANSPOSE2_PD(G,H);
 447             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 448             VV               = _mm_macc_pd(vfeps,Fp,Y);
 449             velec            = _mm_mul_pd(qq10,VV);
 450             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 451             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 452
 453             /* Update potential sum for this i atom from the interaction with this j atom. */
 454             velecsum         = _mm_add_pd(velecsum,velec);
 455
 456             fscal            = felec;
 457
 458             /* Update vectorial force */
 459             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 460             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 461             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 462
 463             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 464             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 465             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             r11              = _mm_mul_pd(rsq11,rinv11);
 472
 473             /* Calculate table index by multiplying r with table scale and truncate to integer */
 474             rt               = _mm_mul_pd(r11,vftabscale);
 475             vfitab           = _mm_cvttpd_epi32(rt);
 476 #ifdef __XOP__
 477             vfeps            = _mm_frcz_pd(rt);
 478 #else
 479             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 480 #endif
 481             twovfeps         = _mm_add_pd(vfeps,vfeps);
 482             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 483
 484             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 485             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 486             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 487             GMX_MM_TRANSPOSE2_PD(Y,F);
 488             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 489             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 490             GMX_MM_TRANSPOSE2_PD(G,H);
 491             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 492             VV               = _mm_macc_pd(vfeps,Fp,Y);
 493             velec            = _mm_mul_pd(qq11,VV);
 494             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 495             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
 496
 497             /* Update potential sum for this i atom from the interaction with this j atom. */
 498             velecsum         = _mm_add_pd(velecsum,velec);
 499
 500             fscal            = felec;
 501
 502             /* Update vectorial force */
 503             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 504             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 505             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 506
 507             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 508             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 509             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             r12              = _mm_mul_pd(rsq12,rinv12);
 516
 517             /* Calculate table index by multiplying r with table scale and truncate to integer */
 518             rt               = _mm_mul_pd(r12,vftabscale);
 519             vfitab           = _mm_cvttpd_epi32(rt);
 520 #ifdef __XOP__
 521             vfeps            = _mm_frcz_pd(rt);
 522 #else
 523             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 524 #endif
 525             twovfeps         = _mm_add_pd(vfeps,vfeps);
 526             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 527
 528             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 529             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 530             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 531             GMX_MM_TRANSPOSE2_PD(Y,F);
 532             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 533             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 534             GMX_MM_TRANSPOSE2_PD(G,H);
 535             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 536             VV               = _mm_macc_pd(vfeps,Fp,Y);
 537             velec            = _mm_mul_pd(qq12,VV);
 538             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 539             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velecsum         = _mm_add_pd(velecsum,velec);
 543
 544             fscal            = felec;
 545
 546             /* Update vectorial force */
 547             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 548             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 549             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 550
 551             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 552             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 553             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             r20              = _mm_mul_pd(rsq20,rinv20);
 560
 561             /* Calculate table index by multiplying r with table scale and truncate to integer */
 562             rt               = _mm_mul_pd(r20,vftabscale);
 563             vfitab           = _mm_cvttpd_epi32(rt);
 564 #ifdef __XOP__
 565             vfeps            = _mm_frcz_pd(rt);
 566 #else
 567             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 568 #endif
 569             twovfeps         = _mm_add_pd(vfeps,vfeps);
 570             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 571
 572             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 573             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 574             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 575             GMX_MM_TRANSPOSE2_PD(Y,F);
 576             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 577             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 578             GMX_MM_TRANSPOSE2_PD(G,H);
 579             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 580             VV               = _mm_macc_pd(vfeps,Fp,Y);
 581             velec            = _mm_mul_pd(qq20,VV);
 582             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 583             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 584
 585             /* Update potential sum for this i atom from the interaction with this j atom. */
 586             velecsum         = _mm_add_pd(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             /* Update vectorial force */
 591             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 592             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 593             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 594
 595             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 596             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 597             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             r21              = _mm_mul_pd(rsq21,rinv21);
 604
 605             /* Calculate table index by multiplying r with table scale and truncate to integer */
 606             rt               = _mm_mul_pd(r21,vftabscale);
 607             vfitab           = _mm_cvttpd_epi32(rt);
 608 #ifdef __XOP__
 609             vfeps            = _mm_frcz_pd(rt);
 610 #else
 611             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 612 #endif
 613             twovfeps         = _mm_add_pd(vfeps,vfeps);
 614             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 615
 616             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 617             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 618             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 619             GMX_MM_TRANSPOSE2_PD(Y,F);
 620             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 621             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 622             GMX_MM_TRANSPOSE2_PD(G,H);
 623             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 624             VV               = _mm_macc_pd(vfeps,Fp,Y);
 625             velec            = _mm_mul_pd(qq21,VV);
 626             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 627             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
 628
 629             /* Update potential sum for this i atom from the interaction with this j atom. */
 630             velecsum         = _mm_add_pd(velecsum,velec);
 631
 632             fscal            = felec;
 633
 634             /* Update vectorial force */
 635             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 636             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 637             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 638
 639             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 640             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 641             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             r22              = _mm_mul_pd(rsq22,rinv22);
 648
 649             /* Calculate table index by multiplying r with table scale and truncate to integer */
 650             rt               = _mm_mul_pd(r22,vftabscale);
 651             vfitab           = _mm_cvttpd_epi32(rt);
 652 #ifdef __XOP__
 653             vfeps            = _mm_frcz_pd(rt);
 654 #else
 655             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 656 #endif
 657             twovfeps         = _mm_add_pd(vfeps,vfeps);
 658             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 659
 660             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 661             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 662             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 663             GMX_MM_TRANSPOSE2_PD(Y,F);
 664             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 665             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 666             GMX_MM_TRANSPOSE2_PD(G,H);
 667             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 668             VV               = _mm_macc_pd(vfeps,Fp,Y);
 669             velec            = _mm_mul_pd(qq22,VV);
 670             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 671             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
 672
 673             /* Update potential sum for this i atom from the interaction with this j atom. */
 674             velecsum         = _mm_add_pd(velecsum,velec);
 675
 676             fscal            = felec;
 677
 678             /* Update vectorial force */
 679             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 680             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 681             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 682
 683             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 684             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 685             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 686
 687             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 688
 689             /* Inner loop uses 444 flops */
 690         }
 691
 692         if(jidx<j_index_end)
 693         {
 694
 695             jnrA             = jjnr[jidx];
 696             j_coord_offsetA  = DIM*jnrA;
 697
 698             /* load j atom coordinates */
 699             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 700                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 701
 702             /* Calculate displacement vector */
 703             dx00             = _mm_sub_pd(ix0,jx0);
 704             dy00             = _mm_sub_pd(iy0,jy0);
 705             dz00             = _mm_sub_pd(iz0,jz0);
 706             dx01             = _mm_sub_pd(ix0,jx1);
 707             dy01             = _mm_sub_pd(iy0,jy1);
 708             dz01             = _mm_sub_pd(iz0,jz1);
 709             dx02             = _mm_sub_pd(ix0,jx2);
 710             dy02             = _mm_sub_pd(iy0,jy2);
 711             dz02             = _mm_sub_pd(iz0,jz2);
 712             dx10             = _mm_sub_pd(ix1,jx0);
 713             dy10             = _mm_sub_pd(iy1,jy0);
 714             dz10             = _mm_sub_pd(iz1,jz0);
 715             dx11             = _mm_sub_pd(ix1,jx1);
 716             dy11             = _mm_sub_pd(iy1,jy1);
 717             dz11             = _mm_sub_pd(iz1,jz1);
 718             dx12             = _mm_sub_pd(ix1,jx2);
 719             dy12             = _mm_sub_pd(iy1,jy2);
 720             dz12             = _mm_sub_pd(iz1,jz2);
 721             dx20             = _mm_sub_pd(ix2,jx0);
 722             dy20             = _mm_sub_pd(iy2,jy0);
 723             dz20             = _mm_sub_pd(iz2,jz0);
 724             dx21             = _mm_sub_pd(ix2,jx1);
 725             dy21             = _mm_sub_pd(iy2,jy1);
 726             dz21             = _mm_sub_pd(iz2,jz1);
 727             dx22             = _mm_sub_pd(ix2,jx2);
 728             dy22             = _mm_sub_pd(iy2,jy2);
 729             dz22             = _mm_sub_pd(iz2,jz2);
 730
 731             /* Calculate squared distance and things based on it */
 732             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 733             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 734             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 735             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 736             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 737             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 738             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 739             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 740             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 741
 742             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 743             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 744             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 745             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 746             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 747             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 748             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 749             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 750             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 751
 752             fjx0             = _mm_setzero_pd();
 753             fjy0             = _mm_setzero_pd();
 754             fjz0             = _mm_setzero_pd();
 755             fjx1             = _mm_setzero_pd();
 756             fjy1             = _mm_setzero_pd();
 757             fjz1             = _mm_setzero_pd();
 758             fjx2             = _mm_setzero_pd();
 759             fjy2             = _mm_setzero_pd();
 760             fjz2             = _mm_setzero_pd();
 761
 762             /**************************
 763              * CALCULATE INTERACTIONS *
 764              **************************/
 765
 766             r00              = _mm_mul_pd(rsq00,rinv00);
 767
 768             /* Calculate table index by multiplying r with table scale and truncate to integer */
 769             rt               = _mm_mul_pd(r00,vftabscale);
 770             vfitab           = _mm_cvttpd_epi32(rt);
 771 #ifdef __XOP__
 772             vfeps            = _mm_frcz_pd(rt);
 773 #else
 774             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 775 #endif
 776             twovfeps         = _mm_add_pd(vfeps,vfeps);
 777             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 778
 779             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 780             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 781             F                = _mm_setzero_pd();
 782             GMX_MM_TRANSPOSE2_PD(Y,F);
 783             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 784             H                = _mm_setzero_pd();
 785             GMX_MM_TRANSPOSE2_PD(G,H);
 786             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 787             VV               = _mm_macc_pd(vfeps,Fp,Y);
 788             velec            = _mm_mul_pd(qq00,VV);
 789             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 790             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 791
 792             /* CUBIC SPLINE TABLE DISPERSION */
 793             vfitab           = _mm_add_epi32(vfitab,ifour);
 794             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 795             F                = _mm_setzero_pd();
 796             GMX_MM_TRANSPOSE2_PD(Y,F);
 797             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 798             H                = _mm_setzero_pd();
 799             GMX_MM_TRANSPOSE2_PD(G,H);
 800             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 801             VV               = _mm_macc_pd(vfeps,Fp,Y);
 802             vvdw6            = _mm_mul_pd(c6_00,VV);
 803             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 804             fvdw6            = _mm_mul_pd(c6_00,FF);
 805
 806             /* CUBIC SPLINE TABLE REPULSION */
 807             vfitab           = _mm_add_epi32(vfitab,ifour);
 808             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 809             F                = _mm_setzero_pd();
 810             GMX_MM_TRANSPOSE2_PD(Y,F);
 811             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 812             H                = _mm_setzero_pd();
 813             GMX_MM_TRANSPOSE2_PD(G,H);
 814             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 815             VV               = _mm_macc_pd(vfeps,Fp,Y);
 816             vvdw12           = _mm_mul_pd(c12_00,VV);
 817             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 818             fvdw12           = _mm_mul_pd(c12_00,FF);
 819             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 820             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 821
 822             /* Update potential sum for this i atom from the interaction with this j atom. */
 823             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 824             velecsum         = _mm_add_pd(velecsum,velec);
 825             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 826             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 827
 828             fscal            = _mm_add_pd(felec,fvdw);
 829
 830             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 831
 832             /* Update vectorial force */
 833             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 834             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 835             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 836
 837             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 838             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 839             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             r01              = _mm_mul_pd(rsq01,rinv01);
 846
 847             /* Calculate table index by multiplying r with table scale and truncate to integer */
 848             rt               = _mm_mul_pd(r01,vftabscale);
 849             vfitab           = _mm_cvttpd_epi32(rt);
 850 #ifdef __XOP__
 851             vfeps            = _mm_frcz_pd(rt);
 852 #else
 853             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 854 #endif
 855             twovfeps         = _mm_add_pd(vfeps,vfeps);
 856             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 857
 858             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 859             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 860             F                = _mm_setzero_pd();
 861             GMX_MM_TRANSPOSE2_PD(Y,F);
 862             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 863             H                = _mm_setzero_pd();
 864             GMX_MM_TRANSPOSE2_PD(G,H);
 865             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 866             VV               = _mm_macc_pd(vfeps,Fp,Y);
 867             velec            = _mm_mul_pd(qq01,VV);
 868             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 869             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
 870
 871             /* Update potential sum for this i atom from the interaction with this j atom. */
 872             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 873             velecsum         = _mm_add_pd(velecsum,velec);
 874
 875             fscal            = felec;
 876
 877             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 878
 879             /* Update vectorial force */
 880             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 881             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 882             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 883
 884             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 885             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 886             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 887
 888             /**************************
 889              * CALCULATE INTERACTIONS *
 890              **************************/
 891
 892             r02              = _mm_mul_pd(rsq02,rinv02);
 893
 894             /* Calculate table index by multiplying r with table scale and truncate to integer */
 895             rt               = _mm_mul_pd(r02,vftabscale);
 896             vfitab           = _mm_cvttpd_epi32(rt);
 897 #ifdef __XOP__
 898             vfeps            = _mm_frcz_pd(rt);
 899 #else
 900             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 901 #endif
 902             twovfeps         = _mm_add_pd(vfeps,vfeps);
 903             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 904
 905             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 906             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 907             F                = _mm_setzero_pd();
 908             GMX_MM_TRANSPOSE2_PD(Y,F);
 909             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 910             H                = _mm_setzero_pd();
 911             GMX_MM_TRANSPOSE2_PD(G,H);
 912             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 913             VV               = _mm_macc_pd(vfeps,Fp,Y);
 914             velec            = _mm_mul_pd(qq02,VV);
 915             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 916             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
 917
 918             /* Update potential sum for this i atom from the interaction with this j atom. */
 919             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 920             velecsum         = _mm_add_pd(velecsum,velec);
 921
 922             fscal            = felec;
 923
 924             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 925
 926             /* Update vectorial force */
 927             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 928             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 929             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 930
 931             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 932             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 933             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 934
 935             /**************************
 936              * CALCULATE INTERACTIONS *
 937              **************************/
 938
 939             r10              = _mm_mul_pd(rsq10,rinv10);
 940
 941             /* Calculate table index by multiplying r with table scale and truncate to integer */
 942             rt               = _mm_mul_pd(r10,vftabscale);
 943             vfitab           = _mm_cvttpd_epi32(rt);
 944 #ifdef __XOP__
 945             vfeps            = _mm_frcz_pd(rt);
 946 #else
 947             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 948 #endif
 949             twovfeps         = _mm_add_pd(vfeps,vfeps);
 950             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 951
 952             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 953             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 954             F                = _mm_setzero_pd();
 955             GMX_MM_TRANSPOSE2_PD(Y,F);
 956             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 957             H                = _mm_setzero_pd();
 958             GMX_MM_TRANSPOSE2_PD(G,H);
 959             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 960             VV               = _mm_macc_pd(vfeps,Fp,Y);
 961             velec            = _mm_mul_pd(qq10,VV);
 962             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 963             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 964
 965             /* Update potential sum for this i atom from the interaction with this j atom. */
 966             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 967             velecsum         = _mm_add_pd(velecsum,velec);
 968
 969             fscal            = felec;
 970
 971             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 972
 973             /* Update vectorial force */
 974             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 975             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 976             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 977
 978             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 979             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 980             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 981
 982             /**************************
 983              * CALCULATE INTERACTIONS *
 984              **************************/
 985
 986             r11              = _mm_mul_pd(rsq11,rinv11);
 987
 988             /* Calculate table index by multiplying r with table scale and truncate to integer */
 989             rt               = _mm_mul_pd(r11,vftabscale);
 990             vfitab           = _mm_cvttpd_epi32(rt);
 991 #ifdef __XOP__
 992             vfeps            = _mm_frcz_pd(rt);
 993 #else
 994             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 995 #endif
 996             twovfeps         = _mm_add_pd(vfeps,vfeps);
 997             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 998
 999             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1000             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1001             F                = _mm_setzero_pd();
1002             GMX_MM_TRANSPOSE2_PD(Y,F);
1003             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1004             H                = _mm_setzero_pd();
1005             GMX_MM_TRANSPOSE2_PD(G,H);
1006             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1007             VV               = _mm_macc_pd(vfeps,Fp,Y);
1008             velec            = _mm_mul_pd(qq11,VV);
1009             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1010             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
1011
1012             /* Update potential sum for this i atom from the interaction with this j atom. */
1013             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1014             velecsum         = _mm_add_pd(velecsum,velec);
1015
1016             fscal            = felec;
1017
1018             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1019
1020             /* Update vectorial force */
1021             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1022             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1023             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1024
1025             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1026             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1027             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1028
1029             /**************************
1030              * CALCULATE INTERACTIONS *
1031              **************************/
1032
1033             r12              = _mm_mul_pd(rsq12,rinv12);
1034
1035             /* Calculate table index by multiplying r with table scale and truncate to integer */
1036             rt               = _mm_mul_pd(r12,vftabscale);
1037             vfitab           = _mm_cvttpd_epi32(rt);
1038 #ifdef __XOP__
1039             vfeps            = _mm_frcz_pd(rt);
1040 #else
1041             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1042 #endif
1043             twovfeps         = _mm_add_pd(vfeps,vfeps);
1044             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1045
1046             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1047             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1048             F                = _mm_setzero_pd();
1049             GMX_MM_TRANSPOSE2_PD(Y,F);
1050             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1051             H                = _mm_setzero_pd();
1052             GMX_MM_TRANSPOSE2_PD(G,H);
1053             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1054             VV               = _mm_macc_pd(vfeps,Fp,Y);
1055             velec            = _mm_mul_pd(qq12,VV);
1056             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1057             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
1058
1059             /* Update potential sum for this i atom from the interaction with this j atom. */
1060             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1061             velecsum         = _mm_add_pd(velecsum,velec);
1062
1063             fscal            = felec;
1064
1065             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1066
1067             /* Update vectorial force */
1068             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1069             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1070             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1071
1072             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1073             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1074             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             r20              = _mm_mul_pd(rsq20,rinv20);
1081
1082             /* Calculate table index by multiplying r with table scale and truncate to integer */
1083             rt               = _mm_mul_pd(r20,vftabscale);
1084             vfitab           = _mm_cvttpd_epi32(rt);
1085 #ifdef __XOP__
1086             vfeps            = _mm_frcz_pd(rt);
1087 #else
1088             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1089 #endif
1090             twovfeps         = _mm_add_pd(vfeps,vfeps);
1091             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1092
1093             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1094             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1095             F                = _mm_setzero_pd();
1096             GMX_MM_TRANSPOSE2_PD(Y,F);
1097             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1098             H                = _mm_setzero_pd();
1099             GMX_MM_TRANSPOSE2_PD(G,H);
1100             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1101             VV               = _mm_macc_pd(vfeps,Fp,Y);
1102             velec            = _mm_mul_pd(qq20,VV);
1103             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1104             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1105
1106             /* Update potential sum for this i atom from the interaction with this j atom. */
1107             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1108             velecsum         = _mm_add_pd(velecsum,velec);
1109
1110             fscal            = felec;
1111
1112             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1113
1114             /* Update vectorial force */
1115             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1116             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1117             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1118
1119             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1120             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1121             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1122
1123             /**************************
1124              * CALCULATE INTERACTIONS *
1125              **************************/
1126
1127             r21              = _mm_mul_pd(rsq21,rinv21);
1128
1129             /* Calculate table index by multiplying r with table scale and truncate to integer */
1130             rt               = _mm_mul_pd(r21,vftabscale);
1131             vfitab           = _mm_cvttpd_epi32(rt);
1132 #ifdef __XOP__
1133             vfeps            = _mm_frcz_pd(rt);
1134 #else
1135             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1136 #endif
1137             twovfeps         = _mm_add_pd(vfeps,vfeps);
1138             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1139
1140             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1141             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1142             F                = _mm_setzero_pd();
1143             GMX_MM_TRANSPOSE2_PD(Y,F);
1144             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1145             H                = _mm_setzero_pd();
1146             GMX_MM_TRANSPOSE2_PD(G,H);
1147             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1148             VV               = _mm_macc_pd(vfeps,Fp,Y);
1149             velec            = _mm_mul_pd(qq21,VV);
1150             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1151             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1152
1153             /* Update potential sum for this i atom from the interaction with this j atom. */
1154             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1155             velecsum         = _mm_add_pd(velecsum,velec);
1156
1157             fscal            = felec;
1158
1159             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1160
1161             /* Update vectorial force */
1162             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1163             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1164             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1165
1166             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1167             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1168             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             r22              = _mm_mul_pd(rsq22,rinv22);
1175
1176             /* Calculate table index by multiplying r with table scale and truncate to integer */
1177             rt               = _mm_mul_pd(r22,vftabscale);
1178             vfitab           = _mm_cvttpd_epi32(rt);
1179 #ifdef __XOP__
1180             vfeps            = _mm_frcz_pd(rt);
1181 #else
1182             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1183 #endif
1184             twovfeps         = _mm_add_pd(vfeps,vfeps);
1185             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1186
1187             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1188             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1189             F                = _mm_setzero_pd();
1190             GMX_MM_TRANSPOSE2_PD(Y,F);
1191             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1192             H                = _mm_setzero_pd();
1193             GMX_MM_TRANSPOSE2_PD(G,H);
1194             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1195             VV               = _mm_macc_pd(vfeps,Fp,Y);
1196             velec            = _mm_mul_pd(qq22,VV);
1197             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1198             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1199
1200             /* Update potential sum for this i atom from the interaction with this j atom. */
1201             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1202             velecsum         = _mm_add_pd(velecsum,velec);
1203
1204             fscal            = felec;
1205
1206             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1207
1208             /* Update vectorial force */
1209             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1210             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1211             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1212
1213             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1214             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1215             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1216
1217             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1218
1219             /* Inner loop uses 444 flops */
1220         }
1221
1222         /* End of innermost loop */
1223
1224         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1225                                               f+i_coord_offset,fshift+i_shift_offset);
1226
1227         ggid                        = gid[iidx];
1228         /* Update potential energies */
1229         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1230         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1231
1232         /* Increment number of inner iterations */
1233         inneriter                  += j_index_end - j_index_start;
1234
1235         /* Outer loop uses 20 flops */
1236     }
1237
1238     /* Increment number of outer iterations */
1239     outeriter        += nri;
1240
1241     /* Update outer/inner flops */
1242
1243     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*444);
1244 }
1245 /*
1246  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_F_avx_128_fma_double
1247  * Electrostatics interaction: CubicSplineTable
1248  * VdW interaction:            CubicSplineTable
1249  * Geometry:                   Water3-Water3
1250  * Calculate force/pot:        Force
1251  */
1252 void
1253 nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_F_avx_128_fma_double
1254                     (t_nblist * gmx_restrict                nlist,
1255                      rvec * gmx_restrict                    xx,
1256                      rvec * gmx_restrict                    ff,
1257                      t_forcerec * gmx_restrict              fr,
1258                      t_mdatoms * gmx_restrict               mdatoms,
1259                      nb_kernel_data_t * gmx_restrict        kernel_data,
1260                      t_nrnb * gmx_restrict                  nrnb)
1261 {
1262     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1263      * just 0 for non-waters.
1264      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1265      * jnr indices corresponding to data put in the four positions in the SIMD register.
1266      */
1267     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1268     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1269     int              jnrA,jnrB;
1270     int              j_coord_offsetA,j_coord_offsetB;
1271     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1272     real             rcutoff_scalar;
1273     real             *shiftvec,*fshift,*x,*f;
1274     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1275     int              vdwioffset0;
1276     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1277     int              vdwioffset1;
1278     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1279     int              vdwioffset2;
1280     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1281     int              vdwjidx0A,vdwjidx0B;
1282     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1283     int              vdwjidx1A,vdwjidx1B;
1284     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1285     int              vdwjidx2A,vdwjidx2B;
1286     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1287     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1288     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1289     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1290     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1291     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1292     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1293     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1294     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1295     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1296     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1297     real             *charge;
1298     int              nvdwtype;
1299     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1300     int              *vdwtype;
1301     real             *vdwparam;
1302     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1303     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1304     __m128i          vfitab;
1305     __m128i          ifour       = _mm_set1_epi32(4);
1306     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
1307     real             *vftab;
1308     __m128d          dummy_mask,cutoff_mask;
1309     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1310     __m128d          one     = _mm_set1_pd(1.0);
1311     __m128d          two     = _mm_set1_pd(2.0);
1312     x                = xx[0];
1313     f                = ff[0];
1314
1315     nri              = nlist->nri;
1316     iinr             = nlist->iinr;
1317     jindex           = nlist->jindex;
1318     jjnr             = nlist->jjnr;
1319     shiftidx         = nlist->shift;
1320     gid              = nlist->gid;
1321     shiftvec         = fr->shift_vec[0];
1322     fshift           = fr->fshift[0];
1323     facel            = _mm_set1_pd(fr->epsfac);
1324     charge           = mdatoms->chargeA;
1325     nvdwtype         = fr->ntype;
1326     vdwparam         = fr->nbfp;
1327     vdwtype          = mdatoms->typeA;
1328
1329     vftab            = kernel_data->table_elec_vdw->data;
1330     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
1331
1332     /* Setup water-specific parameters */
1333     inr              = nlist->iinr[0];
1334     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1335     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1336     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1337     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1338
1339     jq0              = _mm_set1_pd(charge[inr+0]);
1340     jq1              = _mm_set1_pd(charge[inr+1]);
1341     jq2              = _mm_set1_pd(charge[inr+2]);
1342     vdwjidx0A        = 2*vdwtype[inr+0];
1343     qq00             = _mm_mul_pd(iq0,jq0);
1344     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1345     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1346     qq01             = _mm_mul_pd(iq0,jq1);
1347     qq02             = _mm_mul_pd(iq0,jq2);
1348     qq10             = _mm_mul_pd(iq1,jq0);
1349     qq11             = _mm_mul_pd(iq1,jq1);
1350     qq12             = _mm_mul_pd(iq1,jq2);
1351     qq20             = _mm_mul_pd(iq2,jq0);
1352     qq21             = _mm_mul_pd(iq2,jq1);
1353     qq22             = _mm_mul_pd(iq2,jq2);
1354
1355     /* Avoid stupid compiler warnings */
1356     jnrA = jnrB = 0;
1357     j_coord_offsetA = 0;
1358     j_coord_offsetB = 0;
1359
1360     outeriter        = 0;
1361     inneriter        = 0;
1362
1363     /* Start outer loop over neighborlists */
1364     for(iidx=0; iidx<nri; iidx++)
1365     {
1366         /* Load shift vector for this list */
1367         i_shift_offset   = DIM*shiftidx[iidx];
1368
1369         /* Load limits for loop over neighbors */
1370         j_index_start    = jindex[iidx];
1371         j_index_end      = jindex[iidx+1];
1372
1373         /* Get outer coordinate index */
1374         inr              = iinr[iidx];
1375         i_coord_offset   = DIM*inr;
1376
1377         /* Load i particle coords and add shift vector */
1378         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1379                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1380
1381         fix0             = _mm_setzero_pd();
1382         fiy0             = _mm_setzero_pd();
1383         fiz0             = _mm_setzero_pd();
1384         fix1             = _mm_setzero_pd();
1385         fiy1             = _mm_setzero_pd();
1386         fiz1             = _mm_setzero_pd();
1387         fix2             = _mm_setzero_pd();
1388         fiy2             = _mm_setzero_pd();
1389         fiz2             = _mm_setzero_pd();
1390
1391         /* Start inner kernel loop */
1392         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1393         {
1394
1395             /* Get j neighbor index, and coordinate index */
1396             jnrA             = jjnr[jidx];
1397             jnrB             = jjnr[jidx+1];
1398             j_coord_offsetA  = DIM*jnrA;
1399             j_coord_offsetB  = DIM*jnrB;
1400
1401             /* load j atom coordinates */
1402             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1403                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1404
1405             /* Calculate displacement vector */
1406             dx00             = _mm_sub_pd(ix0,jx0);
1407             dy00             = _mm_sub_pd(iy0,jy0);
1408             dz00             = _mm_sub_pd(iz0,jz0);
1409             dx01             = _mm_sub_pd(ix0,jx1);
1410             dy01             = _mm_sub_pd(iy0,jy1);
1411             dz01             = _mm_sub_pd(iz0,jz1);
1412             dx02             = _mm_sub_pd(ix0,jx2);
1413             dy02             = _mm_sub_pd(iy0,jy2);
1414             dz02             = _mm_sub_pd(iz0,jz2);
1415             dx10             = _mm_sub_pd(ix1,jx0);
1416             dy10             = _mm_sub_pd(iy1,jy0);
1417             dz10             = _mm_sub_pd(iz1,jz0);
1418             dx11             = _mm_sub_pd(ix1,jx1);
1419             dy11             = _mm_sub_pd(iy1,jy1);
1420             dz11             = _mm_sub_pd(iz1,jz1);
1421             dx12             = _mm_sub_pd(ix1,jx2);
1422             dy12             = _mm_sub_pd(iy1,jy2);
1423             dz12             = _mm_sub_pd(iz1,jz2);
1424             dx20             = _mm_sub_pd(ix2,jx0);
1425             dy20             = _mm_sub_pd(iy2,jy0);
1426             dz20             = _mm_sub_pd(iz2,jz0);
1427             dx21             = _mm_sub_pd(ix2,jx1);
1428             dy21             = _mm_sub_pd(iy2,jy1);
1429             dz21             = _mm_sub_pd(iz2,jz1);
1430             dx22             = _mm_sub_pd(ix2,jx2);
1431             dy22             = _mm_sub_pd(iy2,jy2);
1432             dz22             = _mm_sub_pd(iz2,jz2);
1433
1434             /* Calculate squared distance and things based on it */
1435             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1436             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1437             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1438             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1439             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1440             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1441             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1442             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1443             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1444
1445             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1446             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1447             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1448             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1449             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1450             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1451             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1452             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1453             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1454
1455             fjx0             = _mm_setzero_pd();
1456             fjy0             = _mm_setzero_pd();
1457             fjz0             = _mm_setzero_pd();
1458             fjx1             = _mm_setzero_pd();
1459             fjy1             = _mm_setzero_pd();
1460             fjz1             = _mm_setzero_pd();
1461             fjx2             = _mm_setzero_pd();
1462             fjy2             = _mm_setzero_pd();
1463             fjz2             = _mm_setzero_pd();
1464
1465             /**************************
1466              * CALCULATE INTERACTIONS *
1467              **************************/
1468
1469             r00              = _mm_mul_pd(rsq00,rinv00);
1470
1471             /* Calculate table index by multiplying r with table scale and truncate to integer */
1472             rt               = _mm_mul_pd(r00,vftabscale);
1473             vfitab           = _mm_cvttpd_epi32(rt);
1474 #ifdef __XOP__
1475             vfeps            = _mm_frcz_pd(rt);
1476 #else
1477             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1478 #endif
1479             twovfeps         = _mm_add_pd(vfeps,vfeps);
1480             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1481
1482             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1483             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1484             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1485             GMX_MM_TRANSPOSE2_PD(Y,F);
1486             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1487             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1488             GMX_MM_TRANSPOSE2_PD(G,H);
1489             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1490             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1491             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1492
1493             /* CUBIC SPLINE TABLE DISPERSION */
1494             vfitab           = _mm_add_epi32(vfitab,ifour);
1495             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1496             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1497             GMX_MM_TRANSPOSE2_PD(Y,F);
1498             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1499             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1500             GMX_MM_TRANSPOSE2_PD(G,H);
1501             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1502             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1503             fvdw6            = _mm_mul_pd(c6_00,FF);
1504
1505             /* CUBIC SPLINE TABLE REPULSION */
1506             vfitab           = _mm_add_epi32(vfitab,ifour);
1507             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1508             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1509             GMX_MM_TRANSPOSE2_PD(Y,F);
1510             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1511             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1512             GMX_MM_TRANSPOSE2_PD(G,H);
1513             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1514             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1515             fvdw12           = _mm_mul_pd(c12_00,FF);
1516             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1517
1518             fscal            = _mm_add_pd(felec,fvdw);
1519
1520             /* Update vectorial force */
1521             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1522             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1523             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1524
1525             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1526             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1527             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1528
1529             /**************************
1530              * CALCULATE INTERACTIONS *
1531              **************************/
1532
1533             r01              = _mm_mul_pd(rsq01,rinv01);
1534
1535             /* Calculate table index by multiplying r with table scale and truncate to integer */
1536             rt               = _mm_mul_pd(r01,vftabscale);
1537             vfitab           = _mm_cvttpd_epi32(rt);
1538 #ifdef __XOP__
1539             vfeps            = _mm_frcz_pd(rt);
1540 #else
1541             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1542 #endif
1543             twovfeps         = _mm_add_pd(vfeps,vfeps);
1544             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1545
1546             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1547             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1548             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1549             GMX_MM_TRANSPOSE2_PD(Y,F);
1550             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1551             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1552             GMX_MM_TRANSPOSE2_PD(G,H);
1553             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1554             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1555             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
1556
1557             fscal            = felec;
1558
1559             /* Update vectorial force */
1560             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1561             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1562             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1563
1564             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1565             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1566             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1567
1568             /**************************
1569              * CALCULATE INTERACTIONS *
1570              **************************/
1571
1572             r02              = _mm_mul_pd(rsq02,rinv02);
1573
1574             /* Calculate table index by multiplying r with table scale and truncate to integer */
1575             rt               = _mm_mul_pd(r02,vftabscale);
1576             vfitab           = _mm_cvttpd_epi32(rt);
1577 #ifdef __XOP__
1578             vfeps            = _mm_frcz_pd(rt);
1579 #else
1580             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1581 #endif
1582             twovfeps         = _mm_add_pd(vfeps,vfeps);
1583             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1584
1585             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1586             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1587             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1588             GMX_MM_TRANSPOSE2_PD(Y,F);
1589             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1590             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1591             GMX_MM_TRANSPOSE2_PD(G,H);
1592             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1593             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1594             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
1595
1596             fscal            = felec;
1597
1598             /* Update vectorial force */
1599             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1600             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1601             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1602
1603             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1604             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1605             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1606
1607             /**************************
1608              * CALCULATE INTERACTIONS *
1609              **************************/
1610
1611             r10              = _mm_mul_pd(rsq10,rinv10);
1612
1613             /* Calculate table index by multiplying r with table scale and truncate to integer */
1614             rt               = _mm_mul_pd(r10,vftabscale);
1615             vfitab           = _mm_cvttpd_epi32(rt);
1616 #ifdef __XOP__
1617             vfeps            = _mm_frcz_pd(rt);
1618 #else
1619             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1620 #endif
1621             twovfeps         = _mm_add_pd(vfeps,vfeps);
1622             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1623
1624             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1625             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1626             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1627             GMX_MM_TRANSPOSE2_PD(Y,F);
1628             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1629             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1630             GMX_MM_TRANSPOSE2_PD(G,H);
1631             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1632             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1633             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1634
1635             fscal            = felec;
1636
1637             /* Update vectorial force */
1638             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1639             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1640             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1641
1642             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1643             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1644             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1645
1646             /**************************
1647              * CALCULATE INTERACTIONS *
1648              **************************/
1649
1650             r11              = _mm_mul_pd(rsq11,rinv11);
1651
1652             /* Calculate table index by multiplying r with table scale and truncate to integer */
1653             rt               = _mm_mul_pd(r11,vftabscale);
1654             vfitab           = _mm_cvttpd_epi32(rt);
1655 #ifdef __XOP__
1656             vfeps            = _mm_frcz_pd(rt);
1657 #else
1658             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1659 #endif
1660             twovfeps         = _mm_add_pd(vfeps,vfeps);
1661             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1662
1663             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1664             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1665             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1666             GMX_MM_TRANSPOSE2_PD(Y,F);
1667             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1668             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1669             GMX_MM_TRANSPOSE2_PD(G,H);
1670             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1671             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1672             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
1673
1674             fscal            = felec;
1675
1676             /* Update vectorial force */
1677             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1678             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1679             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1680
1681             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1682             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1683             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1684
1685             /**************************
1686              * CALCULATE INTERACTIONS *
1687              **************************/
1688
1689             r12              = _mm_mul_pd(rsq12,rinv12);
1690
1691             /* Calculate table index by multiplying r with table scale and truncate to integer */
1692             rt               = _mm_mul_pd(r12,vftabscale);
1693             vfitab           = _mm_cvttpd_epi32(rt);
1694 #ifdef __XOP__
1695             vfeps            = _mm_frcz_pd(rt);
1696 #else
1697             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1698 #endif
1699             twovfeps         = _mm_add_pd(vfeps,vfeps);
1700             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1701
1702             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1703             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1704             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1705             GMX_MM_TRANSPOSE2_PD(Y,F);
1706             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1707             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1708             GMX_MM_TRANSPOSE2_PD(G,H);
1709             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1710             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1711             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
1712
1713             fscal            = felec;
1714
1715             /* Update vectorial force */
1716             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1717             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1718             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1719
1720             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1721             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1722             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1723
1724             /**************************
1725              * CALCULATE INTERACTIONS *
1726              **************************/
1727
1728             r20              = _mm_mul_pd(rsq20,rinv20);
1729
1730             /* Calculate table index by multiplying r with table scale and truncate to integer */
1731             rt               = _mm_mul_pd(r20,vftabscale);
1732             vfitab           = _mm_cvttpd_epi32(rt);
1733 #ifdef __XOP__
1734             vfeps            = _mm_frcz_pd(rt);
1735 #else
1736             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1737 #endif
1738             twovfeps         = _mm_add_pd(vfeps,vfeps);
1739             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1740
1741             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1742             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1743             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1744             GMX_MM_TRANSPOSE2_PD(Y,F);
1745             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1746             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1747             GMX_MM_TRANSPOSE2_PD(G,H);
1748             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1749             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1750             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1751
1752             fscal            = felec;
1753
1754             /* Update vectorial force */
1755             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1756             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1757             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1758
1759             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1760             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1761             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1762
1763             /**************************
1764              * CALCULATE INTERACTIONS *
1765              **************************/
1766
1767             r21              = _mm_mul_pd(rsq21,rinv21);
1768
1769             /* Calculate table index by multiplying r with table scale and truncate to integer */
1770             rt               = _mm_mul_pd(r21,vftabscale);
1771             vfitab           = _mm_cvttpd_epi32(rt);
1772 #ifdef __XOP__
1773             vfeps            = _mm_frcz_pd(rt);
1774 #else
1775             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1776 #endif
1777             twovfeps         = _mm_add_pd(vfeps,vfeps);
1778             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1779
1780             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1781             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1782             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1783             GMX_MM_TRANSPOSE2_PD(Y,F);
1784             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1785             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1786             GMX_MM_TRANSPOSE2_PD(G,H);
1787             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1788             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1789             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1790
1791             fscal            = felec;
1792
1793             /* Update vectorial force */
1794             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1795             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1796             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1797
1798             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1799             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1800             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1801
1802             /**************************
1803              * CALCULATE INTERACTIONS *
1804              **************************/
1805
1806             r22              = _mm_mul_pd(rsq22,rinv22);
1807
1808             /* Calculate table index by multiplying r with table scale and truncate to integer */
1809             rt               = _mm_mul_pd(r22,vftabscale);
1810             vfitab           = _mm_cvttpd_epi32(rt);
1811 #ifdef __XOP__
1812             vfeps            = _mm_frcz_pd(rt);
1813 #else
1814             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1815 #endif
1816             twovfeps         = _mm_add_pd(vfeps,vfeps);
1817             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1818
1819             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1820             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1821             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1822             GMX_MM_TRANSPOSE2_PD(Y,F);
1823             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1824             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1825             GMX_MM_TRANSPOSE2_PD(G,H);
1826             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1827             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1828             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1829
1830             fscal            = felec;
1831
1832             /* Update vectorial force */
1833             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1834             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1835             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1836
1837             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1838             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1839             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1840
1841             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1842
1843             /* Inner loop uses 400 flops */
1844         }
1845
1846         if(jidx<j_index_end)
1847         {
1848
1849             jnrA             = jjnr[jidx];
1850             j_coord_offsetA  = DIM*jnrA;
1851
1852             /* load j atom coordinates */
1853             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1854                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1855
1856             /* Calculate displacement vector */
1857             dx00             = _mm_sub_pd(ix0,jx0);
1858             dy00             = _mm_sub_pd(iy0,jy0);
1859             dz00             = _mm_sub_pd(iz0,jz0);
1860             dx01             = _mm_sub_pd(ix0,jx1);
1861             dy01             = _mm_sub_pd(iy0,jy1);
1862             dz01             = _mm_sub_pd(iz0,jz1);
1863             dx02             = _mm_sub_pd(ix0,jx2);
1864             dy02             = _mm_sub_pd(iy0,jy2);
1865             dz02             = _mm_sub_pd(iz0,jz2);
1866             dx10             = _mm_sub_pd(ix1,jx0);
1867             dy10             = _mm_sub_pd(iy1,jy0);
1868             dz10             = _mm_sub_pd(iz1,jz0);
1869             dx11             = _mm_sub_pd(ix1,jx1);
1870             dy11             = _mm_sub_pd(iy1,jy1);
1871             dz11             = _mm_sub_pd(iz1,jz1);
1872             dx12             = _mm_sub_pd(ix1,jx2);
1873             dy12             = _mm_sub_pd(iy1,jy2);
1874             dz12             = _mm_sub_pd(iz1,jz2);
1875             dx20             = _mm_sub_pd(ix2,jx0);
1876             dy20             = _mm_sub_pd(iy2,jy0);
1877             dz20             = _mm_sub_pd(iz2,jz0);
1878             dx21             = _mm_sub_pd(ix2,jx1);
1879             dy21             = _mm_sub_pd(iy2,jy1);
1880             dz21             = _mm_sub_pd(iz2,jz1);
1881             dx22             = _mm_sub_pd(ix2,jx2);
1882             dy22             = _mm_sub_pd(iy2,jy2);
1883             dz22             = _mm_sub_pd(iz2,jz2);
1884
1885             /* Calculate squared distance and things based on it */
1886             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1887             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1888             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1889             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1890             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1891             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1892             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1893             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1894             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1895
1896             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1897             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1898             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1899             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1900             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1901             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1902             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1903             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1904             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1905
1906             fjx0             = _mm_setzero_pd();
1907             fjy0             = _mm_setzero_pd();
1908             fjz0             = _mm_setzero_pd();
1909             fjx1             = _mm_setzero_pd();
1910             fjy1             = _mm_setzero_pd();
1911             fjz1             = _mm_setzero_pd();
1912             fjx2             = _mm_setzero_pd();
1913             fjy2             = _mm_setzero_pd();
1914             fjz2             = _mm_setzero_pd();
1915
1916             /**************************
1917              * CALCULATE INTERACTIONS *
1918              **************************/
1919
1920             r00              = _mm_mul_pd(rsq00,rinv00);
1921
1922             /* Calculate table index by multiplying r with table scale and truncate to integer */
1923             rt               = _mm_mul_pd(r00,vftabscale);
1924             vfitab           = _mm_cvttpd_epi32(rt);
1925 #ifdef __XOP__
1926             vfeps            = _mm_frcz_pd(rt);
1927 #else
1928             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1929 #endif
1930             twovfeps         = _mm_add_pd(vfeps,vfeps);
1931             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1932
1933             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1934             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1935             F                = _mm_setzero_pd();
1936             GMX_MM_TRANSPOSE2_PD(Y,F);
1937             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1938             H                = _mm_setzero_pd();
1939             GMX_MM_TRANSPOSE2_PD(G,H);
1940             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1941             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1942             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1943
1944             /* CUBIC SPLINE TABLE DISPERSION */
1945             vfitab           = _mm_add_epi32(vfitab,ifour);
1946             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1947             F                = _mm_setzero_pd();
1948             GMX_MM_TRANSPOSE2_PD(Y,F);
1949             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1950             H                = _mm_setzero_pd();
1951             GMX_MM_TRANSPOSE2_PD(G,H);
1952             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1953             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1954             fvdw6            = _mm_mul_pd(c6_00,FF);
1955
1956             /* CUBIC SPLINE TABLE REPULSION */
1957             vfitab           = _mm_add_epi32(vfitab,ifour);
1958             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1959             F                = _mm_setzero_pd();
1960             GMX_MM_TRANSPOSE2_PD(Y,F);
1961             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1962             H                = _mm_setzero_pd();
1963             GMX_MM_TRANSPOSE2_PD(G,H);
1964             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1965             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1966             fvdw12           = _mm_mul_pd(c12_00,FF);
1967             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1968
1969             fscal            = _mm_add_pd(felec,fvdw);
1970
1971             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1972
1973             /* Update vectorial force */
1974             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1975             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1976             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1977
1978             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1979             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1980             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1981
1982             /**************************
1983              * CALCULATE INTERACTIONS *
1984              **************************/
1985
1986             r01              = _mm_mul_pd(rsq01,rinv01);
1987
1988             /* Calculate table index by multiplying r with table scale and truncate to integer */
1989             rt               = _mm_mul_pd(r01,vftabscale);
1990             vfitab           = _mm_cvttpd_epi32(rt);
1991 #ifdef __XOP__
1992             vfeps            = _mm_frcz_pd(rt);
1993 #else
1994             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1995 #endif
1996             twovfeps         = _mm_add_pd(vfeps,vfeps);
1997             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1998
1999             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2000             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2001             F                = _mm_setzero_pd();
2002             GMX_MM_TRANSPOSE2_PD(Y,F);
2003             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2004             H                = _mm_setzero_pd();
2005             GMX_MM_TRANSPOSE2_PD(G,H);
2006             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2007             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2008             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
2009
2010             fscal            = felec;
2011
2012             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2013
2014             /* Update vectorial force */
2015             fix0             = _mm_macc_pd(dx01,fscal,fix0);
2016             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
2017             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
2018
2019             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
2020             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
2021             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
2022
2023             /**************************
2024              * CALCULATE INTERACTIONS *
2025              **************************/
2026
2027             r02              = _mm_mul_pd(rsq02,rinv02);
2028
2029             /* Calculate table index by multiplying r with table scale and truncate to integer */
2030             rt               = _mm_mul_pd(r02,vftabscale);
2031             vfitab           = _mm_cvttpd_epi32(rt);
2032 #ifdef __XOP__
2033             vfeps            = _mm_frcz_pd(rt);
2034 #else
2035             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2036 #endif
2037             twovfeps         = _mm_add_pd(vfeps,vfeps);
2038             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2039
2040             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2041             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2042             F                = _mm_setzero_pd();
2043             GMX_MM_TRANSPOSE2_PD(Y,F);
2044             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2045             H                = _mm_setzero_pd();
2046             GMX_MM_TRANSPOSE2_PD(G,H);
2047             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2048             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2049             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
2050
2051             fscal            = felec;
2052
2053             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2054
2055             /* Update vectorial force */
2056             fix0             = _mm_macc_pd(dx02,fscal,fix0);
2057             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
2058             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
2059
2060             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
2061             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
2062             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
2063
2064             /**************************
2065              * CALCULATE INTERACTIONS *
2066              **************************/
2067
2068             r10              = _mm_mul_pd(rsq10,rinv10);
2069
2070             /* Calculate table index by multiplying r with table scale and truncate to integer */
2071             rt               = _mm_mul_pd(r10,vftabscale);
2072             vfitab           = _mm_cvttpd_epi32(rt);
2073 #ifdef __XOP__
2074             vfeps            = _mm_frcz_pd(rt);
2075 #else
2076             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2077 #endif
2078             twovfeps         = _mm_add_pd(vfeps,vfeps);
2079             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2080
2081             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2082             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2083             F                = _mm_setzero_pd();
2084             GMX_MM_TRANSPOSE2_PD(Y,F);
2085             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2086             H                = _mm_setzero_pd();
2087             GMX_MM_TRANSPOSE2_PD(G,H);
2088             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2089             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2090             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
2091
2092             fscal            = felec;
2093
2094             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2095
2096             /* Update vectorial force */
2097             fix1             = _mm_macc_pd(dx10,fscal,fix1);
2098             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
2099             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
2100
2101             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
2102             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
2103             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
2104
2105             /**************************
2106              * CALCULATE INTERACTIONS *
2107              **************************/
2108
2109             r11              = _mm_mul_pd(rsq11,rinv11);
2110
2111             /* Calculate table index by multiplying r with table scale and truncate to integer */
2112             rt               = _mm_mul_pd(r11,vftabscale);
2113             vfitab           = _mm_cvttpd_epi32(rt);
2114 #ifdef __XOP__
2115             vfeps            = _mm_frcz_pd(rt);
2116 #else
2117             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2118 #endif
2119             twovfeps         = _mm_add_pd(vfeps,vfeps);
2120             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2121
2122             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2123             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2124             F                = _mm_setzero_pd();
2125             GMX_MM_TRANSPOSE2_PD(Y,F);
2126             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2127             H                = _mm_setzero_pd();
2128             GMX_MM_TRANSPOSE2_PD(G,H);
2129             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2130             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2131             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
2132
2133             fscal            = felec;
2134
2135             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2136
2137             /* Update vectorial force */
2138             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2139             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2140             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2141
2142             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2143             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2144             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2145
2146             /**************************
2147              * CALCULATE INTERACTIONS *
2148              **************************/
2149
2150             r12              = _mm_mul_pd(rsq12,rinv12);
2151
2152             /* Calculate table index by multiplying r with table scale and truncate to integer */
2153             rt               = _mm_mul_pd(r12,vftabscale);
2154             vfitab           = _mm_cvttpd_epi32(rt);
2155 #ifdef __XOP__
2156             vfeps            = _mm_frcz_pd(rt);
2157 #else
2158             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2159 #endif
2160             twovfeps         = _mm_add_pd(vfeps,vfeps);
2161             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2162
2163             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2164             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2165             F                = _mm_setzero_pd();
2166             GMX_MM_TRANSPOSE2_PD(Y,F);
2167             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2168             H                = _mm_setzero_pd();
2169             GMX_MM_TRANSPOSE2_PD(G,H);
2170             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2171             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2172             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
2173
2174             fscal            = felec;
2175
2176             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2177
2178             /* Update vectorial force */
2179             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2180             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2181             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2182
2183             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2184             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2185             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2186
2187             /**************************
2188              * CALCULATE INTERACTIONS *
2189              **************************/
2190
2191             r20              = _mm_mul_pd(rsq20,rinv20);
2192
2193             /* Calculate table index by multiplying r with table scale and truncate to integer */
2194             rt               = _mm_mul_pd(r20,vftabscale);
2195             vfitab           = _mm_cvttpd_epi32(rt);
2196 #ifdef __XOP__
2197             vfeps            = _mm_frcz_pd(rt);
2198 #else
2199             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2200 #endif
2201             twovfeps         = _mm_add_pd(vfeps,vfeps);
2202             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2203
2204             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2205             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2206             F                = _mm_setzero_pd();
2207             GMX_MM_TRANSPOSE2_PD(Y,F);
2208             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2209             H                = _mm_setzero_pd();
2210             GMX_MM_TRANSPOSE2_PD(G,H);
2211             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2212             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2213             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
2214
2215             fscal            = felec;
2216
2217             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2218
2219             /* Update vectorial force */
2220             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2221             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2222             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2223
2224             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2225             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2226             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2227
2228             /**************************
2229              * CALCULATE INTERACTIONS *
2230              **************************/
2231
2232             r21              = _mm_mul_pd(rsq21,rinv21);
2233
2234             /* Calculate table index by multiplying r with table scale and truncate to integer */
2235             rt               = _mm_mul_pd(r21,vftabscale);
2236             vfitab           = _mm_cvttpd_epi32(rt);
2237 #ifdef __XOP__
2238             vfeps            = _mm_frcz_pd(rt);
2239 #else
2240             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2241 #endif
2242             twovfeps         = _mm_add_pd(vfeps,vfeps);
2243             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2244
2245             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2246             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2247             F                = _mm_setzero_pd();
2248             GMX_MM_TRANSPOSE2_PD(Y,F);
2249             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2250             H                = _mm_setzero_pd();
2251             GMX_MM_TRANSPOSE2_PD(G,H);
2252             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2253             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2254             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
2255
2256             fscal            = felec;
2257
2258             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2259
2260             /* Update vectorial force */
2261             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2262             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2263             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2264
2265             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2266             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2267             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2268
2269             /**************************
2270              * CALCULATE INTERACTIONS *
2271              **************************/
2272
2273             r22              = _mm_mul_pd(rsq22,rinv22);
2274
2275             /* Calculate table index by multiplying r with table scale and truncate to integer */
2276             rt               = _mm_mul_pd(r22,vftabscale);
2277             vfitab           = _mm_cvttpd_epi32(rt);
2278 #ifdef __XOP__
2279             vfeps            = _mm_frcz_pd(rt);
2280 #else
2281             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2282 #endif
2283             twovfeps         = _mm_add_pd(vfeps,vfeps);
2284             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
2285
2286             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2287             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2288             F                = _mm_setzero_pd();
2289             GMX_MM_TRANSPOSE2_PD(Y,F);
2290             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2291             H                = _mm_setzero_pd();
2292             GMX_MM_TRANSPOSE2_PD(G,H);
2293             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2294             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2295             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
2296
2297             fscal            = felec;
2298
2299             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2300
2301             /* Update vectorial force */
2302             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2303             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2304             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2305
2306             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2307             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2308             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2309
2310             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2311
2312             /* Inner loop uses 400 flops */
2313         }
2314
2315         /* End of innermost loop */
2316
2317         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2318                                               f+i_coord_offset,fshift+i_shift_offset);
2319
2320         /* Increment number of inner iterations */
2321         inneriter                  += j_index_end - j_index_start;
2322
2323         /* Outer loop uses 18 flops */
2324     }
2325
2326     /* Increment number of outer iterations */
2327     outeriter        += nri;
2328
2329     /* Update outer/inner flops */
2330
2331     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*400);
2332 }