src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_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_VdwLJ_GeomW4W4_VF_avx_128_fma_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_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              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     int              vdwjidx1A,vdwjidx1B;
  77     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  78     int              vdwjidx2A,vdwjidx2B;
  79     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  80     int              vdwjidx3A,vdwjidx3B;
  81     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  82     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  84     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  85     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  86     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  87     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  88     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  89     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  90     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  91     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  92     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  93     real             *charge;
  94     int              nvdwtype;
  95     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  96     int              *vdwtype;
  97     real             *vdwparam;
  98     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  99     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 100     __m128i          vfitab;
 101     __m128i          ifour       = _mm_set1_epi32(4);
 102     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 103     real             *vftab;
 104     __m128d          dummy_mask,cutoff_mask;
 105     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 106     __m128d          one     = _mm_set1_pd(1.0);
 107     __m128d          two     = _mm_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     vftab            = kernel_data->table_elec->data;
 126     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 131     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 132     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     jq1              = _mm_set1_pd(charge[inr+1]);
 136     jq2              = _mm_set1_pd(charge[inr+2]);
 137     jq3              = _mm_set1_pd(charge[inr+3]);
 138     vdwjidx0A        = 2*vdwtype[inr+0];
 139     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 140     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 141     qq11             = _mm_mul_pd(iq1,jq1);
 142     qq12             = _mm_mul_pd(iq1,jq2);
 143     qq13             = _mm_mul_pd(iq1,jq3);
 144     qq21             = _mm_mul_pd(iq2,jq1);
 145     qq22             = _mm_mul_pd(iq2,jq2);
 146     qq23             = _mm_mul_pd(iq2,jq3);
 147     qq31             = _mm_mul_pd(iq3,jq1);
 148     qq32             = _mm_mul_pd(iq3,jq2);
 149     qq33             = _mm_mul_pd(iq3,jq3);
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155
 156     outeriter        = 0;
 157     inneriter        = 0;
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 176
 177         fix0             = _mm_setzero_pd();
 178         fiy0             = _mm_setzero_pd();
 179         fiz0             = _mm_setzero_pd();
 180         fix1             = _mm_setzero_pd();
 181         fiy1             = _mm_setzero_pd();
 182         fiz1             = _mm_setzero_pd();
 183         fix2             = _mm_setzero_pd();
 184         fiy2             = _mm_setzero_pd();
 185         fiz2             = _mm_setzero_pd();
 186         fix3             = _mm_setzero_pd();
 187         fiy3             = _mm_setzero_pd();
 188         fiz3             = _mm_setzero_pd();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm_setzero_pd();
 192         vvdwsum          = _mm_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 207                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_pd(ix0,jx0);
 211             dy00             = _mm_sub_pd(iy0,jy0);
 212             dz00             = _mm_sub_pd(iz0,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             dx13             = _mm_sub_pd(ix1,jx3);
 220             dy13             = _mm_sub_pd(iy1,jy3);
 221             dz13             = _mm_sub_pd(iz1,jz3);
 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             dx23             = _mm_sub_pd(ix2,jx3);
 229             dy23             = _mm_sub_pd(iy2,jy3);
 230             dz23             = _mm_sub_pd(iz2,jz3);
 231             dx31             = _mm_sub_pd(ix3,jx1);
 232             dy31             = _mm_sub_pd(iy3,jy1);
 233             dz31             = _mm_sub_pd(iz3,jz1);
 234             dx32             = _mm_sub_pd(ix3,jx2);
 235             dy32             = _mm_sub_pd(iy3,jy2);
 236             dz32             = _mm_sub_pd(iz3,jz2);
 237             dx33             = _mm_sub_pd(ix3,jx3);
 238             dy33             = _mm_sub_pd(iy3,jy3);
 239             dz33             = _mm_sub_pd(iz3,jz3);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 243             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 244             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 245             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 246             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 247             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 248             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 249             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 250             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 251             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 252
 253             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 254             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 255             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 256             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 257             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 258             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 259             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 260             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 261             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 262
 263             rinvsq00         = gmx_mm_inv_pd(rsq00);
 264
 265             fjx0             = _mm_setzero_pd();
 266             fjy0             = _mm_setzero_pd();
 267             fjz0             = _mm_setzero_pd();
 268             fjx1             = _mm_setzero_pd();
 269             fjy1             = _mm_setzero_pd();
 270             fjz1             = _mm_setzero_pd();
 271             fjx2             = _mm_setzero_pd();
 272             fjy2             = _mm_setzero_pd();
 273             fjz2             = _mm_setzero_pd();
 274             fjx3             = _mm_setzero_pd();
 275             fjy3             = _mm_setzero_pd();
 276             fjz3             = _mm_setzero_pd();
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             /* LENNARD-JONES DISPERSION/REPULSION */
 283
 284             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 285             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 286             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 287             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 288             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 292
 293             fscal            = fvdw;
 294
 295             /* Update vectorial force */
 296             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 297             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 298             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 299
 300             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 301             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 302             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             r11              = _mm_mul_pd(rsq11,rinv11);
 309
 310             /* Calculate table index by multiplying r with table scale and truncate to integer */
 311             rt               = _mm_mul_pd(r11,vftabscale);
 312             vfitab           = _mm_cvttpd_epi32(rt);
 313 #ifdef __XOP__
 314             vfeps            = _mm_frcz_pd(rt);
 315 #else
 316             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 317 #endif
 318             twovfeps         = _mm_add_pd(vfeps,vfeps);
 319             vfitab           = _mm_slli_epi32(vfitab,2);
 320
 321             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 322             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 323             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 324             GMX_MM_TRANSPOSE2_PD(Y,F);
 325             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 326             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 327             GMX_MM_TRANSPOSE2_PD(G,H);
 328             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 329             VV               = _mm_macc_pd(vfeps,Fp,Y);
 330             velec            = _mm_mul_pd(qq11,VV);
 331             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 332             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velecsum         = _mm_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             /* Update vectorial force */
 340             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 341             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 342             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 343
 344             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 345             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 346             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             r12              = _mm_mul_pd(rsq12,rinv12);
 353
 354             /* Calculate table index by multiplying r with table scale and truncate to integer */
 355             rt               = _mm_mul_pd(r12,vftabscale);
 356             vfitab           = _mm_cvttpd_epi32(rt);
 357 #ifdef __XOP__
 358             vfeps            = _mm_frcz_pd(rt);
 359 #else
 360             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 361 #endif
 362             twovfeps         = _mm_add_pd(vfeps,vfeps);
 363             vfitab           = _mm_slli_epi32(vfitab,2);
 364
 365             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 366             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 367             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 368             GMX_MM_TRANSPOSE2_PD(Y,F);
 369             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 370             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 371             GMX_MM_TRANSPOSE2_PD(G,H);
 372             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 373             VV               = _mm_macc_pd(vfeps,Fp,Y);
 374             velec            = _mm_mul_pd(qq12,VV);
 375             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 376             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velecsum         = _mm_add_pd(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             /* Update vectorial force */
 384             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 385             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 386             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 387
 388             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 389             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 390             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             r13              = _mm_mul_pd(rsq13,rinv13);
 397
 398             /* Calculate table index by multiplying r with table scale and truncate to integer */
 399             rt               = _mm_mul_pd(r13,vftabscale);
 400             vfitab           = _mm_cvttpd_epi32(rt);
 401 #ifdef __XOP__
 402             vfeps            = _mm_frcz_pd(rt);
 403 #else
 404             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 405 #endif
 406             twovfeps         = _mm_add_pd(vfeps,vfeps);
 407             vfitab           = _mm_slli_epi32(vfitab,2);
 408
 409             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 410             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 411             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 412             GMX_MM_TRANSPOSE2_PD(Y,F);
 413             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 414             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 415             GMX_MM_TRANSPOSE2_PD(G,H);
 416             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 417             VV               = _mm_macc_pd(vfeps,Fp,Y);
 418             velec            = _mm_mul_pd(qq13,VV);
 419             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 420             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velecsum         = _mm_add_pd(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427             /* Update vectorial force */
 428             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 429             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 430             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 431
 432             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 433             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 434             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             r21              = _mm_mul_pd(rsq21,rinv21);
 441
 442             /* Calculate table index by multiplying r with table scale and truncate to integer */
 443             rt               = _mm_mul_pd(r21,vftabscale);
 444             vfitab           = _mm_cvttpd_epi32(rt);
 445 #ifdef __XOP__
 446             vfeps            = _mm_frcz_pd(rt);
 447 #else
 448             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 449 #endif
 450             twovfeps         = _mm_add_pd(vfeps,vfeps);
 451             vfitab           = _mm_slli_epi32(vfitab,2);
 452
 453             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 454             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 455             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 456             GMX_MM_TRANSPOSE2_PD(Y,F);
 457             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 458             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 459             GMX_MM_TRANSPOSE2_PD(G,H);
 460             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 461             VV               = _mm_macc_pd(vfeps,Fp,Y);
 462             velec            = _mm_mul_pd(qq21,VV);
 463             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 464             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             velecsum         = _mm_add_pd(velecsum,velec);
 468
 469             fscal            = felec;
 470
 471             /* Update vectorial force */
 472             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 473             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 474             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 475
 476             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 477             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 478             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r22              = _mm_mul_pd(rsq22,rinv22);
 485
 486             /* Calculate table index by multiplying r with table scale and truncate to integer */
 487             rt               = _mm_mul_pd(r22,vftabscale);
 488             vfitab           = _mm_cvttpd_epi32(rt);
 489 #ifdef __XOP__
 490             vfeps            = _mm_frcz_pd(rt);
 491 #else
 492             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 493 #endif
 494             twovfeps         = _mm_add_pd(vfeps,vfeps);
 495             vfitab           = _mm_slli_epi32(vfitab,2);
 496
 497             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 498             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 499             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 500             GMX_MM_TRANSPOSE2_PD(Y,F);
 501             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 502             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 503             GMX_MM_TRANSPOSE2_PD(G,H);
 504             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 505             VV               = _mm_macc_pd(vfeps,Fp,Y);
 506             velec            = _mm_mul_pd(qq22,VV);
 507             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 508             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             velecsum         = _mm_add_pd(velecsum,velec);
 512
 513             fscal            = felec;
 514
 515             /* Update vectorial force */
 516             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 517             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 518             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 519
 520             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 521             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 522             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r23              = _mm_mul_pd(rsq23,rinv23);
 529
 530             /* Calculate table index by multiplying r with table scale and truncate to integer */
 531             rt               = _mm_mul_pd(r23,vftabscale);
 532             vfitab           = _mm_cvttpd_epi32(rt);
 533 #ifdef __XOP__
 534             vfeps            = _mm_frcz_pd(rt);
 535 #else
 536             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 537 #endif
 538             twovfeps         = _mm_add_pd(vfeps,vfeps);
 539             vfitab           = _mm_slli_epi32(vfitab,2);
 540
 541             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 542             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 543             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 544             GMX_MM_TRANSPOSE2_PD(Y,F);
 545             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 546             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 547             GMX_MM_TRANSPOSE2_PD(G,H);
 548             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 549             VV               = _mm_macc_pd(vfeps,Fp,Y);
 550             velec            = _mm_mul_pd(qq23,VV);
 551             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 552             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velecsum         = _mm_add_pd(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             /* Update vectorial force */
 560             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 561             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 562             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
 563
 564             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
 565             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
 566             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             r31              = _mm_mul_pd(rsq31,rinv31);
 573
 574             /* Calculate table index by multiplying r with table scale and truncate to integer */
 575             rt               = _mm_mul_pd(r31,vftabscale);
 576             vfitab           = _mm_cvttpd_epi32(rt);
 577 #ifdef __XOP__
 578             vfeps            = _mm_frcz_pd(rt);
 579 #else
 580             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 581 #endif
 582             twovfeps         = _mm_add_pd(vfeps,vfeps);
 583             vfitab           = _mm_slli_epi32(vfitab,2);
 584
 585             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 586             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 587             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 588             GMX_MM_TRANSPOSE2_PD(Y,F);
 589             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 590             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 591             GMX_MM_TRANSPOSE2_PD(G,H);
 592             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 593             VV               = _mm_macc_pd(vfeps,Fp,Y);
 594             velec            = _mm_mul_pd(qq31,VV);
 595             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 596             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
 597
 598             /* Update potential sum for this i atom from the interaction with this j atom. */
 599             velecsum         = _mm_add_pd(velecsum,velec);
 600
 601             fscal            = felec;
 602
 603             /* Update vectorial force */
 604             fix3             = _mm_macc_pd(dx31,fscal,fix3);
 605             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
 606             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
 607
 608             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
 609             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
 610             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             r32              = _mm_mul_pd(rsq32,rinv32);
 617
 618             /* Calculate table index by multiplying r with table scale and truncate to integer */
 619             rt               = _mm_mul_pd(r32,vftabscale);
 620             vfitab           = _mm_cvttpd_epi32(rt);
 621 #ifdef __XOP__
 622             vfeps            = _mm_frcz_pd(rt);
 623 #else
 624             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 625 #endif
 626             twovfeps         = _mm_add_pd(vfeps,vfeps);
 627             vfitab           = _mm_slli_epi32(vfitab,2);
 628
 629             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 630             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 631             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 632             GMX_MM_TRANSPOSE2_PD(Y,F);
 633             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 634             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 635             GMX_MM_TRANSPOSE2_PD(G,H);
 636             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 637             VV               = _mm_macc_pd(vfeps,Fp,Y);
 638             velec            = _mm_mul_pd(qq32,VV);
 639             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 640             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velecsum         = _mm_add_pd(velecsum,velec);
 644
 645             fscal            = felec;
 646
 647             /* Update vectorial force */
 648             fix3             = _mm_macc_pd(dx32,fscal,fix3);
 649             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
 650             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
 651
 652             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
 653             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
 654             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
 655
 656             /**************************
 657              * CALCULATE INTERACTIONS *
 658              **************************/
 659
 660             r33              = _mm_mul_pd(rsq33,rinv33);
 661
 662             /* Calculate table index by multiplying r with table scale and truncate to integer */
 663             rt               = _mm_mul_pd(r33,vftabscale);
 664             vfitab           = _mm_cvttpd_epi32(rt);
 665 #ifdef __XOP__
 666             vfeps            = _mm_frcz_pd(rt);
 667 #else
 668             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 669 #endif
 670             twovfeps         = _mm_add_pd(vfeps,vfeps);
 671             vfitab           = _mm_slli_epi32(vfitab,2);
 672
 673             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 674             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 675             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 676             GMX_MM_TRANSPOSE2_PD(Y,F);
 677             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 678             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 679             GMX_MM_TRANSPOSE2_PD(G,H);
 680             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 681             VV               = _mm_macc_pd(vfeps,Fp,Y);
 682             velec            = _mm_mul_pd(qq33,VV);
 683             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 684             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
 685
 686             /* Update potential sum for this i atom from the interaction with this j atom. */
 687             velecsum         = _mm_add_pd(velecsum,velec);
 688
 689             fscal            = felec;
 690
 691             /* Update vectorial force */
 692             fix3             = _mm_macc_pd(dx33,fscal,fix3);
 693             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
 694             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
 695
 696             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
 697             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
 698             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
 699
 700             gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 701
 702             /* Inner loop uses 452 flops */
 703         }
 704
 705         if(jidx<j_index_end)
 706         {
 707
 708             jnrA             = jjnr[jidx];
 709             j_coord_offsetA  = DIM*jnrA;
 710
 711             /* load j atom coordinates */
 712             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 713                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 714                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 715
 716             /* Calculate displacement vector */
 717             dx00             = _mm_sub_pd(ix0,jx0);
 718             dy00             = _mm_sub_pd(iy0,jy0);
 719             dz00             = _mm_sub_pd(iz0,jz0);
 720             dx11             = _mm_sub_pd(ix1,jx1);
 721             dy11             = _mm_sub_pd(iy1,jy1);
 722             dz11             = _mm_sub_pd(iz1,jz1);
 723             dx12             = _mm_sub_pd(ix1,jx2);
 724             dy12             = _mm_sub_pd(iy1,jy2);
 725             dz12             = _mm_sub_pd(iz1,jz2);
 726             dx13             = _mm_sub_pd(ix1,jx3);
 727             dy13             = _mm_sub_pd(iy1,jy3);
 728             dz13             = _mm_sub_pd(iz1,jz3);
 729             dx21             = _mm_sub_pd(ix2,jx1);
 730             dy21             = _mm_sub_pd(iy2,jy1);
 731             dz21             = _mm_sub_pd(iz2,jz1);
 732             dx22             = _mm_sub_pd(ix2,jx2);
 733             dy22             = _mm_sub_pd(iy2,jy2);
 734             dz22             = _mm_sub_pd(iz2,jz2);
 735             dx23             = _mm_sub_pd(ix2,jx3);
 736             dy23             = _mm_sub_pd(iy2,jy3);
 737             dz23             = _mm_sub_pd(iz2,jz3);
 738             dx31             = _mm_sub_pd(ix3,jx1);
 739             dy31             = _mm_sub_pd(iy3,jy1);
 740             dz31             = _mm_sub_pd(iz3,jz1);
 741             dx32             = _mm_sub_pd(ix3,jx2);
 742             dy32             = _mm_sub_pd(iy3,jy2);
 743             dz32             = _mm_sub_pd(iz3,jz2);
 744             dx33             = _mm_sub_pd(ix3,jx3);
 745             dy33             = _mm_sub_pd(iy3,jy3);
 746             dz33             = _mm_sub_pd(iz3,jz3);
 747
 748             /* Calculate squared distance and things based on it */
 749             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 750             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 751             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 752             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 753             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 754             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 755             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 756             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 757             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 758             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 759
 760             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 761             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 762             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 763             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 764             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 765             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 766             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 767             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 768             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 769
 770             rinvsq00         = gmx_mm_inv_pd(rsq00);
 771
 772             fjx0             = _mm_setzero_pd();
 773             fjy0             = _mm_setzero_pd();
 774             fjz0             = _mm_setzero_pd();
 775             fjx1             = _mm_setzero_pd();
 776             fjy1             = _mm_setzero_pd();
 777             fjz1             = _mm_setzero_pd();
 778             fjx2             = _mm_setzero_pd();
 779             fjy2             = _mm_setzero_pd();
 780             fjz2             = _mm_setzero_pd();
 781             fjx3             = _mm_setzero_pd();
 782             fjy3             = _mm_setzero_pd();
 783             fjz3             = _mm_setzero_pd();
 784
 785             /**************************
 786              * CALCULATE INTERACTIONS *
 787              **************************/
 788
 789             /* LENNARD-JONES DISPERSION/REPULSION */
 790
 791             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 792             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 793             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 794             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 795             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 796
 797             /* Update potential sum for this i atom from the interaction with this j atom. */
 798             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 799             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 800
 801             fscal            = fvdw;
 802
 803             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 804
 805             /* Update vectorial force */
 806             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 807             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 808             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 809
 810             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 811             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 812             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 813
 814             /**************************
 815              * CALCULATE INTERACTIONS *
 816              **************************/
 817
 818             r11              = _mm_mul_pd(rsq11,rinv11);
 819
 820             /* Calculate table index by multiplying r with table scale and truncate to integer */
 821             rt               = _mm_mul_pd(r11,vftabscale);
 822             vfitab           = _mm_cvttpd_epi32(rt);
 823 #ifdef __XOP__
 824             vfeps            = _mm_frcz_pd(rt);
 825 #else
 826             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 827 #endif
 828             twovfeps         = _mm_add_pd(vfeps,vfeps);
 829             vfitab           = _mm_slli_epi32(vfitab,2);
 830
 831             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 832             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 833             F                = _mm_setzero_pd();
 834             GMX_MM_TRANSPOSE2_PD(Y,F);
 835             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 836             H                = _mm_setzero_pd();
 837             GMX_MM_TRANSPOSE2_PD(G,H);
 838             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 839             VV               = _mm_macc_pd(vfeps,Fp,Y);
 840             velec            = _mm_mul_pd(qq11,VV);
 841             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 842             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
 843
 844             /* Update potential sum for this i atom from the interaction with this j atom. */
 845             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 846             velecsum         = _mm_add_pd(velecsum,velec);
 847
 848             fscal            = felec;
 849
 850             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 851
 852             /* Update vectorial force */
 853             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 854             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 855             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 856
 857             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 858             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 859             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             r12              = _mm_mul_pd(rsq12,rinv12);
 866
 867             /* Calculate table index by multiplying r with table scale and truncate to integer */
 868             rt               = _mm_mul_pd(r12,vftabscale);
 869             vfitab           = _mm_cvttpd_epi32(rt);
 870 #ifdef __XOP__
 871             vfeps            = _mm_frcz_pd(rt);
 872 #else
 873             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 874 #endif
 875             twovfeps         = _mm_add_pd(vfeps,vfeps);
 876             vfitab           = _mm_slli_epi32(vfitab,2);
 877
 878             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 879             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 880             F                = _mm_setzero_pd();
 881             GMX_MM_TRANSPOSE2_PD(Y,F);
 882             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 883             H                = _mm_setzero_pd();
 884             GMX_MM_TRANSPOSE2_PD(G,H);
 885             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 886             VV               = _mm_macc_pd(vfeps,Fp,Y);
 887             velec            = _mm_mul_pd(qq12,VV);
 888             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 889             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
 890
 891             /* Update potential sum for this i atom from the interaction with this j atom. */
 892             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 893             velecsum         = _mm_add_pd(velecsum,velec);
 894
 895             fscal            = felec;
 896
 897             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 898
 899             /* Update vectorial force */
 900             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 901             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 902             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 903
 904             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 905             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 906             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 907
 908             /**************************
 909              * CALCULATE INTERACTIONS *
 910              **************************/
 911
 912             r13              = _mm_mul_pd(rsq13,rinv13);
 913
 914             /* Calculate table index by multiplying r with table scale and truncate to integer */
 915             rt               = _mm_mul_pd(r13,vftabscale);
 916             vfitab           = _mm_cvttpd_epi32(rt);
 917 #ifdef __XOP__
 918             vfeps            = _mm_frcz_pd(rt);
 919 #else
 920             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 921 #endif
 922             twovfeps         = _mm_add_pd(vfeps,vfeps);
 923             vfitab           = _mm_slli_epi32(vfitab,2);
 924
 925             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 926             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 927             F                = _mm_setzero_pd();
 928             GMX_MM_TRANSPOSE2_PD(Y,F);
 929             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 930             H                = _mm_setzero_pd();
 931             GMX_MM_TRANSPOSE2_PD(G,H);
 932             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 933             VV               = _mm_macc_pd(vfeps,Fp,Y);
 934             velec            = _mm_mul_pd(qq13,VV);
 935             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 936             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
 937
 938             /* Update potential sum for this i atom from the interaction with this j atom. */
 939             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 940             velecsum         = _mm_add_pd(velecsum,velec);
 941
 942             fscal            = felec;
 943
 944             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 945
 946             /* Update vectorial force */
 947             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 948             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 949             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 950
 951             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 952             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 953             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             r21              = _mm_mul_pd(rsq21,rinv21);
 960
 961             /* Calculate table index by multiplying r with table scale and truncate to integer */
 962             rt               = _mm_mul_pd(r21,vftabscale);
 963             vfitab           = _mm_cvttpd_epi32(rt);
 964 #ifdef __XOP__
 965             vfeps            = _mm_frcz_pd(rt);
 966 #else
 967             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 968 #endif
 969             twovfeps         = _mm_add_pd(vfeps,vfeps);
 970             vfitab           = _mm_slli_epi32(vfitab,2);
 971
 972             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 973             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 974             F                = _mm_setzero_pd();
 975             GMX_MM_TRANSPOSE2_PD(Y,F);
 976             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 977             H                = _mm_setzero_pd();
 978             GMX_MM_TRANSPOSE2_PD(G,H);
 979             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 980             VV               = _mm_macc_pd(vfeps,Fp,Y);
 981             velec            = _mm_mul_pd(qq21,VV);
 982             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 983             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
 984
 985             /* Update potential sum for this i atom from the interaction with this j atom. */
 986             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 987             velecsum         = _mm_add_pd(velecsum,velec);
 988
 989             fscal            = felec;
 990
 991             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 992
 993             /* Update vectorial force */
 994             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 995             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 996             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 997
 998             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 999             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1000             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             r22              = _mm_mul_pd(rsq22,rinv22);
1007
1008             /* Calculate table index by multiplying r with table scale and truncate to integer */
1009             rt               = _mm_mul_pd(r22,vftabscale);
1010             vfitab           = _mm_cvttpd_epi32(rt);
1011 #ifdef __XOP__
1012             vfeps            = _mm_frcz_pd(rt);
1013 #else
1014             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1015 #endif
1016             twovfeps         = _mm_add_pd(vfeps,vfeps);
1017             vfitab           = _mm_slli_epi32(vfitab,2);
1018
1019             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1020             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1021             F                = _mm_setzero_pd();
1022             GMX_MM_TRANSPOSE2_PD(Y,F);
1023             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1024             H                = _mm_setzero_pd();
1025             GMX_MM_TRANSPOSE2_PD(G,H);
1026             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1027             VV               = _mm_macc_pd(vfeps,Fp,Y);
1028             velec            = _mm_mul_pd(qq22,VV);
1029             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1030             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1031
1032             /* Update potential sum for this i atom from the interaction with this j atom. */
1033             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1034             velecsum         = _mm_add_pd(velecsum,velec);
1035
1036             fscal            = felec;
1037
1038             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1039
1040             /* Update vectorial force */
1041             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1042             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1043             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1044
1045             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1046             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1047             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             r23              = _mm_mul_pd(rsq23,rinv23);
1054
1055             /* Calculate table index by multiplying r with table scale and truncate to integer */
1056             rt               = _mm_mul_pd(r23,vftabscale);
1057             vfitab           = _mm_cvttpd_epi32(rt);
1058 #ifdef __XOP__
1059             vfeps            = _mm_frcz_pd(rt);
1060 #else
1061             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1062 #endif
1063             twovfeps         = _mm_add_pd(vfeps,vfeps);
1064             vfitab           = _mm_slli_epi32(vfitab,2);
1065
1066             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1067             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1068             F                = _mm_setzero_pd();
1069             GMX_MM_TRANSPOSE2_PD(Y,F);
1070             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1071             H                = _mm_setzero_pd();
1072             GMX_MM_TRANSPOSE2_PD(G,H);
1073             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1074             VV               = _mm_macc_pd(vfeps,Fp,Y);
1075             velec            = _mm_mul_pd(qq23,VV);
1076             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1077             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
1078
1079             /* Update potential sum for this i atom from the interaction with this j atom. */
1080             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1081             velecsum         = _mm_add_pd(velecsum,velec);
1082
1083             fscal            = felec;
1084
1085             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1086
1087             /* Update vectorial force */
1088             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1089             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1090             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1091
1092             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1093             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1094             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             r31              = _mm_mul_pd(rsq31,rinv31);
1101
1102             /* Calculate table index by multiplying r with table scale and truncate to integer */
1103             rt               = _mm_mul_pd(r31,vftabscale);
1104             vfitab           = _mm_cvttpd_epi32(rt);
1105 #ifdef __XOP__
1106             vfeps            = _mm_frcz_pd(rt);
1107 #else
1108             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1109 #endif
1110             twovfeps         = _mm_add_pd(vfeps,vfeps);
1111             vfitab           = _mm_slli_epi32(vfitab,2);
1112
1113             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1114             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1115             F                = _mm_setzero_pd();
1116             GMX_MM_TRANSPOSE2_PD(Y,F);
1117             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1118             H                = _mm_setzero_pd();
1119             GMX_MM_TRANSPOSE2_PD(G,H);
1120             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1121             VV               = _mm_macc_pd(vfeps,Fp,Y);
1122             velec            = _mm_mul_pd(qq31,VV);
1123             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1124             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
1125
1126             /* Update potential sum for this i atom from the interaction with this j atom. */
1127             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1128             velecsum         = _mm_add_pd(velecsum,velec);
1129
1130             fscal            = felec;
1131
1132             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1133
1134             /* Update vectorial force */
1135             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1136             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1137             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1138
1139             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1140             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1141             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1142
1143             /**************************
1144              * CALCULATE INTERACTIONS *
1145              **************************/
1146
1147             r32              = _mm_mul_pd(rsq32,rinv32);
1148
1149             /* Calculate table index by multiplying r with table scale and truncate to integer */
1150             rt               = _mm_mul_pd(r32,vftabscale);
1151             vfitab           = _mm_cvttpd_epi32(rt);
1152 #ifdef __XOP__
1153             vfeps            = _mm_frcz_pd(rt);
1154 #else
1155             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1156 #endif
1157             twovfeps         = _mm_add_pd(vfeps,vfeps);
1158             vfitab           = _mm_slli_epi32(vfitab,2);
1159
1160             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1161             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1162             F                = _mm_setzero_pd();
1163             GMX_MM_TRANSPOSE2_PD(Y,F);
1164             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1165             H                = _mm_setzero_pd();
1166             GMX_MM_TRANSPOSE2_PD(G,H);
1167             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1168             VV               = _mm_macc_pd(vfeps,Fp,Y);
1169             velec            = _mm_mul_pd(qq32,VV);
1170             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1171             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
1172
1173             /* Update potential sum for this i atom from the interaction with this j atom. */
1174             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1175             velecsum         = _mm_add_pd(velecsum,velec);
1176
1177             fscal            = felec;
1178
1179             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1180
1181             /* Update vectorial force */
1182             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1183             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1184             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1185
1186             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1187             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1188             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1189
1190             /**************************
1191              * CALCULATE INTERACTIONS *
1192              **************************/
1193
1194             r33              = _mm_mul_pd(rsq33,rinv33);
1195
1196             /* Calculate table index by multiplying r with table scale and truncate to integer */
1197             rt               = _mm_mul_pd(r33,vftabscale);
1198             vfitab           = _mm_cvttpd_epi32(rt);
1199 #ifdef __XOP__
1200             vfeps            = _mm_frcz_pd(rt);
1201 #else
1202             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1203 #endif
1204             twovfeps         = _mm_add_pd(vfeps,vfeps);
1205             vfitab           = _mm_slli_epi32(vfitab,2);
1206
1207             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1208             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1209             F                = _mm_setzero_pd();
1210             GMX_MM_TRANSPOSE2_PD(Y,F);
1211             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1212             H                = _mm_setzero_pd();
1213             GMX_MM_TRANSPOSE2_PD(G,H);
1214             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1215             VV               = _mm_macc_pd(vfeps,Fp,Y);
1216             velec            = _mm_mul_pd(qq33,VV);
1217             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1218             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
1219
1220             /* Update potential sum for this i atom from the interaction with this j atom. */
1221             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1222             velecsum         = _mm_add_pd(velecsum,velec);
1223
1224             fscal            = felec;
1225
1226             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1227
1228             /* Update vectorial force */
1229             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1230             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1231             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1232
1233             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1234             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1235             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1236
1237             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1238
1239             /* Inner loop uses 452 flops */
1240         }
1241
1242         /* End of innermost loop */
1243
1244         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1245                                               f+i_coord_offset,fshift+i_shift_offset);
1246
1247         ggid                        = gid[iidx];
1248         /* Update potential energies */
1249         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1250         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1251
1252         /* Increment number of inner iterations */
1253         inneriter                  += j_index_end - j_index_start;
1254
1255         /* Outer loop uses 26 flops */
1256     }
1257
1258     /* Increment number of outer iterations */
1259     outeriter        += nri;
1260
1261     /* Update outer/inner flops */
1262
1263     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*452);
1264 }
1265 /*
1266  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_F_avx_128_fma_double
1267  * Electrostatics interaction: CubicSplineTable
1268  * VdW interaction:            LennardJones
1269  * Geometry:                   Water4-Water4
1270  * Calculate force/pot:        Force
1271  */
1272 void
1273 nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_F_avx_128_fma_double
1274                     (t_nblist * gmx_restrict                nlist,
1275                      rvec * gmx_restrict                    xx,
1276                      rvec * gmx_restrict                    ff,
1277                      t_forcerec * gmx_restrict              fr,
1278                      t_mdatoms * gmx_restrict               mdatoms,
1279                      nb_kernel_data_t * gmx_restrict        kernel_data,
1280                      t_nrnb * gmx_restrict                  nrnb)
1281 {
1282     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1283      * just 0 for non-waters.
1284      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1285      * jnr indices corresponding to data put in the four positions in the SIMD register.
1286      */
1287     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1288     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1289     int              jnrA,jnrB;
1290     int              j_coord_offsetA,j_coord_offsetB;
1291     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1292     real             rcutoff_scalar;
1293     real             *shiftvec,*fshift,*x,*f;
1294     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1295     int              vdwioffset0;
1296     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1297     int              vdwioffset1;
1298     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1299     int              vdwioffset2;
1300     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1301     int              vdwioffset3;
1302     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1303     int              vdwjidx0A,vdwjidx0B;
1304     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1305     int              vdwjidx1A,vdwjidx1B;
1306     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1307     int              vdwjidx2A,vdwjidx2B;
1308     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1309     int              vdwjidx3A,vdwjidx3B;
1310     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1311     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1312     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1313     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1314     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1315     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1316     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1317     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1318     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1319     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1320     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1321     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1322     real             *charge;
1323     int              nvdwtype;
1324     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1325     int              *vdwtype;
1326     real             *vdwparam;
1327     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1328     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1329     __m128i          vfitab;
1330     __m128i          ifour       = _mm_set1_epi32(4);
1331     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
1332     real             *vftab;
1333     __m128d          dummy_mask,cutoff_mask;
1334     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1335     __m128d          one     = _mm_set1_pd(1.0);
1336     __m128d          two     = _mm_set1_pd(2.0);
1337     x                = xx[0];
1338     f                = ff[0];
1339
1340     nri              = nlist->nri;
1341     iinr             = nlist->iinr;
1342     jindex           = nlist->jindex;
1343     jjnr             = nlist->jjnr;
1344     shiftidx         = nlist->shift;
1345     gid              = nlist->gid;
1346     shiftvec         = fr->shift_vec[0];
1347     fshift           = fr->fshift[0];
1348     facel            = _mm_set1_pd(fr->epsfac);
1349     charge           = mdatoms->chargeA;
1350     nvdwtype         = fr->ntype;
1351     vdwparam         = fr->nbfp;
1352     vdwtype          = mdatoms->typeA;
1353
1354     vftab            = kernel_data->table_elec->data;
1355     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
1356
1357     /* Setup water-specific parameters */
1358     inr              = nlist->iinr[0];
1359     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1360     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1361     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1362     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1363
1364     jq1              = _mm_set1_pd(charge[inr+1]);
1365     jq2              = _mm_set1_pd(charge[inr+2]);
1366     jq3              = _mm_set1_pd(charge[inr+3]);
1367     vdwjidx0A        = 2*vdwtype[inr+0];
1368     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1369     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1370     qq11             = _mm_mul_pd(iq1,jq1);
1371     qq12             = _mm_mul_pd(iq1,jq2);
1372     qq13             = _mm_mul_pd(iq1,jq3);
1373     qq21             = _mm_mul_pd(iq2,jq1);
1374     qq22             = _mm_mul_pd(iq2,jq2);
1375     qq23             = _mm_mul_pd(iq2,jq3);
1376     qq31             = _mm_mul_pd(iq3,jq1);
1377     qq32             = _mm_mul_pd(iq3,jq2);
1378     qq33             = _mm_mul_pd(iq3,jq3);
1379
1380     /* Avoid stupid compiler warnings */
1381     jnrA = jnrB = 0;
1382     j_coord_offsetA = 0;
1383     j_coord_offsetB = 0;
1384
1385     outeriter        = 0;
1386     inneriter        = 0;
1387
1388     /* Start outer loop over neighborlists */
1389     for(iidx=0; iidx<nri; iidx++)
1390     {
1391         /* Load shift vector for this list */
1392         i_shift_offset   = DIM*shiftidx[iidx];
1393
1394         /* Load limits for loop over neighbors */
1395         j_index_start    = jindex[iidx];
1396         j_index_end      = jindex[iidx+1];
1397
1398         /* Get outer coordinate index */
1399         inr              = iinr[iidx];
1400         i_coord_offset   = DIM*inr;
1401
1402         /* Load i particle coords and add shift vector */
1403         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1404                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1405
1406         fix0             = _mm_setzero_pd();
1407         fiy0             = _mm_setzero_pd();
1408         fiz0             = _mm_setzero_pd();
1409         fix1             = _mm_setzero_pd();
1410         fiy1             = _mm_setzero_pd();
1411         fiz1             = _mm_setzero_pd();
1412         fix2             = _mm_setzero_pd();
1413         fiy2             = _mm_setzero_pd();
1414         fiz2             = _mm_setzero_pd();
1415         fix3             = _mm_setzero_pd();
1416         fiy3             = _mm_setzero_pd();
1417         fiz3             = _mm_setzero_pd();
1418
1419         /* Start inner kernel loop */
1420         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1421         {
1422
1423             /* Get j neighbor index, and coordinate index */
1424             jnrA             = jjnr[jidx];
1425             jnrB             = jjnr[jidx+1];
1426             j_coord_offsetA  = DIM*jnrA;
1427             j_coord_offsetB  = DIM*jnrB;
1428
1429             /* load j atom coordinates */
1430             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1431                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1432                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1433
1434             /* Calculate displacement vector */
1435             dx00             = _mm_sub_pd(ix0,jx0);
1436             dy00             = _mm_sub_pd(iy0,jy0);
1437             dz00             = _mm_sub_pd(iz0,jz0);
1438             dx11             = _mm_sub_pd(ix1,jx1);
1439             dy11             = _mm_sub_pd(iy1,jy1);
1440             dz11             = _mm_sub_pd(iz1,jz1);
1441             dx12             = _mm_sub_pd(ix1,jx2);
1442             dy12             = _mm_sub_pd(iy1,jy2);
1443             dz12             = _mm_sub_pd(iz1,jz2);
1444             dx13             = _mm_sub_pd(ix1,jx3);
1445             dy13             = _mm_sub_pd(iy1,jy3);
1446             dz13             = _mm_sub_pd(iz1,jz3);
1447             dx21             = _mm_sub_pd(ix2,jx1);
1448             dy21             = _mm_sub_pd(iy2,jy1);
1449             dz21             = _mm_sub_pd(iz2,jz1);
1450             dx22             = _mm_sub_pd(ix2,jx2);
1451             dy22             = _mm_sub_pd(iy2,jy2);
1452             dz22             = _mm_sub_pd(iz2,jz2);
1453             dx23             = _mm_sub_pd(ix2,jx3);
1454             dy23             = _mm_sub_pd(iy2,jy3);
1455             dz23             = _mm_sub_pd(iz2,jz3);
1456             dx31             = _mm_sub_pd(ix3,jx1);
1457             dy31             = _mm_sub_pd(iy3,jy1);
1458             dz31             = _mm_sub_pd(iz3,jz1);
1459             dx32             = _mm_sub_pd(ix3,jx2);
1460             dy32             = _mm_sub_pd(iy3,jy2);
1461             dz32             = _mm_sub_pd(iz3,jz2);
1462             dx33             = _mm_sub_pd(ix3,jx3);
1463             dy33             = _mm_sub_pd(iy3,jy3);
1464             dz33             = _mm_sub_pd(iz3,jz3);
1465
1466             /* Calculate squared distance and things based on it */
1467             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1468             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1469             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1470             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1471             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1472             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1473             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1474             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1475             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1476             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1477
1478             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1479             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1480             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1481             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1482             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1483             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1484             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1485             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1486             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1487
1488             rinvsq00         = gmx_mm_inv_pd(rsq00);
1489
1490             fjx0             = _mm_setzero_pd();
1491             fjy0             = _mm_setzero_pd();
1492             fjz0             = _mm_setzero_pd();
1493             fjx1             = _mm_setzero_pd();
1494             fjy1             = _mm_setzero_pd();
1495             fjz1             = _mm_setzero_pd();
1496             fjx2             = _mm_setzero_pd();
1497             fjy2             = _mm_setzero_pd();
1498             fjz2             = _mm_setzero_pd();
1499             fjx3             = _mm_setzero_pd();
1500             fjy3             = _mm_setzero_pd();
1501             fjz3             = _mm_setzero_pd();
1502
1503             /**************************
1504              * CALCULATE INTERACTIONS *
1505              **************************/
1506
1507             /* LENNARD-JONES DISPERSION/REPULSION */
1508
1509             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1510             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1511
1512             fscal            = fvdw;
1513
1514             /* Update vectorial force */
1515             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1516             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1517             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1518
1519             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1520             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1521             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1522
1523             /**************************
1524              * CALCULATE INTERACTIONS *
1525              **************************/
1526
1527             r11              = _mm_mul_pd(rsq11,rinv11);
1528
1529             /* Calculate table index by multiplying r with table scale and truncate to integer */
1530             rt               = _mm_mul_pd(r11,vftabscale);
1531             vfitab           = _mm_cvttpd_epi32(rt);
1532 #ifdef __XOP__
1533             vfeps            = _mm_frcz_pd(rt);
1534 #else
1535             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1536 #endif
1537             twovfeps         = _mm_add_pd(vfeps,vfeps);
1538             vfitab           = _mm_slli_epi32(vfitab,2);
1539
1540             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1541             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1542             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1543             GMX_MM_TRANSPOSE2_PD(Y,F);
1544             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1545             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1546             GMX_MM_TRANSPOSE2_PD(G,H);
1547             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1548             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1549             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
1550
1551             fscal            = felec;
1552
1553             /* Update vectorial force */
1554             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1555             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1556             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1557
1558             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1559             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1560             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1561
1562             /**************************
1563              * CALCULATE INTERACTIONS *
1564              **************************/
1565
1566             r12              = _mm_mul_pd(rsq12,rinv12);
1567
1568             /* Calculate table index by multiplying r with table scale and truncate to integer */
1569             rt               = _mm_mul_pd(r12,vftabscale);
1570             vfitab           = _mm_cvttpd_epi32(rt);
1571 #ifdef __XOP__
1572             vfeps            = _mm_frcz_pd(rt);
1573 #else
1574             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1575 #endif
1576             twovfeps         = _mm_add_pd(vfeps,vfeps);
1577             vfitab           = _mm_slli_epi32(vfitab,2);
1578
1579             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1580             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1581             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1582             GMX_MM_TRANSPOSE2_PD(Y,F);
1583             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1584             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1585             GMX_MM_TRANSPOSE2_PD(G,H);
1586             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1587             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1588             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
1589
1590             fscal            = felec;
1591
1592             /* Update vectorial force */
1593             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1594             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1595             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1596
1597             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1598             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1599             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1600
1601             /**************************
1602              * CALCULATE INTERACTIONS *
1603              **************************/
1604
1605             r13              = _mm_mul_pd(rsq13,rinv13);
1606
1607             /* Calculate table index by multiplying r with table scale and truncate to integer */
1608             rt               = _mm_mul_pd(r13,vftabscale);
1609             vfitab           = _mm_cvttpd_epi32(rt);
1610 #ifdef __XOP__
1611             vfeps            = _mm_frcz_pd(rt);
1612 #else
1613             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1614 #endif
1615             twovfeps         = _mm_add_pd(vfeps,vfeps);
1616             vfitab           = _mm_slli_epi32(vfitab,2);
1617
1618             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1619             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1620             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1621             GMX_MM_TRANSPOSE2_PD(Y,F);
1622             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1623             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1624             GMX_MM_TRANSPOSE2_PD(G,H);
1625             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1626             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1627             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
1628
1629             fscal            = felec;
1630
1631             /* Update vectorial force */
1632             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1633             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1634             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1635
1636             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1637             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1638             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1639
1640             /**************************
1641              * CALCULATE INTERACTIONS *
1642              **************************/
1643
1644             r21              = _mm_mul_pd(rsq21,rinv21);
1645
1646             /* Calculate table index by multiplying r with table scale and truncate to integer */
1647             rt               = _mm_mul_pd(r21,vftabscale);
1648             vfitab           = _mm_cvttpd_epi32(rt);
1649 #ifdef __XOP__
1650             vfeps            = _mm_frcz_pd(rt);
1651 #else
1652             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1653 #endif
1654             twovfeps         = _mm_add_pd(vfeps,vfeps);
1655             vfitab           = _mm_slli_epi32(vfitab,2);
1656
1657             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1658             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1659             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1660             GMX_MM_TRANSPOSE2_PD(Y,F);
1661             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1662             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1663             GMX_MM_TRANSPOSE2_PD(G,H);
1664             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1665             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1666             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1667
1668             fscal            = felec;
1669
1670             /* Update vectorial force */
1671             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1672             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1673             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1674
1675             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1676             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1677             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1678
1679             /**************************
1680              * CALCULATE INTERACTIONS *
1681              **************************/
1682
1683             r22              = _mm_mul_pd(rsq22,rinv22);
1684
1685             /* Calculate table index by multiplying r with table scale and truncate to integer */
1686             rt               = _mm_mul_pd(r22,vftabscale);
1687             vfitab           = _mm_cvttpd_epi32(rt);
1688 #ifdef __XOP__
1689             vfeps            = _mm_frcz_pd(rt);
1690 #else
1691             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1692 #endif
1693             twovfeps         = _mm_add_pd(vfeps,vfeps);
1694             vfitab           = _mm_slli_epi32(vfitab,2);
1695
1696             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1697             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1698             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1699             GMX_MM_TRANSPOSE2_PD(Y,F);
1700             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1701             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1702             GMX_MM_TRANSPOSE2_PD(G,H);
1703             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1704             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1705             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1706
1707             fscal            = felec;
1708
1709             /* Update vectorial force */
1710             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1711             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1712             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1713
1714             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1715             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1716             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1717
1718             /**************************
1719              * CALCULATE INTERACTIONS *
1720              **************************/
1721
1722             r23              = _mm_mul_pd(rsq23,rinv23);
1723
1724             /* Calculate table index by multiplying r with table scale and truncate to integer */
1725             rt               = _mm_mul_pd(r23,vftabscale);
1726             vfitab           = _mm_cvttpd_epi32(rt);
1727 #ifdef __XOP__
1728             vfeps            = _mm_frcz_pd(rt);
1729 #else
1730             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1731 #endif
1732             twovfeps         = _mm_add_pd(vfeps,vfeps);
1733             vfitab           = _mm_slli_epi32(vfitab,2);
1734
1735             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1736             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1737             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1738             GMX_MM_TRANSPOSE2_PD(Y,F);
1739             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1740             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1741             GMX_MM_TRANSPOSE2_PD(G,H);
1742             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1743             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1744             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
1745
1746             fscal            = felec;
1747
1748             /* Update vectorial force */
1749             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1750             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1751             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1752
1753             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1754             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1755             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1756
1757             /**************************
1758              * CALCULATE INTERACTIONS *
1759              **************************/
1760
1761             r31              = _mm_mul_pd(rsq31,rinv31);
1762
1763             /* Calculate table index by multiplying r with table scale and truncate to integer */
1764             rt               = _mm_mul_pd(r31,vftabscale);
1765             vfitab           = _mm_cvttpd_epi32(rt);
1766 #ifdef __XOP__
1767             vfeps            = _mm_frcz_pd(rt);
1768 #else
1769             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1770 #endif
1771             twovfeps         = _mm_add_pd(vfeps,vfeps);
1772             vfitab           = _mm_slli_epi32(vfitab,2);
1773
1774             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1775             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1776             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1777             GMX_MM_TRANSPOSE2_PD(Y,F);
1778             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1779             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1780             GMX_MM_TRANSPOSE2_PD(G,H);
1781             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1782             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1783             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
1784
1785             fscal            = felec;
1786
1787             /* Update vectorial force */
1788             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1789             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1790             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1791
1792             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1793             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1794             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1795
1796             /**************************
1797              * CALCULATE INTERACTIONS *
1798              **************************/
1799
1800             r32              = _mm_mul_pd(rsq32,rinv32);
1801
1802             /* Calculate table index by multiplying r with table scale and truncate to integer */
1803             rt               = _mm_mul_pd(r32,vftabscale);
1804             vfitab           = _mm_cvttpd_epi32(rt);
1805 #ifdef __XOP__
1806             vfeps            = _mm_frcz_pd(rt);
1807 #else
1808             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1809 #endif
1810             twovfeps         = _mm_add_pd(vfeps,vfeps);
1811             vfitab           = _mm_slli_epi32(vfitab,2);
1812
1813             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1814             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1815             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1816             GMX_MM_TRANSPOSE2_PD(Y,F);
1817             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1818             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1819             GMX_MM_TRANSPOSE2_PD(G,H);
1820             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1821             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1822             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
1823
1824             fscal            = felec;
1825
1826             /* Update vectorial force */
1827             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1828             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1829             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1830
1831             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1832             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1833             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1834
1835             /**************************
1836              * CALCULATE INTERACTIONS *
1837              **************************/
1838
1839             r33              = _mm_mul_pd(rsq33,rinv33);
1840
1841             /* Calculate table index by multiplying r with table scale and truncate to integer */
1842             rt               = _mm_mul_pd(r33,vftabscale);
1843             vfitab           = _mm_cvttpd_epi32(rt);
1844 #ifdef __XOP__
1845             vfeps            = _mm_frcz_pd(rt);
1846 #else
1847             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1848 #endif
1849             twovfeps         = _mm_add_pd(vfeps,vfeps);
1850             vfitab           = _mm_slli_epi32(vfitab,2);
1851
1852             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1853             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1854             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1855             GMX_MM_TRANSPOSE2_PD(Y,F);
1856             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1857             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1858             GMX_MM_TRANSPOSE2_PD(G,H);
1859             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1860             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1861             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
1862
1863             fscal            = felec;
1864
1865             /* Update vectorial force */
1866             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1867             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1868             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1869
1870             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1871             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1872             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1873
1874             gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1875
1876             /* Inner loop uses 411 flops */
1877         }
1878
1879         if(jidx<j_index_end)
1880         {
1881
1882             jnrA             = jjnr[jidx];
1883             j_coord_offsetA  = DIM*jnrA;
1884
1885             /* load j atom coordinates */
1886             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1887                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1888                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1889
1890             /* Calculate displacement vector */
1891             dx00             = _mm_sub_pd(ix0,jx0);
1892             dy00             = _mm_sub_pd(iy0,jy0);
1893             dz00             = _mm_sub_pd(iz0,jz0);
1894             dx11             = _mm_sub_pd(ix1,jx1);
1895             dy11             = _mm_sub_pd(iy1,jy1);
1896             dz11             = _mm_sub_pd(iz1,jz1);
1897             dx12             = _mm_sub_pd(ix1,jx2);
1898             dy12             = _mm_sub_pd(iy1,jy2);
1899             dz12             = _mm_sub_pd(iz1,jz2);
1900             dx13             = _mm_sub_pd(ix1,jx3);
1901             dy13             = _mm_sub_pd(iy1,jy3);
1902             dz13             = _mm_sub_pd(iz1,jz3);
1903             dx21             = _mm_sub_pd(ix2,jx1);
1904             dy21             = _mm_sub_pd(iy2,jy1);
1905             dz21             = _mm_sub_pd(iz2,jz1);
1906             dx22             = _mm_sub_pd(ix2,jx2);
1907             dy22             = _mm_sub_pd(iy2,jy2);
1908             dz22             = _mm_sub_pd(iz2,jz2);
1909             dx23             = _mm_sub_pd(ix2,jx3);
1910             dy23             = _mm_sub_pd(iy2,jy3);
1911             dz23             = _mm_sub_pd(iz2,jz3);
1912             dx31             = _mm_sub_pd(ix3,jx1);
1913             dy31             = _mm_sub_pd(iy3,jy1);
1914             dz31             = _mm_sub_pd(iz3,jz1);
1915             dx32             = _mm_sub_pd(ix3,jx2);
1916             dy32             = _mm_sub_pd(iy3,jy2);
1917             dz32             = _mm_sub_pd(iz3,jz2);
1918             dx33             = _mm_sub_pd(ix3,jx3);
1919             dy33             = _mm_sub_pd(iy3,jy3);
1920             dz33             = _mm_sub_pd(iz3,jz3);
1921
1922             /* Calculate squared distance and things based on it */
1923             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1924             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1925             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1926             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1927             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1928             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1929             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1930             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1931             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1932             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1933
1934             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1935             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1936             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1937             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1938             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1939             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1940             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1941             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1942             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1943
1944             rinvsq00         = gmx_mm_inv_pd(rsq00);
1945
1946             fjx0             = _mm_setzero_pd();
1947             fjy0             = _mm_setzero_pd();
1948             fjz0             = _mm_setzero_pd();
1949             fjx1             = _mm_setzero_pd();
1950             fjy1             = _mm_setzero_pd();
1951             fjz1             = _mm_setzero_pd();
1952             fjx2             = _mm_setzero_pd();
1953             fjy2             = _mm_setzero_pd();
1954             fjz2             = _mm_setzero_pd();
1955             fjx3             = _mm_setzero_pd();
1956             fjy3             = _mm_setzero_pd();
1957             fjz3             = _mm_setzero_pd();
1958
1959             /**************************
1960              * CALCULATE INTERACTIONS *
1961              **************************/
1962
1963             /* LENNARD-JONES DISPERSION/REPULSION */
1964
1965             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1966             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1967
1968             fscal            = fvdw;
1969
1970             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1971
1972             /* Update vectorial force */
1973             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1974             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1975             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1976
1977             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1978             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1979             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1980
1981             /**************************
1982              * CALCULATE INTERACTIONS *
1983              **************************/
1984
1985             r11              = _mm_mul_pd(rsq11,rinv11);
1986
1987             /* Calculate table index by multiplying r with table scale and truncate to integer */
1988             rt               = _mm_mul_pd(r11,vftabscale);
1989             vfitab           = _mm_cvttpd_epi32(rt);
1990 #ifdef __XOP__
1991             vfeps            = _mm_frcz_pd(rt);
1992 #else
1993             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1994 #endif
1995             twovfeps         = _mm_add_pd(vfeps,vfeps);
1996             vfitab           = _mm_slli_epi32(vfitab,2);
1997
1998             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1999             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2000             F                = _mm_setzero_pd();
2001             GMX_MM_TRANSPOSE2_PD(Y,F);
2002             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2003             H                = _mm_setzero_pd();
2004             GMX_MM_TRANSPOSE2_PD(G,H);
2005             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2006             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2007             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
2008
2009             fscal            = felec;
2010
2011             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2012
2013             /* Update vectorial force */
2014             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2015             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2016             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2017
2018             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2019             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2020             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2021
2022             /**************************
2023              * CALCULATE INTERACTIONS *
2024              **************************/
2025
2026             r12              = _mm_mul_pd(rsq12,rinv12);
2027
2028             /* Calculate table index by multiplying r with table scale and truncate to integer */
2029             rt               = _mm_mul_pd(r12,vftabscale);
2030             vfitab           = _mm_cvttpd_epi32(rt);
2031 #ifdef __XOP__
2032             vfeps            = _mm_frcz_pd(rt);
2033 #else
2034             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2035 #endif
2036             twovfeps         = _mm_add_pd(vfeps,vfeps);
2037             vfitab           = _mm_slli_epi32(vfitab,2);
2038
2039             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2040             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2041             F                = _mm_setzero_pd();
2042             GMX_MM_TRANSPOSE2_PD(Y,F);
2043             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2044             H                = _mm_setzero_pd();
2045             GMX_MM_TRANSPOSE2_PD(G,H);
2046             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2047             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2048             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
2049
2050             fscal            = felec;
2051
2052             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2053
2054             /* Update vectorial force */
2055             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2056             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2057             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2058
2059             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2060             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2061             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2062
2063             /**************************
2064              * CALCULATE INTERACTIONS *
2065              **************************/
2066
2067             r13              = _mm_mul_pd(rsq13,rinv13);
2068
2069             /* Calculate table index by multiplying r with table scale and truncate to integer */
2070             rt               = _mm_mul_pd(r13,vftabscale);
2071             vfitab           = _mm_cvttpd_epi32(rt);
2072 #ifdef __XOP__
2073             vfeps            = _mm_frcz_pd(rt);
2074 #else
2075             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2076 #endif
2077             twovfeps         = _mm_add_pd(vfeps,vfeps);
2078             vfitab           = _mm_slli_epi32(vfitab,2);
2079
2080             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2081             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2082             F                = _mm_setzero_pd();
2083             GMX_MM_TRANSPOSE2_PD(Y,F);
2084             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2085             H                = _mm_setzero_pd();
2086             GMX_MM_TRANSPOSE2_PD(G,H);
2087             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2088             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2089             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
2090
2091             fscal            = felec;
2092
2093             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2094
2095             /* Update vectorial force */
2096             fix1             = _mm_macc_pd(dx13,fscal,fix1);
2097             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
2098             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
2099
2100             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
2101             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
2102             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
2103
2104             /**************************
2105              * CALCULATE INTERACTIONS *
2106              **************************/
2107
2108             r21              = _mm_mul_pd(rsq21,rinv21);
2109
2110             /* Calculate table index by multiplying r with table scale and truncate to integer */
2111             rt               = _mm_mul_pd(r21,vftabscale);
2112             vfitab           = _mm_cvttpd_epi32(rt);
2113 #ifdef __XOP__
2114             vfeps            = _mm_frcz_pd(rt);
2115 #else
2116             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2117 #endif
2118             twovfeps         = _mm_add_pd(vfeps,vfeps);
2119             vfitab           = _mm_slli_epi32(vfitab,2);
2120
2121             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2122             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2123             F                = _mm_setzero_pd();
2124             GMX_MM_TRANSPOSE2_PD(Y,F);
2125             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2126             H                = _mm_setzero_pd();
2127             GMX_MM_TRANSPOSE2_PD(G,H);
2128             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2129             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2130             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
2131
2132             fscal            = felec;
2133
2134             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2135
2136             /* Update vectorial force */
2137             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2138             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2139             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2140
2141             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2142             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2143             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2144
2145             /**************************
2146              * CALCULATE INTERACTIONS *
2147              **************************/
2148
2149             r22              = _mm_mul_pd(rsq22,rinv22);
2150
2151             /* Calculate table index by multiplying r with table scale and truncate to integer */
2152             rt               = _mm_mul_pd(r22,vftabscale);
2153             vfitab           = _mm_cvttpd_epi32(rt);
2154 #ifdef __XOP__
2155             vfeps            = _mm_frcz_pd(rt);
2156 #else
2157             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2158 #endif
2159             twovfeps         = _mm_add_pd(vfeps,vfeps);
2160             vfitab           = _mm_slli_epi32(vfitab,2);
2161
2162             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2163             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2164             F                = _mm_setzero_pd();
2165             GMX_MM_TRANSPOSE2_PD(Y,F);
2166             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2167             H                = _mm_setzero_pd();
2168             GMX_MM_TRANSPOSE2_PD(G,H);
2169             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2170             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2171             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
2172
2173             fscal            = felec;
2174
2175             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2176
2177             /* Update vectorial force */
2178             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2179             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2180             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2181
2182             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2183             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2184             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2185
2186             /**************************
2187              * CALCULATE INTERACTIONS *
2188              **************************/
2189
2190             r23              = _mm_mul_pd(rsq23,rinv23);
2191
2192             /* Calculate table index by multiplying r with table scale and truncate to integer */
2193             rt               = _mm_mul_pd(r23,vftabscale);
2194             vfitab           = _mm_cvttpd_epi32(rt);
2195 #ifdef __XOP__
2196             vfeps            = _mm_frcz_pd(rt);
2197 #else
2198             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2199 #endif
2200             twovfeps         = _mm_add_pd(vfeps,vfeps);
2201             vfitab           = _mm_slli_epi32(vfitab,2);
2202
2203             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2204             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2205             F                = _mm_setzero_pd();
2206             GMX_MM_TRANSPOSE2_PD(Y,F);
2207             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2208             H                = _mm_setzero_pd();
2209             GMX_MM_TRANSPOSE2_PD(G,H);
2210             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2211             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2212             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
2213
2214             fscal            = felec;
2215
2216             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2217
2218             /* Update vectorial force */
2219             fix2             = _mm_macc_pd(dx23,fscal,fix2);
2220             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
2221             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
2222
2223             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
2224             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
2225             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
2226
2227             /**************************
2228              * CALCULATE INTERACTIONS *
2229              **************************/
2230
2231             r31              = _mm_mul_pd(rsq31,rinv31);
2232
2233             /* Calculate table index by multiplying r with table scale and truncate to integer */
2234             rt               = _mm_mul_pd(r31,vftabscale);
2235             vfitab           = _mm_cvttpd_epi32(rt);
2236 #ifdef __XOP__
2237             vfeps            = _mm_frcz_pd(rt);
2238 #else
2239             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2240 #endif
2241             twovfeps         = _mm_add_pd(vfeps,vfeps);
2242             vfitab           = _mm_slli_epi32(vfitab,2);
2243
2244             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2245             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2246             F                = _mm_setzero_pd();
2247             GMX_MM_TRANSPOSE2_PD(Y,F);
2248             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2249             H                = _mm_setzero_pd();
2250             GMX_MM_TRANSPOSE2_PD(G,H);
2251             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2252             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2253             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
2254
2255             fscal            = felec;
2256
2257             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2258
2259             /* Update vectorial force */
2260             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2261             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2262             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2263
2264             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2265             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2266             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2267
2268             /**************************
2269              * CALCULATE INTERACTIONS *
2270              **************************/
2271
2272             r32              = _mm_mul_pd(rsq32,rinv32);
2273
2274             /* Calculate table index by multiplying r with table scale and truncate to integer */
2275             rt               = _mm_mul_pd(r32,vftabscale);
2276             vfitab           = _mm_cvttpd_epi32(rt);
2277 #ifdef __XOP__
2278             vfeps            = _mm_frcz_pd(rt);
2279 #else
2280             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2281 #endif
2282             twovfeps         = _mm_add_pd(vfeps,vfeps);
2283             vfitab           = _mm_slli_epi32(vfitab,2);
2284
2285             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2286             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2287             F                = _mm_setzero_pd();
2288             GMX_MM_TRANSPOSE2_PD(Y,F);
2289             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2290             H                = _mm_setzero_pd();
2291             GMX_MM_TRANSPOSE2_PD(G,H);
2292             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2293             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2294             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
2295
2296             fscal            = felec;
2297
2298             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2299
2300             /* Update vectorial force */
2301             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2302             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2303             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2304
2305             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2306             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2307             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2308
2309             /**************************
2310              * CALCULATE INTERACTIONS *
2311              **************************/
2312
2313             r33              = _mm_mul_pd(rsq33,rinv33);
2314
2315             /* Calculate table index by multiplying r with table scale and truncate to integer */
2316             rt               = _mm_mul_pd(r33,vftabscale);
2317             vfitab           = _mm_cvttpd_epi32(rt);
2318 #ifdef __XOP__
2319             vfeps            = _mm_frcz_pd(rt);
2320 #else
2321             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2322 #endif
2323             twovfeps         = _mm_add_pd(vfeps,vfeps);
2324             vfitab           = _mm_slli_epi32(vfitab,2);
2325
2326             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2327             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2328             F                = _mm_setzero_pd();
2329             GMX_MM_TRANSPOSE2_PD(Y,F);
2330             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2331             H                = _mm_setzero_pd();
2332             GMX_MM_TRANSPOSE2_PD(G,H);
2333             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2334             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2335             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
2336
2337             fscal            = felec;
2338
2339             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2340
2341             /* Update vectorial force */
2342             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2343             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2344             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2345
2346             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2347             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2348             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2349
2350             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2351
2352             /* Inner loop uses 411 flops */
2353         }
2354
2355         /* End of innermost loop */
2356
2357         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2358                                               f+i_coord_offset,fshift+i_shift_offset);
2359
2360         /* Increment number of inner iterations */
2361         inneriter                  += j_index_end - j_index_start;
2362
2363         /* Outer loop uses 24 flops */
2364     }
2365
2366     /* Increment number of outer iterations */
2367     outeriter        += nri;
2368
2369     /* Update outer/inner flops */
2370
2371     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*411);
2372 }