src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwNone_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 101     real             *vftab;
 102     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119
 120     vftab            = kernel_data->table_elec->data;
 121     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 126     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 127     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 160                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 161
 162         fix1             = _mm_setzero_ps();
 163         fiy1             = _mm_setzero_ps();
 164         fiz1             = _mm_setzero_ps();
 165         fix2             = _mm_setzero_ps();
 166         fiy2             = _mm_setzero_ps();
 167         fiz2             = _mm_setzero_ps();
 168         fix3             = _mm_setzero_ps();
 169         fiy3             = _mm_setzero_ps();
 170         fiz3             = _mm_setzero_ps();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_ps();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186             j_coord_offsetC  = DIM*jnrC;
 187             j_coord_offsetD  = DIM*jnrD;
 188
 189             /* load j atom coordinates */
 190             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                               x+j_coord_offsetC,x+j_coord_offsetD,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx10             = _mm_sub_ps(ix1,jx0);
 196             dy10             = _mm_sub_ps(iy1,jy0);
 197             dz10             = _mm_sub_ps(iz1,jz0);
 198             dx20             = _mm_sub_ps(ix2,jx0);
 199             dy20             = _mm_sub_ps(iy2,jy0);
 200             dz20             = _mm_sub_ps(iz2,jz0);
 201             dx30             = _mm_sub_ps(ix3,jx0);
 202             dy30             = _mm_sub_ps(iy3,jy0);
 203             dz30             = _mm_sub_ps(iz3,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 207             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 208             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 209
 210             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 211             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 212             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 216                                                               charge+jnrC+0,charge+jnrD+0);
 217
 218             fjx0             = _mm_setzero_ps();
 219             fjy0             = _mm_setzero_ps();
 220             fjz0             = _mm_setzero_ps();
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             r10              = _mm_mul_ps(rsq10,rinv10);
 227
 228             /* Compute parameters for interactions between i and j atoms */
 229             qq10             = _mm_mul_ps(iq1,jq0);
 230
 231             /* Calculate table index by multiplying r with table scale and truncate to integer */
 232             rt               = _mm_mul_ps(r10,vftabscale);
 233             vfitab           = _mm_cvttps_epi32(rt);
 234 #ifdef __XOP__
 235             vfeps            = _mm_frcz_ps(rt);
 236 #else
 237             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 238 #endif
 239             twovfeps         = _mm_add_ps(vfeps,vfeps);
 240             vfitab           = _mm_slli_epi32(vfitab,2);
 241
 242             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 243             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 244             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 245             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 246             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 247             _MM_TRANSPOSE4_PS(Y,F,G,H);
 248             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 249             VV               = _mm_macc_ps(vfeps,Fp,Y);
 250             velec            = _mm_mul_ps(qq10,VV);
 251             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 252             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 253
 254             /* Update potential sum for this i atom from the interaction with this j atom. */
 255             velecsum         = _mm_add_ps(velecsum,velec);
 256
 257             fscal            = felec;
 258
 259              /* Update vectorial force */
 260             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 261             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 262             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 263
 264             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 265             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 266             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             r20              = _mm_mul_ps(rsq20,rinv20);
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq20             = _mm_mul_ps(iq2,jq0);
 276
 277             /* Calculate table index by multiplying r with table scale and truncate to integer */
 278             rt               = _mm_mul_ps(r20,vftabscale);
 279             vfitab           = _mm_cvttps_epi32(rt);
 280 #ifdef __XOP__
 281             vfeps            = _mm_frcz_ps(rt);
 282 #else
 283             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 284 #endif
 285             twovfeps         = _mm_add_ps(vfeps,vfeps);
 286             vfitab           = _mm_slli_epi32(vfitab,2);
 287
 288             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 289             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 290             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 291             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 292             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 293             _MM_TRANSPOSE4_PS(Y,F,G,H);
 294             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 295             VV               = _mm_macc_ps(vfeps,Fp,Y);
 296             velec            = _mm_mul_ps(qq20,VV);
 297             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 298             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm_add_ps(velecsum,velec);
 302
 303             fscal            = felec;
 304
 305              /* Update vectorial force */
 306             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 307             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 308             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 309
 310             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 311             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 312             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r30              = _mm_mul_ps(rsq30,rinv30);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq30             = _mm_mul_ps(iq3,jq0);
 322
 323             /* Calculate table index by multiplying r with table scale and truncate to integer */
 324             rt               = _mm_mul_ps(r30,vftabscale);
 325             vfitab           = _mm_cvttps_epi32(rt);
 326 #ifdef __XOP__
 327             vfeps            = _mm_frcz_ps(rt);
 328 #else
 329             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 330 #endif
 331             twovfeps         = _mm_add_ps(vfeps,vfeps);
 332             vfitab           = _mm_slli_epi32(vfitab,2);
 333
 334             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 335             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 336             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 337             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 338             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 339             _MM_TRANSPOSE4_PS(Y,F,G,H);
 340             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 341             VV               = _mm_macc_ps(vfeps,Fp,Y);
 342             velec            = _mm_mul_ps(qq30,VV);
 343             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 344             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351              /* Update vectorial force */
 352             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 353             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 354             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 355
 356             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 357             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 358             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 359
 360             fjptrA             = f+j_coord_offsetA;
 361             fjptrB             = f+j_coord_offsetB;
 362             fjptrC             = f+j_coord_offsetC;
 363             fjptrD             = f+j_coord_offsetD;
 364
 365             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 366
 367             /* Inner loop uses 138 flops */
 368         }
 369
 370         if(jidx<j_index_end)
 371         {
 372
 373             /* Get j neighbor index, and coordinate index */
 374             jnrlistA         = jjnr[jidx];
 375             jnrlistB         = jjnr[jidx+1];
 376             jnrlistC         = jjnr[jidx+2];
 377             jnrlistD         = jjnr[jidx+3];
 378             /* Sign of each element will be negative for non-real atoms.
 379              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 380              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 381              */
 382             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 383             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 384             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 385             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 386             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 387             j_coord_offsetA  = DIM*jnrA;
 388             j_coord_offsetB  = DIM*jnrB;
 389             j_coord_offsetC  = DIM*jnrC;
 390             j_coord_offsetD  = DIM*jnrD;
 391
 392             /* load j atom coordinates */
 393             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 394                                               x+j_coord_offsetC,x+j_coord_offsetD,
 395                                               &jx0,&jy0,&jz0);
 396
 397             /* Calculate displacement vector */
 398             dx10             = _mm_sub_ps(ix1,jx0);
 399             dy10             = _mm_sub_ps(iy1,jy0);
 400             dz10             = _mm_sub_ps(iz1,jz0);
 401             dx20             = _mm_sub_ps(ix2,jx0);
 402             dy20             = _mm_sub_ps(iy2,jy0);
 403             dz20             = _mm_sub_ps(iz2,jz0);
 404             dx30             = _mm_sub_ps(ix3,jx0);
 405             dy30             = _mm_sub_ps(iy3,jy0);
 406             dz30             = _mm_sub_ps(iz3,jz0);
 407
 408             /* Calculate squared distance and things based on it */
 409             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 410             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 411             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 412
 413             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 414             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 415             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 416
 417             /* Load parameters for j particles */
 418             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 419                                                               charge+jnrC+0,charge+jnrD+0);
 420
 421             fjx0             = _mm_setzero_ps();
 422             fjy0             = _mm_setzero_ps();
 423             fjz0             = _mm_setzero_ps();
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             r10              = _mm_mul_ps(rsq10,rinv10);
 430             r10              = _mm_andnot_ps(dummy_mask,r10);
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             qq10             = _mm_mul_ps(iq1,jq0);
 434
 435             /* Calculate table index by multiplying r with table scale and truncate to integer */
 436             rt               = _mm_mul_ps(r10,vftabscale);
 437             vfitab           = _mm_cvttps_epi32(rt);
 438 #ifdef __XOP__
 439             vfeps            = _mm_frcz_ps(rt);
 440 #else
 441             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 442 #endif
 443             twovfeps         = _mm_add_ps(vfeps,vfeps);
 444             vfitab           = _mm_slli_epi32(vfitab,2);
 445
 446             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 447             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 448             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 449             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 450             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 451             _MM_TRANSPOSE4_PS(Y,F,G,H);
 452             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 453             VV               = _mm_macc_ps(vfeps,Fp,Y);
 454             velec            = _mm_mul_ps(qq10,VV);
 455             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 456             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velec            = _mm_andnot_ps(dummy_mask,velec);
 460             velecsum         = _mm_add_ps(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 465
 466              /* Update vectorial force */
 467             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 468             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 469             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 470
 471             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 472             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 473             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             r20              = _mm_mul_ps(rsq20,rinv20);
 480             r20              = _mm_andnot_ps(dummy_mask,r20);
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             qq20             = _mm_mul_ps(iq2,jq0);
 484
 485             /* Calculate table index by multiplying r with table scale and truncate to integer */
 486             rt               = _mm_mul_ps(r20,vftabscale);
 487             vfitab           = _mm_cvttps_epi32(rt);
 488 #ifdef __XOP__
 489             vfeps            = _mm_frcz_ps(rt);
 490 #else
 491             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 492 #endif
 493             twovfeps         = _mm_add_ps(vfeps,vfeps);
 494             vfitab           = _mm_slli_epi32(vfitab,2);
 495
 496             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 497             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 498             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 499             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 500             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 501             _MM_TRANSPOSE4_PS(Y,F,G,H);
 502             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 503             VV               = _mm_macc_ps(vfeps,Fp,Y);
 504             velec            = _mm_mul_ps(qq20,VV);
 505             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 506             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velec            = _mm_andnot_ps(dummy_mask,velec);
 510             velecsum         = _mm_add_ps(velecsum,velec);
 511
 512             fscal            = felec;
 513
 514             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 515
 516              /* Update vectorial force */
 517             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 518             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 519             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 520
 521             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 522             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 523             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             r30              = _mm_mul_ps(rsq30,rinv30);
 530             r30              = _mm_andnot_ps(dummy_mask,r30);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq30             = _mm_mul_ps(iq3,jq0);
 534
 535             /* Calculate table index by multiplying r with table scale and truncate to integer */
 536             rt               = _mm_mul_ps(r30,vftabscale);
 537             vfitab           = _mm_cvttps_epi32(rt);
 538 #ifdef __XOP__
 539             vfeps            = _mm_frcz_ps(rt);
 540 #else
 541             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 542 #endif
 543             twovfeps         = _mm_add_ps(vfeps,vfeps);
 544             vfitab           = _mm_slli_epi32(vfitab,2);
 545
 546             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 547             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 548             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 549             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 550             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 551             _MM_TRANSPOSE4_PS(Y,F,G,H);
 552             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 553             VV               = _mm_macc_ps(vfeps,Fp,Y);
 554             velec            = _mm_mul_ps(qq30,VV);
 555             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 556             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velec            = _mm_andnot_ps(dummy_mask,velec);
 560             velecsum         = _mm_add_ps(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 565
 566              /* Update vectorial force */
 567             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 568             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 569             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 570
 571             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 572             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 573             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 574
 575             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 576             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 577             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 578             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 579
 580             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 581
 582             /* Inner loop uses 141 flops */
 583         }
 584
 585         /* End of innermost loop */
 586
 587         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 588                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 589
 590         ggid                        = gid[iidx];
 591         /* Update potential energies */
 592         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 593
 594         /* Increment number of inner iterations */
 595         inneriter                  += j_index_end - j_index_start;
 596
 597         /* Outer loop uses 19 flops */
 598     }
 599
 600     /* Increment number of outer iterations */
 601     outeriter        += nri;
 602
 603     /* Update outer/inner flops */
 604
 605     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*141);
 606 }
 607 /*
 608  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_128_fma_single
 609  * Electrostatics interaction: CubicSplineTable
 610  * VdW interaction:            None
 611  * Geometry:                   Water4-Particle
 612  * Calculate force/pot:        Force
 613  */
 614 void
 615 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_128_fma_single
 616                     (t_nblist                    * gmx_restrict       nlist,
 617                      rvec                        * gmx_restrict          xx,
 618                      rvec                        * gmx_restrict          ff,
 619                      t_forcerec                  * gmx_restrict          fr,
 620                      t_mdatoms                   * gmx_restrict     mdatoms,
 621                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 622                      t_nrnb                      * gmx_restrict        nrnb)
 623 {
 624     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 625      * just 0 for non-waters.
 626      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 627      * jnr indices corresponding to data put in the four positions in the SIMD register.
 628      */
 629     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 630     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 631     int              jnrA,jnrB,jnrC,jnrD;
 632     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 633     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 634     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 635     real             rcutoff_scalar;
 636     real             *shiftvec,*fshift,*x,*f;
 637     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 638     real             scratch[4*DIM];
 639     __m128           fscal,rcutoff,rcutoff2,jidxall;
 640     int              vdwioffset1;
 641     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 642     int              vdwioffset2;
 643     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 644     int              vdwioffset3;
 645     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 646     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 647     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 648     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 649     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 650     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 651     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 652     real             *charge;
 653     __m128i          vfitab;
 654     __m128i          ifour       = _mm_set1_epi32(4);
 655     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 656     real             *vftab;
 657     __m128           dummy_mask,cutoff_mask;
 658     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 659     __m128           one     = _mm_set1_ps(1.0);
 660     __m128           two     = _mm_set1_ps(2.0);
 661     x                = xx[0];
 662     f                = ff[0];
 663
 664     nri              = nlist->nri;
 665     iinr             = nlist->iinr;
 666     jindex           = nlist->jindex;
 667     jjnr             = nlist->jjnr;
 668     shiftidx         = nlist->shift;
 669     gid              = nlist->gid;
 670     shiftvec         = fr->shift_vec[0];
 671     fshift           = fr->fshift[0];
 672     facel            = _mm_set1_ps(fr->epsfac);
 673     charge           = mdatoms->chargeA;
 674
 675     vftab            = kernel_data->table_elec->data;
 676     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 677
 678     /* Setup water-specific parameters */
 679     inr              = nlist->iinr[0];
 680     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 681     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 682     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 683
 684     /* Avoid stupid compiler warnings */
 685     jnrA = jnrB = jnrC = jnrD = 0;
 686     j_coord_offsetA = 0;
 687     j_coord_offsetB = 0;
 688     j_coord_offsetC = 0;
 689     j_coord_offsetD = 0;
 690
 691     outeriter        = 0;
 692     inneriter        = 0;
 693
 694     for(iidx=0;iidx<4*DIM;iidx++)
 695     {
 696         scratch[iidx] = 0.0;
 697     }
 698
 699     /* Start outer loop over neighborlists */
 700     for(iidx=0; iidx<nri; iidx++)
 701     {
 702         /* Load shift vector for this list */
 703         i_shift_offset   = DIM*shiftidx[iidx];
 704
 705         /* Load limits for loop over neighbors */
 706         j_index_start    = jindex[iidx];
 707         j_index_end      = jindex[iidx+1];
 708
 709         /* Get outer coordinate index */
 710         inr              = iinr[iidx];
 711         i_coord_offset   = DIM*inr;
 712
 713         /* Load i particle coords and add shift vector */
 714         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 715                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 716
 717         fix1             = _mm_setzero_ps();
 718         fiy1             = _mm_setzero_ps();
 719         fiz1             = _mm_setzero_ps();
 720         fix2             = _mm_setzero_ps();
 721         fiy2             = _mm_setzero_ps();
 722         fiz2             = _mm_setzero_ps();
 723         fix3             = _mm_setzero_ps();
 724         fiy3             = _mm_setzero_ps();
 725         fiz3             = _mm_setzero_ps();
 726
 727         /* Start inner kernel loop */
 728         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 729         {
 730
 731             /* Get j neighbor index, and coordinate index */
 732             jnrA             = jjnr[jidx];
 733             jnrB             = jjnr[jidx+1];
 734             jnrC             = jjnr[jidx+2];
 735             jnrD             = jjnr[jidx+3];
 736             j_coord_offsetA  = DIM*jnrA;
 737             j_coord_offsetB  = DIM*jnrB;
 738             j_coord_offsetC  = DIM*jnrC;
 739             j_coord_offsetD  = DIM*jnrD;
 740
 741             /* load j atom coordinates */
 742             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 743                                               x+j_coord_offsetC,x+j_coord_offsetD,
 744                                               &jx0,&jy0,&jz0);
 745
 746             /* Calculate displacement vector */
 747             dx10             = _mm_sub_ps(ix1,jx0);
 748             dy10             = _mm_sub_ps(iy1,jy0);
 749             dz10             = _mm_sub_ps(iz1,jz0);
 750             dx20             = _mm_sub_ps(ix2,jx0);
 751             dy20             = _mm_sub_ps(iy2,jy0);
 752             dz20             = _mm_sub_ps(iz2,jz0);
 753             dx30             = _mm_sub_ps(ix3,jx0);
 754             dy30             = _mm_sub_ps(iy3,jy0);
 755             dz30             = _mm_sub_ps(iz3,jz0);
 756
 757             /* Calculate squared distance and things based on it */
 758             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 759             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 760             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 761
 762             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 763             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 764             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 765
 766             /* Load parameters for j particles */
 767             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 768                                                               charge+jnrC+0,charge+jnrD+0);
 769
 770             fjx0             = _mm_setzero_ps();
 771             fjy0             = _mm_setzero_ps();
 772             fjz0             = _mm_setzero_ps();
 773
 774             /**************************
 775              * CALCULATE INTERACTIONS *
 776              **************************/
 777
 778             r10              = _mm_mul_ps(rsq10,rinv10);
 779
 780             /* Compute parameters for interactions between i and j atoms */
 781             qq10             = _mm_mul_ps(iq1,jq0);
 782
 783             /* Calculate table index by multiplying r with table scale and truncate to integer */
 784             rt               = _mm_mul_ps(r10,vftabscale);
 785             vfitab           = _mm_cvttps_epi32(rt);
 786 #ifdef __XOP__
 787             vfeps            = _mm_frcz_ps(rt);
 788 #else
 789             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 790 #endif
 791             twovfeps         = _mm_add_ps(vfeps,vfeps);
 792             vfitab           = _mm_slli_epi32(vfitab,2);
 793
 794             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 795             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 796             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 797             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 798             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 799             _MM_TRANSPOSE4_PS(Y,F,G,H);
 800             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 801             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 802             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 803
 804             fscal            = felec;
 805
 806              /* Update vectorial force */
 807             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 808             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 809             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 810
 811             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 812             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 813             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 814
 815             /**************************
 816              * CALCULATE INTERACTIONS *
 817              **************************/
 818
 819             r20              = _mm_mul_ps(rsq20,rinv20);
 820
 821             /* Compute parameters for interactions between i and j atoms */
 822             qq20             = _mm_mul_ps(iq2,jq0);
 823
 824             /* Calculate table index by multiplying r with table scale and truncate to integer */
 825             rt               = _mm_mul_ps(r20,vftabscale);
 826             vfitab           = _mm_cvttps_epi32(rt);
 827 #ifdef __XOP__
 828             vfeps            = _mm_frcz_ps(rt);
 829 #else
 830             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 831 #endif
 832             twovfeps         = _mm_add_ps(vfeps,vfeps);
 833             vfitab           = _mm_slli_epi32(vfitab,2);
 834
 835             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 836             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 837             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 838             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 839             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 840             _MM_TRANSPOSE4_PS(Y,F,G,H);
 841             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 842             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 843             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 844
 845             fscal            = felec;
 846
 847              /* Update vectorial force */
 848             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 849             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 850             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 851
 852             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 853             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 854             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 855
 856             /**************************
 857              * CALCULATE INTERACTIONS *
 858              **************************/
 859
 860             r30              = _mm_mul_ps(rsq30,rinv30);
 861
 862             /* Compute parameters for interactions between i and j atoms */
 863             qq30             = _mm_mul_ps(iq3,jq0);
 864
 865             /* Calculate table index by multiplying r with table scale and truncate to integer */
 866             rt               = _mm_mul_ps(r30,vftabscale);
 867             vfitab           = _mm_cvttps_epi32(rt);
 868 #ifdef __XOP__
 869             vfeps            = _mm_frcz_ps(rt);
 870 #else
 871             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 872 #endif
 873             twovfeps         = _mm_add_ps(vfeps,vfeps);
 874             vfitab           = _mm_slli_epi32(vfitab,2);
 875
 876             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 877             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 878             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 879             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 880             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 881             _MM_TRANSPOSE4_PS(Y,F,G,H);
 882             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 883             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 884             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 885
 886             fscal            = felec;
 887
 888              /* Update vectorial force */
 889             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 890             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 891             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 892
 893             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 894             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 895             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 896
 897             fjptrA             = f+j_coord_offsetA;
 898             fjptrB             = f+j_coord_offsetB;
 899             fjptrC             = f+j_coord_offsetC;
 900             fjptrD             = f+j_coord_offsetD;
 901
 902             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 903
 904             /* Inner loop uses 126 flops */
 905         }
 906
 907         if(jidx<j_index_end)
 908         {
 909
 910             /* Get j neighbor index, and coordinate index */
 911             jnrlistA         = jjnr[jidx];
 912             jnrlistB         = jjnr[jidx+1];
 913             jnrlistC         = jjnr[jidx+2];
 914             jnrlistD         = jjnr[jidx+3];
 915             /* Sign of each element will be negative for non-real atoms.
 916              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 917              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 918              */
 919             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 920             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 921             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 922             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 923             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 924             j_coord_offsetA  = DIM*jnrA;
 925             j_coord_offsetB  = DIM*jnrB;
 926             j_coord_offsetC  = DIM*jnrC;
 927             j_coord_offsetD  = DIM*jnrD;
 928
 929             /* load j atom coordinates */
 930             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 931                                               x+j_coord_offsetC,x+j_coord_offsetD,
 932                                               &jx0,&jy0,&jz0);
 933
 934             /* Calculate displacement vector */
 935             dx10             = _mm_sub_ps(ix1,jx0);
 936             dy10             = _mm_sub_ps(iy1,jy0);
 937             dz10             = _mm_sub_ps(iz1,jz0);
 938             dx20             = _mm_sub_ps(ix2,jx0);
 939             dy20             = _mm_sub_ps(iy2,jy0);
 940             dz20             = _mm_sub_ps(iz2,jz0);
 941             dx30             = _mm_sub_ps(ix3,jx0);
 942             dy30             = _mm_sub_ps(iy3,jy0);
 943             dz30             = _mm_sub_ps(iz3,jz0);
 944
 945             /* Calculate squared distance and things based on it */
 946             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 947             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 948             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 949
 950             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 951             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 952             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 953
 954             /* Load parameters for j particles */
 955             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 956                                                               charge+jnrC+0,charge+jnrD+0);
 957
 958             fjx0             = _mm_setzero_ps();
 959             fjy0             = _mm_setzero_ps();
 960             fjz0             = _mm_setzero_ps();
 961
 962             /**************************
 963              * CALCULATE INTERACTIONS *
 964              **************************/
 965
 966             r10              = _mm_mul_ps(rsq10,rinv10);
 967             r10              = _mm_andnot_ps(dummy_mask,r10);
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq10             = _mm_mul_ps(iq1,jq0);
 971
 972             /* Calculate table index by multiplying r with table scale and truncate to integer */
 973             rt               = _mm_mul_ps(r10,vftabscale);
 974             vfitab           = _mm_cvttps_epi32(rt);
 975 #ifdef __XOP__
 976             vfeps            = _mm_frcz_ps(rt);
 977 #else
 978             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 979 #endif
 980             twovfeps         = _mm_add_ps(vfeps,vfeps);
 981             vfitab           = _mm_slli_epi32(vfitab,2);
 982
 983             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 984             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 985             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 986             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 987             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 988             _MM_TRANSPOSE4_PS(Y,F,G,H);
 989             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 990             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 991             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 992
 993             fscal            = felec;
 994
 995             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 996
 997              /* Update vectorial force */
 998             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 999             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1000             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1001
1002             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1003             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1004             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             r20              = _mm_mul_ps(rsq20,rinv20);
1011             r20              = _mm_andnot_ps(dummy_mask,r20);
1012
1013             /* Compute parameters for interactions between i and j atoms */
1014             qq20             = _mm_mul_ps(iq2,jq0);
1015
1016             /* Calculate table index by multiplying r with table scale and truncate to integer */
1017             rt               = _mm_mul_ps(r20,vftabscale);
1018             vfitab           = _mm_cvttps_epi32(rt);
1019 #ifdef __XOP__
1020             vfeps            = _mm_frcz_ps(rt);
1021 #else
1022             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1023 #endif
1024             twovfeps         = _mm_add_ps(vfeps,vfeps);
1025             vfitab           = _mm_slli_epi32(vfitab,2);
1026
1027             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1028             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1029             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1030             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1031             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1032             _MM_TRANSPOSE4_PS(Y,F,G,H);
1033             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1034             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1035             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1036
1037             fscal            = felec;
1038
1039             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1040
1041              /* Update vectorial force */
1042             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1043             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1044             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1045
1046             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1047             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1048             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             r30              = _mm_mul_ps(rsq30,rinv30);
1055             r30              = _mm_andnot_ps(dummy_mask,r30);
1056
1057             /* Compute parameters for interactions between i and j atoms */
1058             qq30             = _mm_mul_ps(iq3,jq0);
1059
1060             /* Calculate table index by multiplying r with table scale and truncate to integer */
1061             rt               = _mm_mul_ps(r30,vftabscale);
1062             vfitab           = _mm_cvttps_epi32(rt);
1063 #ifdef __XOP__
1064             vfeps            = _mm_frcz_ps(rt);
1065 #else
1066             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1067 #endif
1068             twovfeps         = _mm_add_ps(vfeps,vfeps);
1069             vfitab           = _mm_slli_epi32(vfitab,2);
1070
1071             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1072             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1073             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1074             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1075             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1076             _MM_TRANSPOSE4_PS(Y,F,G,H);
1077             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1078             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1079             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1080
1081             fscal            = felec;
1082
1083             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1084
1085              /* Update vectorial force */
1086             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1087             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1088             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1089
1090             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1091             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1092             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1093
1094             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1095             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1096             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1097             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1098
1099             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1100
1101             /* Inner loop uses 129 flops */
1102         }
1103
1104         /* End of innermost loop */
1105
1106         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1107                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1108
1109         /* Increment number of inner iterations */
1110         inneriter                  += j_index_end - j_index_start;
1111
1112         /* Outer loop uses 18 flops */
1113     }
1114
1115     /* Increment number of outer iterations */
1116     outeriter        += nri;
1117
1118     /* Update outer/inner flops */
1119
1120     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*129);
1121 }