src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwNone_GeomW4P1_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_ElecEwSh_VdwNone_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_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              vdwioffset1;
  67     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  68     int              vdwioffset2;
  69     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  70     int              vdwioffset3;
  71     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  75     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  76     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m128i          ewitab;
  80     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  81     real             *ewtab;
  82     __m128d          dummy_mask,cutoff_mask;
  83     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  84     __m128d          one     = _mm_set1_pd(1.0);
  85     __m128d          two     = _mm_set1_pd(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_pd(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99
 100     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 101     ewtab            = fr->ic->tabq_coul_FDV0;
 102     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 103     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 104
 105     /* Setup water-specific parameters */
 106     inr              = nlist->iinr[0];
 107     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 108     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 109     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 110
 111     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 112     rcutoff_scalar   = fr->rcoulomb;
 113     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 114     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 140                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 141
 142         fix1             = _mm_setzero_pd();
 143         fiy1             = _mm_setzero_pd();
 144         fiz1             = _mm_setzero_pd();
 145         fix2             = _mm_setzero_pd();
 146         fiy2             = _mm_setzero_pd();
 147         fiz2             = _mm_setzero_pd();
 148         fix3             = _mm_setzero_pd();
 149         fiy3             = _mm_setzero_pd();
 150         fiz3             = _mm_setzero_pd();
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm_setzero_pd();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             j_coord_offsetA  = DIM*jnrA;
 163             j_coord_offsetB  = DIM*jnrB;
 164
 165             /* load j atom coordinates */
 166             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 167                                               &jx0,&jy0,&jz0);
 168
 169             /* Calculate displacement vector */
 170             dx10             = _mm_sub_pd(ix1,jx0);
 171             dy10             = _mm_sub_pd(iy1,jy0);
 172             dz10             = _mm_sub_pd(iz1,jz0);
 173             dx20             = _mm_sub_pd(ix2,jx0);
 174             dy20             = _mm_sub_pd(iy2,jy0);
 175             dz20             = _mm_sub_pd(iz2,jz0);
 176             dx30             = _mm_sub_pd(ix3,jx0);
 177             dy30             = _mm_sub_pd(iy3,jy0);
 178             dz30             = _mm_sub_pd(iz3,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 182             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 183             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 184
 185             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 186             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 187             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 188
 189             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 190             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 191             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 195
 196             fjx0             = _mm_setzero_pd();
 197             fjy0             = _mm_setzero_pd();
 198             fjz0             = _mm_setzero_pd();
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             if (gmx_mm_any_lt(rsq10,rcutoff2))
 205             {
 206
 207             r10              = _mm_mul_pd(rsq10,rinv10);
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq10             = _mm_mul_pd(iq1,jq0);
 211
 212             /* EWALD ELECTROSTATICS */
 213
 214             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 215             ewrt             = _mm_mul_pd(r10,ewtabscale);
 216             ewitab           = _mm_cvttpd_epi32(ewrt);
 217 #ifdef __XOP__
 218             eweps            = _mm_frcz_pd(ewrt);
 219 #else
 220             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 221 #endif
 222             twoeweps         = _mm_add_pd(eweps,eweps);
 223             ewitab           = _mm_slli_epi32(ewitab,2);
 224             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 225             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 226             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 227             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 228             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 229             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 230             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 231             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 232             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 233             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 234
 235             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 236
 237             /* Update potential sum for this i atom from the interaction with this j atom. */
 238             velec            = _mm_and_pd(velec,cutoff_mask);
 239             velecsum         = _mm_add_pd(velecsum,velec);
 240
 241             fscal            = felec;
 242
 243             fscal            = _mm_and_pd(fscal,cutoff_mask);
 244
 245             /* Update vectorial force */
 246             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 247             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 248             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 249
 250             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 251             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 252             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 253
 254             }
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             if (gmx_mm_any_lt(rsq20,rcutoff2))
 261             {
 262
 263             r20              = _mm_mul_pd(rsq20,rinv20);
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             qq20             = _mm_mul_pd(iq2,jq0);
 267
 268             /* EWALD ELECTROSTATICS */
 269
 270             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 271             ewrt             = _mm_mul_pd(r20,ewtabscale);
 272             ewitab           = _mm_cvttpd_epi32(ewrt);
 273 #ifdef __XOP__
 274             eweps            = _mm_frcz_pd(ewrt);
 275 #else
 276             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 277 #endif
 278             twoeweps         = _mm_add_pd(eweps,eweps);
 279             ewitab           = _mm_slli_epi32(ewitab,2);
 280             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 281             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 282             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 283             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 284             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 285             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 286             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 287             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 288             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 289             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 290
 291             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velec            = _mm_and_pd(velec,cutoff_mask);
 295             velecsum         = _mm_add_pd(velecsum,velec);
 296
 297             fscal            = felec;
 298
 299             fscal            = _mm_and_pd(fscal,cutoff_mask);
 300
 301             /* Update vectorial force */
 302             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 303             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 304             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 305
 306             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 307             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 308             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 309
 310             }
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm_any_lt(rsq30,rcutoff2))
 317             {
 318
 319             r30              = _mm_mul_pd(rsq30,rinv30);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq30             = _mm_mul_pd(iq3,jq0);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 327             ewrt             = _mm_mul_pd(r30,ewtabscale);
 328             ewitab           = _mm_cvttpd_epi32(ewrt);
 329 #ifdef __XOP__
 330             eweps            = _mm_frcz_pd(ewrt);
 331 #else
 332             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 333 #endif
 334             twoeweps         = _mm_add_pd(eweps,eweps);
 335             ewitab           = _mm_slli_epi32(ewitab,2);
 336             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 337             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 338             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 339             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 340             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 341             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 342             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 343             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 344             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 345             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 346
 347             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velec            = _mm_and_pd(velec,cutoff_mask);
 351             velecsum         = _mm_add_pd(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             fscal            = _mm_and_pd(fscal,cutoff_mask);
 356
 357             /* Update vectorial force */
 358             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 359             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 360             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 361
 362             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 363             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 364             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 365
 366             }
 367
 368             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 369
 370             /* Inner loop uses 150 flops */
 371         }
 372
 373         if(jidx<j_index_end)
 374         {
 375
 376             jnrA             = jjnr[jidx];
 377             j_coord_offsetA  = DIM*jnrA;
 378
 379             /* load j atom coordinates */
 380             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 381                                               &jx0,&jy0,&jz0);
 382
 383             /* Calculate displacement vector */
 384             dx10             = _mm_sub_pd(ix1,jx0);
 385             dy10             = _mm_sub_pd(iy1,jy0);
 386             dz10             = _mm_sub_pd(iz1,jz0);
 387             dx20             = _mm_sub_pd(ix2,jx0);
 388             dy20             = _mm_sub_pd(iy2,jy0);
 389             dz20             = _mm_sub_pd(iz2,jz0);
 390             dx30             = _mm_sub_pd(ix3,jx0);
 391             dy30             = _mm_sub_pd(iy3,jy0);
 392             dz30             = _mm_sub_pd(iz3,jz0);
 393
 394             /* Calculate squared distance and things based on it */
 395             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 396             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 397             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 398
 399             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 400             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 401             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 402
 403             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 404             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 405             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 406
 407             /* Load parameters for j particles */
 408             jq0              = _mm_load_sd(charge+jnrA+0);
 409
 410             fjx0             = _mm_setzero_pd();
 411             fjy0             = _mm_setzero_pd();
 412             fjz0             = _mm_setzero_pd();
 413
 414             /**************************
 415              * CALCULATE INTERACTIONS *
 416              **************************/
 417
 418             if (gmx_mm_any_lt(rsq10,rcutoff2))
 419             {
 420
 421             r10              = _mm_mul_pd(rsq10,rinv10);
 422
 423             /* Compute parameters for interactions between i and j atoms */
 424             qq10             = _mm_mul_pd(iq1,jq0);
 425
 426             /* EWALD ELECTROSTATICS */
 427
 428             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 429             ewrt             = _mm_mul_pd(r10,ewtabscale);
 430             ewitab           = _mm_cvttpd_epi32(ewrt);
 431 #ifdef __XOP__
 432             eweps            = _mm_frcz_pd(ewrt);
 433 #else
 434             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 435 #endif
 436             twoeweps         = _mm_add_pd(eweps,eweps);
 437             ewitab           = _mm_slli_epi32(ewitab,2);
 438             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 439             ewtabD           = _mm_setzero_pd();
 440             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 441             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 442             ewtabFn          = _mm_setzero_pd();
 443             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 444             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 445             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 446             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 447             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 448
 449             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 450
 451             /* Update potential sum for this i atom from the interaction with this j atom. */
 452             velec            = _mm_and_pd(velec,cutoff_mask);
 453             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 454             velecsum         = _mm_add_pd(velecsum,velec);
 455
 456             fscal            = felec;
 457
 458             fscal            = _mm_and_pd(fscal,cutoff_mask);
 459
 460             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 461
 462             /* Update vectorial force */
 463             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 464             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 465             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 466
 467             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 468             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 469             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 470
 471             }
 472
 473             /**************************
 474              * CALCULATE INTERACTIONS *
 475              **************************/
 476
 477             if (gmx_mm_any_lt(rsq20,rcutoff2))
 478             {
 479
 480             r20              = _mm_mul_pd(rsq20,rinv20);
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             qq20             = _mm_mul_pd(iq2,jq0);
 484
 485             /* EWALD ELECTROSTATICS */
 486
 487             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 488             ewrt             = _mm_mul_pd(r20,ewtabscale);
 489             ewitab           = _mm_cvttpd_epi32(ewrt);
 490 #ifdef __XOP__
 491             eweps            = _mm_frcz_pd(ewrt);
 492 #else
 493             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 494 #endif
 495             twoeweps         = _mm_add_pd(eweps,eweps);
 496             ewitab           = _mm_slli_epi32(ewitab,2);
 497             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 498             ewtabD           = _mm_setzero_pd();
 499             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 500             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 501             ewtabFn          = _mm_setzero_pd();
 502             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 503             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 504             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 505             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 506             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 507
 508             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             velec            = _mm_and_pd(velec,cutoff_mask);
 512             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 513             velecsum         = _mm_add_pd(velecsum,velec);
 514
 515             fscal            = felec;
 516
 517             fscal            = _mm_and_pd(fscal,cutoff_mask);
 518
 519             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 520
 521             /* Update vectorial force */
 522             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 523             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 524             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 525
 526             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 527             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 528             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 529
 530             }
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             if (gmx_mm_any_lt(rsq30,rcutoff2))
 537             {
 538
 539             r30              = _mm_mul_pd(rsq30,rinv30);
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq30             = _mm_mul_pd(iq3,jq0);
 543
 544             /* EWALD ELECTROSTATICS */
 545
 546             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 547             ewrt             = _mm_mul_pd(r30,ewtabscale);
 548             ewitab           = _mm_cvttpd_epi32(ewrt);
 549 #ifdef __XOP__
 550             eweps            = _mm_frcz_pd(ewrt);
 551 #else
 552             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 553 #endif
 554             twoeweps         = _mm_add_pd(eweps,eweps);
 555             ewitab           = _mm_slli_epi32(ewitab,2);
 556             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 557             ewtabD           = _mm_setzero_pd();
 558             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 559             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 560             ewtabFn          = _mm_setzero_pd();
 561             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 562             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 563             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 564             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 565             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 566
 567             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm_and_pd(velec,cutoff_mask);
 571             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 572             velecsum         = _mm_add_pd(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm_and_pd(fscal,cutoff_mask);
 577
 578             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 579
 580             /* Update vectorial force */
 581             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 582             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 583             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 584
 585             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 586             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 587             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 588
 589             }
 590
 591             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 592
 593             /* Inner loop uses 150 flops */
 594         }
 595
 596         /* End of innermost loop */
 597
 598         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 599                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 600
 601         ggid                        = gid[iidx];
 602         /* Update potential energies */
 603         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 604
 605         /* Increment number of inner iterations */
 606         inneriter                  += j_index_end - j_index_start;
 607
 608         /* Outer loop uses 19 flops */
 609     }
 610
 611     /* Increment number of outer iterations */
 612     outeriter        += nri;
 613
 614     /* Update outer/inner flops */
 615
 616     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*150);
 617 }
 618 /*
 619  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_avx_128_fma_double
 620  * Electrostatics interaction: Ewald
 621  * VdW interaction:            None
 622  * Geometry:                   Water4-Particle
 623  * Calculate force/pot:        Force
 624  */
 625 void
 626 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_avx_128_fma_double
 627                     (t_nblist * gmx_restrict                nlist,
 628                      rvec * gmx_restrict                    xx,
 629                      rvec * gmx_restrict                    ff,
 630                      t_forcerec * gmx_restrict              fr,
 631                      t_mdatoms * gmx_restrict               mdatoms,
 632                      nb_kernel_data_t * gmx_restrict        kernel_data,
 633                      t_nrnb * gmx_restrict                  nrnb)
 634 {
 635     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 636      * just 0 for non-waters.
 637      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 638      * jnr indices corresponding to data put in the four positions in the SIMD register.
 639      */
 640     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 641     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 642     int              jnrA,jnrB;
 643     int              j_coord_offsetA,j_coord_offsetB;
 644     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 645     real             rcutoff_scalar;
 646     real             *shiftvec,*fshift,*x,*f;
 647     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 648     int              vdwioffset1;
 649     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 650     int              vdwioffset2;
 651     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 652     int              vdwioffset3;
 653     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 654     int              vdwjidx0A,vdwjidx0B;
 655     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 656     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 657     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 658     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 659     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 660     real             *charge;
 661     __m128i          ewitab;
 662     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 663     real             *ewtab;
 664     __m128d          dummy_mask,cutoff_mask;
 665     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 666     __m128d          one     = _mm_set1_pd(1.0);
 667     __m128d          two     = _mm_set1_pd(2.0);
 668     x                = xx[0];
 669     f                = ff[0];
 670
 671     nri              = nlist->nri;
 672     iinr             = nlist->iinr;
 673     jindex           = nlist->jindex;
 674     jjnr             = nlist->jjnr;
 675     shiftidx         = nlist->shift;
 676     gid              = nlist->gid;
 677     shiftvec         = fr->shift_vec[0];
 678     fshift           = fr->fshift[0];
 679     facel            = _mm_set1_pd(fr->epsfac);
 680     charge           = mdatoms->chargeA;
 681
 682     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 683     ewtab            = fr->ic->tabq_coul_F;
 684     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 685     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 686
 687     /* Setup water-specific parameters */
 688     inr              = nlist->iinr[0];
 689     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 690     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 691     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 692
 693     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 694     rcutoff_scalar   = fr->rcoulomb;
 695     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 696     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 697
 698     /* Avoid stupid compiler warnings */
 699     jnrA = jnrB = 0;
 700     j_coord_offsetA = 0;
 701     j_coord_offsetB = 0;
 702
 703     outeriter        = 0;
 704     inneriter        = 0;
 705
 706     /* Start outer loop over neighborlists */
 707     for(iidx=0; iidx<nri; iidx++)
 708     {
 709         /* Load shift vector for this list */
 710         i_shift_offset   = DIM*shiftidx[iidx];
 711
 712         /* Load limits for loop over neighbors */
 713         j_index_start    = jindex[iidx];
 714         j_index_end      = jindex[iidx+1];
 715
 716         /* Get outer coordinate index */
 717         inr              = iinr[iidx];
 718         i_coord_offset   = DIM*inr;
 719
 720         /* Load i particle coords and add shift vector */
 721         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 722                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 723
 724         fix1             = _mm_setzero_pd();
 725         fiy1             = _mm_setzero_pd();
 726         fiz1             = _mm_setzero_pd();
 727         fix2             = _mm_setzero_pd();
 728         fiy2             = _mm_setzero_pd();
 729         fiz2             = _mm_setzero_pd();
 730         fix3             = _mm_setzero_pd();
 731         fiy3             = _mm_setzero_pd();
 732         fiz3             = _mm_setzero_pd();
 733
 734         /* Start inner kernel loop */
 735         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 736         {
 737
 738             /* Get j neighbor index, and coordinate index */
 739             jnrA             = jjnr[jidx];
 740             jnrB             = jjnr[jidx+1];
 741             j_coord_offsetA  = DIM*jnrA;
 742             j_coord_offsetB  = DIM*jnrB;
 743
 744             /* load j atom coordinates */
 745             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 746                                               &jx0,&jy0,&jz0);
 747
 748             /* Calculate displacement vector */
 749             dx10             = _mm_sub_pd(ix1,jx0);
 750             dy10             = _mm_sub_pd(iy1,jy0);
 751             dz10             = _mm_sub_pd(iz1,jz0);
 752             dx20             = _mm_sub_pd(ix2,jx0);
 753             dy20             = _mm_sub_pd(iy2,jy0);
 754             dz20             = _mm_sub_pd(iz2,jz0);
 755             dx30             = _mm_sub_pd(ix3,jx0);
 756             dy30             = _mm_sub_pd(iy3,jy0);
 757             dz30             = _mm_sub_pd(iz3,jz0);
 758
 759             /* Calculate squared distance and things based on it */
 760             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 761             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 762             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 763
 764             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 765             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 766             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 767
 768             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 769             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 770             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 771
 772             /* Load parameters for j particles */
 773             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 774
 775             fjx0             = _mm_setzero_pd();
 776             fjy0             = _mm_setzero_pd();
 777             fjz0             = _mm_setzero_pd();
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             if (gmx_mm_any_lt(rsq10,rcutoff2))
 784             {
 785
 786             r10              = _mm_mul_pd(rsq10,rinv10);
 787
 788             /* Compute parameters for interactions between i and j atoms */
 789             qq10             = _mm_mul_pd(iq1,jq0);
 790
 791             /* EWALD ELECTROSTATICS */
 792
 793             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 794             ewrt             = _mm_mul_pd(r10,ewtabscale);
 795             ewitab           = _mm_cvttpd_epi32(ewrt);
 796 #ifdef __XOP__
 797             eweps            = _mm_frcz_pd(ewrt);
 798 #else
 799             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 800 #endif
 801             twoeweps         = _mm_add_pd(eweps,eweps);
 802             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 803                                          &ewtabF,&ewtabFn);
 804             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 805             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 806
 807             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 808
 809             fscal            = felec;
 810
 811             fscal            = _mm_and_pd(fscal,cutoff_mask);
 812
 813             /* Update vectorial force */
 814             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 815             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 816             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 817
 818             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 819             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 820             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 821
 822             }
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             if (gmx_mm_any_lt(rsq20,rcutoff2))
 829             {
 830
 831             r20              = _mm_mul_pd(rsq20,rinv20);
 832
 833             /* Compute parameters for interactions between i and j atoms */
 834             qq20             = _mm_mul_pd(iq2,jq0);
 835
 836             /* EWALD ELECTROSTATICS */
 837
 838             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 839             ewrt             = _mm_mul_pd(r20,ewtabscale);
 840             ewitab           = _mm_cvttpd_epi32(ewrt);
 841 #ifdef __XOP__
 842             eweps            = _mm_frcz_pd(ewrt);
 843 #else
 844             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 845 #endif
 846             twoeweps         = _mm_add_pd(eweps,eweps);
 847             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 848                                          &ewtabF,&ewtabFn);
 849             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 850             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 851
 852             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 853
 854             fscal            = felec;
 855
 856             fscal            = _mm_and_pd(fscal,cutoff_mask);
 857
 858             /* Update vectorial force */
 859             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 860             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 861             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 862
 863             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 864             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 865             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 866
 867             }
 868
 869             /**************************
 870              * CALCULATE INTERACTIONS *
 871              **************************/
 872
 873             if (gmx_mm_any_lt(rsq30,rcutoff2))
 874             {
 875
 876             r30              = _mm_mul_pd(rsq30,rinv30);
 877
 878             /* Compute parameters for interactions between i and j atoms */
 879             qq30             = _mm_mul_pd(iq3,jq0);
 880
 881             /* EWALD ELECTROSTATICS */
 882
 883             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 884             ewrt             = _mm_mul_pd(r30,ewtabscale);
 885             ewitab           = _mm_cvttpd_epi32(ewrt);
 886 #ifdef __XOP__
 887             eweps            = _mm_frcz_pd(ewrt);
 888 #else
 889             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 890 #endif
 891             twoeweps         = _mm_add_pd(eweps,eweps);
 892             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 893                                          &ewtabF,&ewtabFn);
 894             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 895             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 896
 897             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 898
 899             fscal            = felec;
 900
 901             fscal            = _mm_and_pd(fscal,cutoff_mask);
 902
 903             /* Update vectorial force */
 904             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 905             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 906             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 907
 908             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 909             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 910             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 911
 912             }
 913
 914             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 915
 916             /* Inner loop uses 129 flops */
 917         }
 918
 919         if(jidx<j_index_end)
 920         {
 921
 922             jnrA             = jjnr[jidx];
 923             j_coord_offsetA  = DIM*jnrA;
 924
 925             /* load j atom coordinates */
 926             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 927                                               &jx0,&jy0,&jz0);
 928
 929             /* Calculate displacement vector */
 930             dx10             = _mm_sub_pd(ix1,jx0);
 931             dy10             = _mm_sub_pd(iy1,jy0);
 932             dz10             = _mm_sub_pd(iz1,jz0);
 933             dx20             = _mm_sub_pd(ix2,jx0);
 934             dy20             = _mm_sub_pd(iy2,jy0);
 935             dz20             = _mm_sub_pd(iz2,jz0);
 936             dx30             = _mm_sub_pd(ix3,jx0);
 937             dy30             = _mm_sub_pd(iy3,jy0);
 938             dz30             = _mm_sub_pd(iz3,jz0);
 939
 940             /* Calculate squared distance and things based on it */
 941             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 942             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 943             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 944
 945             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 946             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 947             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 948
 949             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 950             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 951             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 952
 953             /* Load parameters for j particles */
 954             jq0              = _mm_load_sd(charge+jnrA+0);
 955
 956             fjx0             = _mm_setzero_pd();
 957             fjy0             = _mm_setzero_pd();
 958             fjz0             = _mm_setzero_pd();
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             if (gmx_mm_any_lt(rsq10,rcutoff2))
 965             {
 966
 967             r10              = _mm_mul_pd(rsq10,rinv10);
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq10             = _mm_mul_pd(iq1,jq0);
 971
 972             /* EWALD ELECTROSTATICS */
 973
 974             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 975             ewrt             = _mm_mul_pd(r10,ewtabscale);
 976             ewitab           = _mm_cvttpd_epi32(ewrt);
 977 #ifdef __XOP__
 978             eweps            = _mm_frcz_pd(ewrt);
 979 #else
 980             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 981 #endif
 982             twoeweps         = _mm_add_pd(eweps,eweps);
 983             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 984             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 985             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 986
 987             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 988
 989             fscal            = felec;
 990
 991             fscal            = _mm_and_pd(fscal,cutoff_mask);
 992
 993             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 994
 995             /* Update vectorial force */
 996             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 997             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 998             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 999
1000             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1001             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1002             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1003
1004             }
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             if (gmx_mm_any_lt(rsq20,rcutoff2))
1011             {
1012
1013             r20              = _mm_mul_pd(rsq20,rinv20);
1014
1015             /* Compute parameters for interactions between i and j atoms */
1016             qq20             = _mm_mul_pd(iq2,jq0);
1017
1018             /* EWALD ELECTROSTATICS */
1019
1020             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1021             ewrt             = _mm_mul_pd(r20,ewtabscale);
1022             ewitab           = _mm_cvttpd_epi32(ewrt);
1023 #ifdef __XOP__
1024             eweps            = _mm_frcz_pd(ewrt);
1025 #else
1026             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1027 #endif
1028             twoeweps         = _mm_add_pd(eweps,eweps);
1029             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1030             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1031             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1032
1033             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1034
1035             fscal            = felec;
1036
1037             fscal            = _mm_and_pd(fscal,cutoff_mask);
1038
1039             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1040
1041             /* Update vectorial force */
1042             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1043             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1044             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1045
1046             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1047             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1048             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1049
1050             }
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             if (gmx_mm_any_lt(rsq30,rcutoff2))
1057             {
1058
1059             r30              = _mm_mul_pd(rsq30,rinv30);
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq30             = _mm_mul_pd(iq3,jq0);
1063
1064             /* EWALD ELECTROSTATICS */
1065
1066             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1067             ewrt             = _mm_mul_pd(r30,ewtabscale);
1068             ewitab           = _mm_cvttpd_epi32(ewrt);
1069 #ifdef __XOP__
1070             eweps            = _mm_frcz_pd(ewrt);
1071 #else
1072             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1073 #endif
1074             twoeweps         = _mm_add_pd(eweps,eweps);
1075             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1076             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1077             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1078
1079             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1080
1081             fscal            = felec;
1082
1083             fscal            = _mm_and_pd(fscal,cutoff_mask);
1084
1085             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1086
1087             /* Update vectorial force */
1088             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1089             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1090             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1091
1092             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1093             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1094             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1095
1096             }
1097
1098             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1099
1100             /* Inner loop uses 129 flops */
1101         }
1102
1103         /* End of innermost loop */
1104
1105         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1106                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1107
1108         /* Increment number of inner iterations */
1109         inneriter                  += j_index_end - j_index_start;
1110
1111         /* Outer loop uses 18 flops */
1112     }
1113
1114     /* Increment number of outer iterations */
1115     outeriter        += nri;
1116
1117     /* Update outer/inner flops */
1118
1119     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*129);
1120 }