src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_VF_avx_128_fma_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_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              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  94     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  95     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  96     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  97     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  98     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  99     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 100     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 101     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 102     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 103     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 104     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 105     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 106     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 107     real             *charge;
 108     int              nvdwtype;
 109     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 110     int              *vdwtype;
 111     real             *vdwparam;
 112     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 113     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 114     __m128i          vfitab;
 115     __m128i          ifour       = _mm_set1_epi32(4);
 116     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 117     real             *vftab;
 118     __m128           dummy_mask,cutoff_mask;
 119     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 120     __m128           one     = _mm_set1_ps(1.0);
 121     __m128           two     = _mm_set1_ps(2.0);
 122     x                = xx[0];
 123     f                = ff[0];
 124
 125     nri              = nlist->nri;
 126     iinr             = nlist->iinr;
 127     jindex           = nlist->jindex;
 128     jjnr             = nlist->jjnr;
 129     shiftidx         = nlist->shift;
 130     gid              = nlist->gid;
 131     shiftvec         = fr->shift_vec[0];
 132     fshift           = fr->fshift[0];
 133     facel            = _mm_set1_ps(fr->epsfac);
 134     charge           = mdatoms->chargeA;
 135     nvdwtype         = fr->ntype;
 136     vdwparam         = fr->nbfp;
 137     vdwtype          = mdatoms->typeA;
 138
 139     vftab            = kernel_data->table_vdw->data;
 140     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 145     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 146     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 147     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149     jq0              = _mm_set1_ps(charge[inr+0]);
 150     jq1              = _mm_set1_ps(charge[inr+1]);
 151     jq2              = _mm_set1_ps(charge[inr+2]);
 152     vdwjidx0A        = 2*vdwtype[inr+0];
 153     qq00             = _mm_mul_ps(iq0,jq0);
 154     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 155     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 156     qq01             = _mm_mul_ps(iq0,jq1);
 157     qq02             = _mm_mul_ps(iq0,jq2);
 158     qq10             = _mm_mul_ps(iq1,jq0);
 159     qq11             = _mm_mul_ps(iq1,jq1);
 160     qq12             = _mm_mul_ps(iq1,jq2);
 161     qq20             = _mm_mul_ps(iq2,jq0);
 162     qq21             = _mm_mul_ps(iq2,jq1);
 163     qq22             = _mm_mul_ps(iq2,jq2);
 164
 165     /* Avoid stupid compiler warnings */
 166     jnrA = jnrB = jnrC = jnrD = 0;
 167     j_coord_offsetA = 0;
 168     j_coord_offsetB = 0;
 169     j_coord_offsetC = 0;
 170     j_coord_offsetD = 0;
 171
 172     outeriter        = 0;
 173     inneriter        = 0;
 174
 175     for(iidx=0;iidx<4*DIM;iidx++)
 176     {
 177         scratch[iidx] = 0.0;
 178     }
 179
 180     /* Start outer loop over neighborlists */
 181     for(iidx=0; iidx<nri; iidx++)
 182     {
 183         /* Load shift vector for this list */
 184         i_shift_offset   = DIM*shiftidx[iidx];
 185
 186         /* Load limits for loop over neighbors */
 187         j_index_start    = jindex[iidx];
 188         j_index_end      = jindex[iidx+1];
 189
 190         /* Get outer coordinate index */
 191         inr              = iinr[iidx];
 192         i_coord_offset   = DIM*inr;
 193
 194         /* Load i particle coords and add shift vector */
 195         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 196                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 197
 198         fix0             = _mm_setzero_ps();
 199         fiy0             = _mm_setzero_ps();
 200         fiz0             = _mm_setzero_ps();
 201         fix1             = _mm_setzero_ps();
 202         fiy1             = _mm_setzero_ps();
 203         fiz1             = _mm_setzero_ps();
 204         fix2             = _mm_setzero_ps();
 205         fiy2             = _mm_setzero_ps();
 206         fiz2             = _mm_setzero_ps();
 207
 208         /* Reset potential sums */
 209         velecsum         = _mm_setzero_ps();
 210         vvdwsum          = _mm_setzero_ps();
 211
 212         /* Start inner kernel loop */
 213         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 214         {
 215
 216             /* Get j neighbor index, and coordinate index */
 217             jnrA             = jjnr[jidx];
 218             jnrB             = jjnr[jidx+1];
 219             jnrC             = jjnr[jidx+2];
 220             jnrD             = jjnr[jidx+3];
 221             j_coord_offsetA  = DIM*jnrA;
 222             j_coord_offsetB  = DIM*jnrB;
 223             j_coord_offsetC  = DIM*jnrC;
 224             j_coord_offsetD  = DIM*jnrD;
 225
 226             /* load j atom coordinates */
 227             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 228                                               x+j_coord_offsetC,x+j_coord_offsetD,
 229                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 230
 231             /* Calculate displacement vector */
 232             dx00             = _mm_sub_ps(ix0,jx0);
 233             dy00             = _mm_sub_ps(iy0,jy0);
 234             dz00             = _mm_sub_ps(iz0,jz0);
 235             dx01             = _mm_sub_ps(ix0,jx1);
 236             dy01             = _mm_sub_ps(iy0,jy1);
 237             dz01             = _mm_sub_ps(iz0,jz1);
 238             dx02             = _mm_sub_ps(ix0,jx2);
 239             dy02             = _mm_sub_ps(iy0,jy2);
 240             dz02             = _mm_sub_ps(iz0,jz2);
 241             dx10             = _mm_sub_ps(ix1,jx0);
 242             dy10             = _mm_sub_ps(iy1,jy0);
 243             dz10             = _mm_sub_ps(iz1,jz0);
 244             dx11             = _mm_sub_ps(ix1,jx1);
 245             dy11             = _mm_sub_ps(iy1,jy1);
 246             dz11             = _mm_sub_ps(iz1,jz1);
 247             dx12             = _mm_sub_ps(ix1,jx2);
 248             dy12             = _mm_sub_ps(iy1,jy2);
 249             dz12             = _mm_sub_ps(iz1,jz2);
 250             dx20             = _mm_sub_ps(ix2,jx0);
 251             dy20             = _mm_sub_ps(iy2,jy0);
 252             dz20             = _mm_sub_ps(iz2,jz0);
 253             dx21             = _mm_sub_ps(ix2,jx1);
 254             dy21             = _mm_sub_ps(iy2,jy1);
 255             dz21             = _mm_sub_ps(iz2,jz1);
 256             dx22             = _mm_sub_ps(ix2,jx2);
 257             dy22             = _mm_sub_ps(iy2,jy2);
 258             dz22             = _mm_sub_ps(iz2,jz2);
 259
 260             /* Calculate squared distance and things based on it */
 261             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 262             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 263             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 264             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 265             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 266             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 267             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 268             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 269             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 270
 271             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 272             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 273             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 274             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 275             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 276             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 277             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 278             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 279             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 280
 281             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 282             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 283             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 284             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 285             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 286             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 287             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 288             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 289             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 290
 291             fjx0             = _mm_setzero_ps();
 292             fjy0             = _mm_setzero_ps();
 293             fjz0             = _mm_setzero_ps();
 294             fjx1             = _mm_setzero_ps();
 295             fjy1             = _mm_setzero_ps();
 296             fjz1             = _mm_setzero_ps();
 297             fjx2             = _mm_setzero_ps();
 298             fjy2             = _mm_setzero_ps();
 299             fjz2             = _mm_setzero_ps();
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             r00              = _mm_mul_ps(rsq00,rinv00);
 306
 307             /* Calculate table index by multiplying r with table scale and truncate to integer */
 308             rt               = _mm_mul_ps(r00,vftabscale);
 309             vfitab           = _mm_cvttps_epi32(rt);
 310 #ifdef __XOP__
 311             vfeps            = _mm_frcz_ps(rt);
 312 #else
 313             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 314 #endif
 315             twovfeps         = _mm_add_ps(vfeps,vfeps);
 316             vfitab           = _mm_slli_epi32(vfitab,3);
 317
 318             /* COULOMB ELECTROSTATICS */
 319             velec            = _mm_mul_ps(qq00,rinv00);
 320             felec            = _mm_mul_ps(velec,rinvsq00);
 321
 322             /* CUBIC SPLINE TABLE DISPERSION */
 323             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 324             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 325             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 326             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 327             _MM_TRANSPOSE4_PS(Y,F,G,H);
 328             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 329             VV               = _mm_macc_ps(vfeps,Fp,Y);
 330             vvdw6            = _mm_mul_ps(c6_00,VV);
 331             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 332             fvdw6            = _mm_mul_ps(c6_00,FF);
 333
 334             /* CUBIC SPLINE TABLE REPULSION */
 335             vfitab           = _mm_add_epi32(vfitab,ifour);
 336             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 337             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 338             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 339             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 340             _MM_TRANSPOSE4_PS(Y,F,G,H);
 341             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 342             VV               = _mm_macc_ps(vfeps,Fp,Y);
 343             vvdw12           = _mm_mul_ps(c12_00,VV);
 344             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 345             fvdw12           = _mm_mul_ps(c12_00,FF);
 346             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 347             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velecsum         = _mm_add_ps(velecsum,velec);
 351             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 352
 353             fscal            = _mm_add_ps(felec,fvdw);
 354
 355              /* Update vectorial force */
 356             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 357             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 358             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 359
 360             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 361             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 362             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             /* COULOMB ELECTROSTATICS */
 369             velec            = _mm_mul_ps(qq01,rinv01);
 370             felec            = _mm_mul_ps(velec,rinvsq01);
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm_add_ps(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377              /* Update vectorial force */
 378             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 379             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 380             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 381
 382             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 383             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 384             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             /* COULOMB ELECTROSTATICS */
 391             velec            = _mm_mul_ps(qq02,rinv02);
 392             felec            = _mm_mul_ps(velec,rinvsq02);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velecsum         = _mm_add_ps(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399              /* Update vectorial force */
 400             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 401             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 402             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 403
 404             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 405             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 406             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 407
 408             /**************************
 409              * CALCULATE INTERACTIONS *
 410              **************************/
 411
 412             /* COULOMB ELECTROSTATICS */
 413             velec            = _mm_mul_ps(qq10,rinv10);
 414             felec            = _mm_mul_ps(velec,rinvsq10);
 415
 416             /* Update potential sum for this i atom from the interaction with this j atom. */
 417             velecsum         = _mm_add_ps(velecsum,velec);
 418
 419             fscal            = felec;
 420
 421              /* Update vectorial force */
 422             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 423             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 424             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 425
 426             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 427             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 428             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 429
 430             /**************************
 431              * CALCULATE INTERACTIONS *
 432              **************************/
 433
 434             /* COULOMB ELECTROSTATICS */
 435             velec            = _mm_mul_ps(qq11,rinv11);
 436             felec            = _mm_mul_ps(velec,rinvsq11);
 437
 438             /* Update potential sum for this i atom from the interaction with this j atom. */
 439             velecsum         = _mm_add_ps(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443              /* Update vectorial force */
 444             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 445             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 446             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 447
 448             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 449             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 450             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             /* COULOMB ELECTROSTATICS */
 457             velec            = _mm_mul_ps(qq12,rinv12);
 458             felec            = _mm_mul_ps(velec,rinvsq12);
 459
 460             /* Update potential sum for this i atom from the interaction with this j atom. */
 461             velecsum         = _mm_add_ps(velecsum,velec);
 462
 463             fscal            = felec;
 464
 465              /* Update vectorial force */
 466             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 467             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 468             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 469
 470             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 471             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 472             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             /* COULOMB ELECTROSTATICS */
 479             velec            = _mm_mul_ps(qq20,rinv20);
 480             felec            = _mm_mul_ps(velec,rinvsq20);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velecsum         = _mm_add_ps(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487              /* Update vectorial force */
 488             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 489             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 490             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 491
 492             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 493             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 494             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             /* COULOMB ELECTROSTATICS */
 501             velec            = _mm_mul_ps(qq21,rinv21);
 502             felec            = _mm_mul_ps(velec,rinvsq21);
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velecsum         = _mm_add_ps(velecsum,velec);
 506
 507             fscal            = felec;
 508
 509              /* Update vectorial force */
 510             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 511             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 512             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 513
 514             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 515             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 516             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             /* COULOMB ELECTROSTATICS */
 523             velec            = _mm_mul_ps(qq22,rinv22);
 524             felec            = _mm_mul_ps(velec,rinvsq22);
 525
 526             /* Update potential sum for this i atom from the interaction with this j atom. */
 527             velecsum         = _mm_add_ps(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531              /* Update vectorial force */
 532             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 533             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 534             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 535
 536             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 537             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 538             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 539
 540             fjptrA             = f+j_coord_offsetA;
 541             fjptrB             = f+j_coord_offsetB;
 542             fjptrC             = f+j_coord_offsetC;
 543             fjptrD             = f+j_coord_offsetD;
 544
 545             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 546                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 547
 548             /* Inner loop uses 314 flops */
 549         }
 550
 551         if(jidx<j_index_end)
 552         {
 553
 554             /* Get j neighbor index, and coordinate index */
 555             jnrlistA         = jjnr[jidx];
 556             jnrlistB         = jjnr[jidx+1];
 557             jnrlistC         = jjnr[jidx+2];
 558             jnrlistD         = jjnr[jidx+3];
 559             /* Sign of each element will be negative for non-real atoms.
 560              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 561              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 562              */
 563             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 564             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 565             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 566             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 567             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 568             j_coord_offsetA  = DIM*jnrA;
 569             j_coord_offsetB  = DIM*jnrB;
 570             j_coord_offsetC  = DIM*jnrC;
 571             j_coord_offsetD  = DIM*jnrD;
 572
 573             /* load j atom coordinates */
 574             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 575                                               x+j_coord_offsetC,x+j_coord_offsetD,
 576                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 577
 578             /* Calculate displacement vector */
 579             dx00             = _mm_sub_ps(ix0,jx0);
 580             dy00             = _mm_sub_ps(iy0,jy0);
 581             dz00             = _mm_sub_ps(iz0,jz0);
 582             dx01             = _mm_sub_ps(ix0,jx1);
 583             dy01             = _mm_sub_ps(iy0,jy1);
 584             dz01             = _mm_sub_ps(iz0,jz1);
 585             dx02             = _mm_sub_ps(ix0,jx2);
 586             dy02             = _mm_sub_ps(iy0,jy2);
 587             dz02             = _mm_sub_ps(iz0,jz2);
 588             dx10             = _mm_sub_ps(ix1,jx0);
 589             dy10             = _mm_sub_ps(iy1,jy0);
 590             dz10             = _mm_sub_ps(iz1,jz0);
 591             dx11             = _mm_sub_ps(ix1,jx1);
 592             dy11             = _mm_sub_ps(iy1,jy1);
 593             dz11             = _mm_sub_ps(iz1,jz1);
 594             dx12             = _mm_sub_ps(ix1,jx2);
 595             dy12             = _mm_sub_ps(iy1,jy2);
 596             dz12             = _mm_sub_ps(iz1,jz2);
 597             dx20             = _mm_sub_ps(ix2,jx0);
 598             dy20             = _mm_sub_ps(iy2,jy0);
 599             dz20             = _mm_sub_ps(iz2,jz0);
 600             dx21             = _mm_sub_ps(ix2,jx1);
 601             dy21             = _mm_sub_ps(iy2,jy1);
 602             dz21             = _mm_sub_ps(iz2,jz1);
 603             dx22             = _mm_sub_ps(ix2,jx2);
 604             dy22             = _mm_sub_ps(iy2,jy2);
 605             dz22             = _mm_sub_ps(iz2,jz2);
 606
 607             /* Calculate squared distance and things based on it */
 608             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 609             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 610             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 611             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 612             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 613             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 614             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 615             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 616             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 617
 618             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 619             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 620             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 621             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 622             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 623             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 624             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 625             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 626             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 627
 628             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 629             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 630             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 631             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 632             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 633             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 634             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 635             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 636             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 637
 638             fjx0             = _mm_setzero_ps();
 639             fjy0             = _mm_setzero_ps();
 640             fjz0             = _mm_setzero_ps();
 641             fjx1             = _mm_setzero_ps();
 642             fjy1             = _mm_setzero_ps();
 643             fjz1             = _mm_setzero_ps();
 644             fjx2             = _mm_setzero_ps();
 645             fjy2             = _mm_setzero_ps();
 646             fjz2             = _mm_setzero_ps();
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             r00              = _mm_mul_ps(rsq00,rinv00);
 653             r00              = _mm_andnot_ps(dummy_mask,r00);
 654
 655             /* Calculate table index by multiplying r with table scale and truncate to integer */
 656             rt               = _mm_mul_ps(r00,vftabscale);
 657             vfitab           = _mm_cvttps_epi32(rt);
 658 #ifdef __XOP__
 659             vfeps            = _mm_frcz_ps(rt);
 660 #else
 661             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 662 #endif
 663             twovfeps         = _mm_add_ps(vfeps,vfeps);
 664             vfitab           = _mm_slli_epi32(vfitab,3);
 665
 666             /* COULOMB ELECTROSTATICS */
 667             velec            = _mm_mul_ps(qq00,rinv00);
 668             felec            = _mm_mul_ps(velec,rinvsq00);
 669
 670             /* CUBIC SPLINE TABLE DISPERSION */
 671             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 672             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 673             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 674             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 675             _MM_TRANSPOSE4_PS(Y,F,G,H);
 676             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 677             VV               = _mm_macc_ps(vfeps,Fp,Y);
 678             vvdw6            = _mm_mul_ps(c6_00,VV);
 679             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 680             fvdw6            = _mm_mul_ps(c6_00,FF);
 681
 682             /* CUBIC SPLINE TABLE REPULSION */
 683             vfitab           = _mm_add_epi32(vfitab,ifour);
 684             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 685             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 686             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 687             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 688             _MM_TRANSPOSE4_PS(Y,F,G,H);
 689             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 690             VV               = _mm_macc_ps(vfeps,Fp,Y);
 691             vvdw12           = _mm_mul_ps(c12_00,VV);
 692             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 693             fvdw12           = _mm_mul_ps(c12_00,FF);
 694             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 695             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 696
 697             /* Update potential sum for this i atom from the interaction with this j atom. */
 698             velec            = _mm_andnot_ps(dummy_mask,velec);
 699             velecsum         = _mm_add_ps(velecsum,velec);
 700             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 701             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 702
 703             fscal            = _mm_add_ps(felec,fvdw);
 704
 705             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 706
 707              /* Update vectorial force */
 708             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 709             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 710             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 711
 712             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 713             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 714             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 715
 716             /**************************
 717              * CALCULATE INTERACTIONS *
 718              **************************/
 719
 720             /* COULOMB ELECTROSTATICS */
 721             velec            = _mm_mul_ps(qq01,rinv01);
 722             felec            = _mm_mul_ps(velec,rinvsq01);
 723
 724             /* Update potential sum for this i atom from the interaction with this j atom. */
 725             velec            = _mm_andnot_ps(dummy_mask,velec);
 726             velecsum         = _mm_add_ps(velecsum,velec);
 727
 728             fscal            = felec;
 729
 730             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 731
 732              /* Update vectorial force */
 733             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 734             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 735             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 736
 737             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 738             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 739             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 740
 741             /**************************
 742              * CALCULATE INTERACTIONS *
 743              **************************/
 744
 745             /* COULOMB ELECTROSTATICS */
 746             velec            = _mm_mul_ps(qq02,rinv02);
 747             felec            = _mm_mul_ps(velec,rinvsq02);
 748
 749             /* Update potential sum for this i atom from the interaction with this j atom. */
 750             velec            = _mm_andnot_ps(dummy_mask,velec);
 751             velecsum         = _mm_add_ps(velecsum,velec);
 752
 753             fscal            = felec;
 754
 755             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 756
 757              /* Update vectorial force */
 758             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 759             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 760             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 761
 762             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 763             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 764             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 765
 766             /**************************
 767              * CALCULATE INTERACTIONS *
 768              **************************/
 769
 770             /* COULOMB ELECTROSTATICS */
 771             velec            = _mm_mul_ps(qq10,rinv10);
 772             felec            = _mm_mul_ps(velec,rinvsq10);
 773
 774             /* Update potential sum for this i atom from the interaction with this j atom. */
 775             velec            = _mm_andnot_ps(dummy_mask,velec);
 776             velecsum         = _mm_add_ps(velecsum,velec);
 777
 778             fscal            = felec;
 779
 780             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 781
 782              /* Update vectorial force */
 783             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 784             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 785             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 786
 787             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 788             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 789             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             /* COULOMB ELECTROSTATICS */
 796             velec            = _mm_mul_ps(qq11,rinv11);
 797             felec            = _mm_mul_ps(velec,rinvsq11);
 798
 799             /* Update potential sum for this i atom from the interaction with this j atom. */
 800             velec            = _mm_andnot_ps(dummy_mask,velec);
 801             velecsum         = _mm_add_ps(velecsum,velec);
 802
 803             fscal            = felec;
 804
 805             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 806
 807              /* Update vectorial force */
 808             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 809             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 810             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 811
 812             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 813             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 814             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             /* COULOMB ELECTROSTATICS */
 821             velec            = _mm_mul_ps(qq12,rinv12);
 822             felec            = _mm_mul_ps(velec,rinvsq12);
 823
 824             /* Update potential sum for this i atom from the interaction with this j atom. */
 825             velec            = _mm_andnot_ps(dummy_mask,velec);
 826             velecsum         = _mm_add_ps(velecsum,velec);
 827
 828             fscal            = felec;
 829
 830             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 831
 832              /* Update vectorial force */
 833             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 834             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 835             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 836
 837             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 838             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 839             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             /* COULOMB ELECTROSTATICS */
 846             velec            = _mm_mul_ps(qq20,rinv20);
 847             felec            = _mm_mul_ps(velec,rinvsq20);
 848
 849             /* Update potential sum for this i atom from the interaction with this j atom. */
 850             velec            = _mm_andnot_ps(dummy_mask,velec);
 851             velecsum         = _mm_add_ps(velecsum,velec);
 852
 853             fscal            = felec;
 854
 855             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 856
 857              /* Update vectorial force */
 858             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 859             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 860             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 861
 862             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 863             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 864             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             /* COULOMB ELECTROSTATICS */
 871             velec            = _mm_mul_ps(qq21,rinv21);
 872             felec            = _mm_mul_ps(velec,rinvsq21);
 873
 874             /* Update potential sum for this i atom from the interaction with this j atom. */
 875             velec            = _mm_andnot_ps(dummy_mask,velec);
 876             velecsum         = _mm_add_ps(velecsum,velec);
 877
 878             fscal            = felec;
 879
 880             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 881
 882              /* Update vectorial force */
 883             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 884             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 885             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 886
 887             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 888             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 889             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             /* COULOMB ELECTROSTATICS */
 896             velec            = _mm_mul_ps(qq22,rinv22);
 897             felec            = _mm_mul_ps(velec,rinvsq22);
 898
 899             /* Update potential sum for this i atom from the interaction with this j atom. */
 900             velec            = _mm_andnot_ps(dummy_mask,velec);
 901             velecsum         = _mm_add_ps(velecsum,velec);
 902
 903             fscal            = felec;
 904
 905             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 906
 907              /* Update vectorial force */
 908             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 909             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 910             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 911
 912             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 913             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 914             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 915
 916             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 917             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 918             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 919             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 920
 921             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 922                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 923
 924             /* Inner loop uses 315 flops */
 925         }
 926
 927         /* End of innermost loop */
 928
 929         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 930                                               f+i_coord_offset,fshift+i_shift_offset);
 931
 932         ggid                        = gid[iidx];
 933         /* Update potential energies */
 934         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 935         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 936
 937         /* Increment number of inner iterations */
 938         inneriter                  += j_index_end - j_index_start;
 939
 940         /* Outer loop uses 20 flops */
 941     }
 942
 943     /* Increment number of outer iterations */
 944     outeriter        += nri;
 945
 946     /* Update outer/inner flops */
 947
 948     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*315);
 949 }
 950 /*
 951  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_F_avx_128_fma_single
 952  * Electrostatics interaction: Coulomb
 953  * VdW interaction:            CubicSplineTable
 954  * Geometry:                   Water3-Water3
 955  * Calculate force/pot:        Force
 956  */
 957 void
 958 nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_F_avx_128_fma_single
 959                     (t_nblist                    * gmx_restrict       nlist,
 960                      rvec                        * gmx_restrict          xx,
 961                      rvec                        * gmx_restrict          ff,
 962                      t_forcerec                  * gmx_restrict          fr,
 963                      t_mdatoms                   * gmx_restrict     mdatoms,
 964                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 965                      t_nrnb                      * gmx_restrict        nrnb)
 966 {
 967     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 968      * just 0 for non-waters.
 969      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 970      * jnr indices corresponding to data put in the four positions in the SIMD register.
 971      */
 972     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 973     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 974     int              jnrA,jnrB,jnrC,jnrD;
 975     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 976     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 977     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 978     real             rcutoff_scalar;
 979     real             *shiftvec,*fshift,*x,*f;
 980     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 981     real             scratch[4*DIM];
 982     __m128           fscal,rcutoff,rcutoff2,jidxall;
 983     int              vdwioffset0;
 984     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 985     int              vdwioffset1;
 986     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 987     int              vdwioffset2;
 988     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 989     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 990     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 991     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 992     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 993     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 994     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 995     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 996     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 997     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 998     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 999     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1000     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1001     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1002     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1003     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1004     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1005     real             *charge;
1006     int              nvdwtype;
1007     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1008     int              *vdwtype;
1009     real             *vdwparam;
1010     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1011     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1012     __m128i          vfitab;
1013     __m128i          ifour       = _mm_set1_epi32(4);
1014     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
1015     real             *vftab;
1016     __m128           dummy_mask,cutoff_mask;
1017     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1018     __m128           one     = _mm_set1_ps(1.0);
1019     __m128           two     = _mm_set1_ps(2.0);
1020     x                = xx[0];
1021     f                = ff[0];
1022
1023     nri              = nlist->nri;
1024     iinr             = nlist->iinr;
1025     jindex           = nlist->jindex;
1026     jjnr             = nlist->jjnr;
1027     shiftidx         = nlist->shift;
1028     gid              = nlist->gid;
1029     shiftvec         = fr->shift_vec[0];
1030     fshift           = fr->fshift[0];
1031     facel            = _mm_set1_ps(fr->epsfac);
1032     charge           = mdatoms->chargeA;
1033     nvdwtype         = fr->ntype;
1034     vdwparam         = fr->nbfp;
1035     vdwtype          = mdatoms->typeA;
1036
1037     vftab            = kernel_data->table_vdw->data;
1038     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
1039
1040     /* Setup water-specific parameters */
1041     inr              = nlist->iinr[0];
1042     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1043     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1044     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1045     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1046
1047     jq0              = _mm_set1_ps(charge[inr+0]);
1048     jq1              = _mm_set1_ps(charge[inr+1]);
1049     jq2              = _mm_set1_ps(charge[inr+2]);
1050     vdwjidx0A        = 2*vdwtype[inr+0];
1051     qq00             = _mm_mul_ps(iq0,jq0);
1052     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1053     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1054     qq01             = _mm_mul_ps(iq0,jq1);
1055     qq02             = _mm_mul_ps(iq0,jq2);
1056     qq10             = _mm_mul_ps(iq1,jq0);
1057     qq11             = _mm_mul_ps(iq1,jq1);
1058     qq12             = _mm_mul_ps(iq1,jq2);
1059     qq20             = _mm_mul_ps(iq2,jq0);
1060     qq21             = _mm_mul_ps(iq2,jq1);
1061     qq22             = _mm_mul_ps(iq2,jq2);
1062
1063     /* Avoid stupid compiler warnings */
1064     jnrA = jnrB = jnrC = jnrD = 0;
1065     j_coord_offsetA = 0;
1066     j_coord_offsetB = 0;
1067     j_coord_offsetC = 0;
1068     j_coord_offsetD = 0;
1069
1070     outeriter        = 0;
1071     inneriter        = 0;
1072
1073     for(iidx=0;iidx<4*DIM;iidx++)
1074     {
1075         scratch[iidx] = 0.0;
1076     }
1077
1078     /* Start outer loop over neighborlists */
1079     for(iidx=0; iidx<nri; iidx++)
1080     {
1081         /* Load shift vector for this list */
1082         i_shift_offset   = DIM*shiftidx[iidx];
1083
1084         /* Load limits for loop over neighbors */
1085         j_index_start    = jindex[iidx];
1086         j_index_end      = jindex[iidx+1];
1087
1088         /* Get outer coordinate index */
1089         inr              = iinr[iidx];
1090         i_coord_offset   = DIM*inr;
1091
1092         /* Load i particle coords and add shift vector */
1093         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1094                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1095
1096         fix0             = _mm_setzero_ps();
1097         fiy0             = _mm_setzero_ps();
1098         fiz0             = _mm_setzero_ps();
1099         fix1             = _mm_setzero_ps();
1100         fiy1             = _mm_setzero_ps();
1101         fiz1             = _mm_setzero_ps();
1102         fix2             = _mm_setzero_ps();
1103         fiy2             = _mm_setzero_ps();
1104         fiz2             = _mm_setzero_ps();
1105
1106         /* Start inner kernel loop */
1107         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1108         {
1109
1110             /* Get j neighbor index, and coordinate index */
1111             jnrA             = jjnr[jidx];
1112             jnrB             = jjnr[jidx+1];
1113             jnrC             = jjnr[jidx+2];
1114             jnrD             = jjnr[jidx+3];
1115             j_coord_offsetA  = DIM*jnrA;
1116             j_coord_offsetB  = DIM*jnrB;
1117             j_coord_offsetC  = DIM*jnrC;
1118             j_coord_offsetD  = DIM*jnrD;
1119
1120             /* load j atom coordinates */
1121             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1122                                               x+j_coord_offsetC,x+j_coord_offsetD,
1123                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1124
1125             /* Calculate displacement vector */
1126             dx00             = _mm_sub_ps(ix0,jx0);
1127             dy00             = _mm_sub_ps(iy0,jy0);
1128             dz00             = _mm_sub_ps(iz0,jz0);
1129             dx01             = _mm_sub_ps(ix0,jx1);
1130             dy01             = _mm_sub_ps(iy0,jy1);
1131             dz01             = _mm_sub_ps(iz0,jz1);
1132             dx02             = _mm_sub_ps(ix0,jx2);
1133             dy02             = _mm_sub_ps(iy0,jy2);
1134             dz02             = _mm_sub_ps(iz0,jz2);
1135             dx10             = _mm_sub_ps(ix1,jx0);
1136             dy10             = _mm_sub_ps(iy1,jy0);
1137             dz10             = _mm_sub_ps(iz1,jz0);
1138             dx11             = _mm_sub_ps(ix1,jx1);
1139             dy11             = _mm_sub_ps(iy1,jy1);
1140             dz11             = _mm_sub_ps(iz1,jz1);
1141             dx12             = _mm_sub_ps(ix1,jx2);
1142             dy12             = _mm_sub_ps(iy1,jy2);
1143             dz12             = _mm_sub_ps(iz1,jz2);
1144             dx20             = _mm_sub_ps(ix2,jx0);
1145             dy20             = _mm_sub_ps(iy2,jy0);
1146             dz20             = _mm_sub_ps(iz2,jz0);
1147             dx21             = _mm_sub_ps(ix2,jx1);
1148             dy21             = _mm_sub_ps(iy2,jy1);
1149             dz21             = _mm_sub_ps(iz2,jz1);
1150             dx22             = _mm_sub_ps(ix2,jx2);
1151             dy22             = _mm_sub_ps(iy2,jy2);
1152             dz22             = _mm_sub_ps(iz2,jz2);
1153
1154             /* Calculate squared distance and things based on it */
1155             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1156             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1157             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1158             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1159             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1160             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1161             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1162             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1163             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1164
1165             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1166             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1167             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1168             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1169             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1170             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1171             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1172             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1173             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1174
1175             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1176             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1177             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1178             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1179             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1180             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1181             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1182             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1183             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1184
1185             fjx0             = _mm_setzero_ps();
1186             fjy0             = _mm_setzero_ps();
1187             fjz0             = _mm_setzero_ps();
1188             fjx1             = _mm_setzero_ps();
1189             fjy1             = _mm_setzero_ps();
1190             fjz1             = _mm_setzero_ps();
1191             fjx2             = _mm_setzero_ps();
1192             fjy2             = _mm_setzero_ps();
1193             fjz2             = _mm_setzero_ps();
1194
1195             /**************************
1196              * CALCULATE INTERACTIONS *
1197              **************************/
1198
1199             r00              = _mm_mul_ps(rsq00,rinv00);
1200
1201             /* Calculate table index by multiplying r with table scale and truncate to integer */
1202             rt               = _mm_mul_ps(r00,vftabscale);
1203             vfitab           = _mm_cvttps_epi32(rt);
1204 #ifdef __XOP__
1205             vfeps            = _mm_frcz_ps(rt);
1206 #else
1207             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1208 #endif
1209             twovfeps         = _mm_add_ps(vfeps,vfeps);
1210             vfitab           = _mm_slli_epi32(vfitab,3);
1211
1212             /* COULOMB ELECTROSTATICS */
1213             velec            = _mm_mul_ps(qq00,rinv00);
1214             felec            = _mm_mul_ps(velec,rinvsq00);
1215
1216             /* CUBIC SPLINE TABLE DISPERSION */
1217             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1218             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1219             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1220             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1221             _MM_TRANSPOSE4_PS(Y,F,G,H);
1222             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1223             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1224             fvdw6            = _mm_mul_ps(c6_00,FF);
1225
1226             /* CUBIC SPLINE TABLE REPULSION */
1227             vfitab           = _mm_add_epi32(vfitab,ifour);
1228             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1229             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1230             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1231             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1232             _MM_TRANSPOSE4_PS(Y,F,G,H);
1233             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1234             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1235             fvdw12           = _mm_mul_ps(c12_00,FF);
1236             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1237
1238             fscal            = _mm_add_ps(felec,fvdw);
1239
1240              /* Update vectorial force */
1241             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1242             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1243             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1244
1245             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1246             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1247             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1248
1249             /**************************
1250              * CALCULATE INTERACTIONS *
1251              **************************/
1252
1253             /* COULOMB ELECTROSTATICS */
1254             velec            = _mm_mul_ps(qq01,rinv01);
1255             felec            = _mm_mul_ps(velec,rinvsq01);
1256
1257             fscal            = felec;
1258
1259              /* Update vectorial force */
1260             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1261             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1262             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1263
1264             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1265             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1266             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1267
1268             /**************************
1269              * CALCULATE INTERACTIONS *
1270              **************************/
1271
1272             /* COULOMB ELECTROSTATICS */
1273             velec            = _mm_mul_ps(qq02,rinv02);
1274             felec            = _mm_mul_ps(velec,rinvsq02);
1275
1276             fscal            = felec;
1277
1278              /* Update vectorial force */
1279             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1280             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1281             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1282
1283             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1284             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1285             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1286
1287             /**************************
1288              * CALCULATE INTERACTIONS *
1289              **************************/
1290
1291             /* COULOMB ELECTROSTATICS */
1292             velec            = _mm_mul_ps(qq10,rinv10);
1293             felec            = _mm_mul_ps(velec,rinvsq10);
1294
1295             fscal            = felec;
1296
1297              /* Update vectorial force */
1298             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1299             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1300             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1301
1302             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1303             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1304             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1305
1306             /**************************
1307              * CALCULATE INTERACTIONS *
1308              **************************/
1309
1310             /* COULOMB ELECTROSTATICS */
1311             velec            = _mm_mul_ps(qq11,rinv11);
1312             felec            = _mm_mul_ps(velec,rinvsq11);
1313
1314             fscal            = felec;
1315
1316              /* Update vectorial force */
1317             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1318             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1319             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1320
1321             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1322             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1323             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1324
1325             /**************************
1326              * CALCULATE INTERACTIONS *
1327              **************************/
1328
1329             /* COULOMB ELECTROSTATICS */
1330             velec            = _mm_mul_ps(qq12,rinv12);
1331             felec            = _mm_mul_ps(velec,rinvsq12);
1332
1333             fscal            = felec;
1334
1335              /* Update vectorial force */
1336             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1337             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1338             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1339
1340             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1341             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1342             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1343
1344             /**************************
1345              * CALCULATE INTERACTIONS *
1346              **************************/
1347
1348             /* COULOMB ELECTROSTATICS */
1349             velec            = _mm_mul_ps(qq20,rinv20);
1350             felec            = _mm_mul_ps(velec,rinvsq20);
1351
1352             fscal            = felec;
1353
1354              /* Update vectorial force */
1355             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1356             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1357             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1358
1359             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1360             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1361             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1362
1363             /**************************
1364              * CALCULATE INTERACTIONS *
1365              **************************/
1366
1367             /* COULOMB ELECTROSTATICS */
1368             velec            = _mm_mul_ps(qq21,rinv21);
1369             felec            = _mm_mul_ps(velec,rinvsq21);
1370
1371             fscal            = felec;
1372
1373              /* Update vectorial force */
1374             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1375             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1376             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1377
1378             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1379             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1380             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1381
1382             /**************************
1383              * CALCULATE INTERACTIONS *
1384              **************************/
1385
1386             /* COULOMB ELECTROSTATICS */
1387             velec            = _mm_mul_ps(qq22,rinv22);
1388             felec            = _mm_mul_ps(velec,rinvsq22);
1389
1390             fscal            = felec;
1391
1392              /* Update vectorial force */
1393             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1394             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1395             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1396
1397             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1398             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1399             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1400
1401             fjptrA             = f+j_coord_offsetA;
1402             fjptrB             = f+j_coord_offsetB;
1403             fjptrC             = f+j_coord_offsetC;
1404             fjptrD             = f+j_coord_offsetD;
1405
1406             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1407                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1408
1409             /* Inner loop uses 297 flops */
1410         }
1411
1412         if(jidx<j_index_end)
1413         {
1414
1415             /* Get j neighbor index, and coordinate index */
1416             jnrlistA         = jjnr[jidx];
1417             jnrlistB         = jjnr[jidx+1];
1418             jnrlistC         = jjnr[jidx+2];
1419             jnrlistD         = jjnr[jidx+3];
1420             /* Sign of each element will be negative for non-real atoms.
1421              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1422              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1423              */
1424             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1425             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1426             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1427             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1428             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1429             j_coord_offsetA  = DIM*jnrA;
1430             j_coord_offsetB  = DIM*jnrB;
1431             j_coord_offsetC  = DIM*jnrC;
1432             j_coord_offsetD  = DIM*jnrD;
1433
1434             /* load j atom coordinates */
1435             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1436                                               x+j_coord_offsetC,x+j_coord_offsetD,
1437                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1438
1439             /* Calculate displacement vector */
1440             dx00             = _mm_sub_ps(ix0,jx0);
1441             dy00             = _mm_sub_ps(iy0,jy0);
1442             dz00             = _mm_sub_ps(iz0,jz0);
1443             dx01             = _mm_sub_ps(ix0,jx1);
1444             dy01             = _mm_sub_ps(iy0,jy1);
1445             dz01             = _mm_sub_ps(iz0,jz1);
1446             dx02             = _mm_sub_ps(ix0,jx2);
1447             dy02             = _mm_sub_ps(iy0,jy2);
1448             dz02             = _mm_sub_ps(iz0,jz2);
1449             dx10             = _mm_sub_ps(ix1,jx0);
1450             dy10             = _mm_sub_ps(iy1,jy0);
1451             dz10             = _mm_sub_ps(iz1,jz0);
1452             dx11             = _mm_sub_ps(ix1,jx1);
1453             dy11             = _mm_sub_ps(iy1,jy1);
1454             dz11             = _mm_sub_ps(iz1,jz1);
1455             dx12             = _mm_sub_ps(ix1,jx2);
1456             dy12             = _mm_sub_ps(iy1,jy2);
1457             dz12             = _mm_sub_ps(iz1,jz2);
1458             dx20             = _mm_sub_ps(ix2,jx0);
1459             dy20             = _mm_sub_ps(iy2,jy0);
1460             dz20             = _mm_sub_ps(iz2,jz0);
1461             dx21             = _mm_sub_ps(ix2,jx1);
1462             dy21             = _mm_sub_ps(iy2,jy1);
1463             dz21             = _mm_sub_ps(iz2,jz1);
1464             dx22             = _mm_sub_ps(ix2,jx2);
1465             dy22             = _mm_sub_ps(iy2,jy2);
1466             dz22             = _mm_sub_ps(iz2,jz2);
1467
1468             /* Calculate squared distance and things based on it */
1469             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1470             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1471             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1472             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1473             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1474             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1475             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1476             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1477             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1478
1479             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1480             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1481             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1482             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1483             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1484             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1485             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1486             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1487             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1488
1489             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1490             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1491             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1492             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1493             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1494             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1495             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1496             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1497             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1498
1499             fjx0             = _mm_setzero_ps();
1500             fjy0             = _mm_setzero_ps();
1501             fjz0             = _mm_setzero_ps();
1502             fjx1             = _mm_setzero_ps();
1503             fjy1             = _mm_setzero_ps();
1504             fjz1             = _mm_setzero_ps();
1505             fjx2             = _mm_setzero_ps();
1506             fjy2             = _mm_setzero_ps();
1507             fjz2             = _mm_setzero_ps();
1508
1509             /**************************
1510              * CALCULATE INTERACTIONS *
1511              **************************/
1512
1513             r00              = _mm_mul_ps(rsq00,rinv00);
1514             r00              = _mm_andnot_ps(dummy_mask,r00);
1515
1516             /* Calculate table index by multiplying r with table scale and truncate to integer */
1517             rt               = _mm_mul_ps(r00,vftabscale);
1518             vfitab           = _mm_cvttps_epi32(rt);
1519 #ifdef __XOP__
1520             vfeps            = _mm_frcz_ps(rt);
1521 #else
1522             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1523 #endif
1524             twovfeps         = _mm_add_ps(vfeps,vfeps);
1525             vfitab           = _mm_slli_epi32(vfitab,3);
1526
1527             /* COULOMB ELECTROSTATICS */
1528             velec            = _mm_mul_ps(qq00,rinv00);
1529             felec            = _mm_mul_ps(velec,rinvsq00);
1530
1531             /* CUBIC SPLINE TABLE DISPERSION */
1532             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1533             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1534             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1535             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1536             _MM_TRANSPOSE4_PS(Y,F,G,H);
1537             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1538             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1539             fvdw6            = _mm_mul_ps(c6_00,FF);
1540
1541             /* CUBIC SPLINE TABLE REPULSION */
1542             vfitab           = _mm_add_epi32(vfitab,ifour);
1543             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1544             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1545             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1546             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1547             _MM_TRANSPOSE4_PS(Y,F,G,H);
1548             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1549             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1550             fvdw12           = _mm_mul_ps(c12_00,FF);
1551             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1552
1553             fscal            = _mm_add_ps(felec,fvdw);
1554
1555             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1556
1557              /* Update vectorial force */
1558             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1559             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1560             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1561
1562             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1563             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1564             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1565
1566             /**************************
1567              * CALCULATE INTERACTIONS *
1568              **************************/
1569
1570             /* COULOMB ELECTROSTATICS */
1571             velec            = _mm_mul_ps(qq01,rinv01);
1572             felec            = _mm_mul_ps(velec,rinvsq01);
1573
1574             fscal            = felec;
1575
1576             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1577
1578              /* Update vectorial force */
1579             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1580             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1581             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1582
1583             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1584             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1585             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1586
1587             /**************************
1588              * CALCULATE INTERACTIONS *
1589              **************************/
1590
1591             /* COULOMB ELECTROSTATICS */
1592             velec            = _mm_mul_ps(qq02,rinv02);
1593             felec            = _mm_mul_ps(velec,rinvsq02);
1594
1595             fscal            = felec;
1596
1597             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1598
1599              /* Update vectorial force */
1600             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1601             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1602             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1603
1604             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1605             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1606             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1607
1608             /**************************
1609              * CALCULATE INTERACTIONS *
1610              **************************/
1611
1612             /* COULOMB ELECTROSTATICS */
1613             velec            = _mm_mul_ps(qq10,rinv10);
1614             felec            = _mm_mul_ps(velec,rinvsq10);
1615
1616             fscal            = felec;
1617
1618             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1619
1620              /* Update vectorial force */
1621             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1622             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1623             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1624
1625             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1626             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1627             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1628
1629             /**************************
1630              * CALCULATE INTERACTIONS *
1631              **************************/
1632
1633             /* COULOMB ELECTROSTATICS */
1634             velec            = _mm_mul_ps(qq11,rinv11);
1635             felec            = _mm_mul_ps(velec,rinvsq11);
1636
1637             fscal            = felec;
1638
1639             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1640
1641              /* Update vectorial force */
1642             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1643             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1644             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1645
1646             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1647             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1648             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1649
1650             /**************************
1651              * CALCULATE INTERACTIONS *
1652              **************************/
1653
1654             /* COULOMB ELECTROSTATICS */
1655             velec            = _mm_mul_ps(qq12,rinv12);
1656             felec            = _mm_mul_ps(velec,rinvsq12);
1657
1658             fscal            = felec;
1659
1660             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1661
1662              /* Update vectorial force */
1663             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1664             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1665             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1666
1667             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1668             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1669             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1670
1671             /**************************
1672              * CALCULATE INTERACTIONS *
1673              **************************/
1674
1675             /* COULOMB ELECTROSTATICS */
1676             velec            = _mm_mul_ps(qq20,rinv20);
1677             felec            = _mm_mul_ps(velec,rinvsq20);
1678
1679             fscal            = felec;
1680
1681             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1682
1683              /* Update vectorial force */
1684             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1685             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1686             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1687
1688             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1689             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1690             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1691
1692             /**************************
1693              * CALCULATE INTERACTIONS *
1694              **************************/
1695
1696             /* COULOMB ELECTROSTATICS */
1697             velec            = _mm_mul_ps(qq21,rinv21);
1698             felec            = _mm_mul_ps(velec,rinvsq21);
1699
1700             fscal            = felec;
1701
1702             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1703
1704              /* Update vectorial force */
1705             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1706             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1707             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1708
1709             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1710             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1711             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1712
1713             /**************************
1714              * CALCULATE INTERACTIONS *
1715              **************************/
1716
1717             /* COULOMB ELECTROSTATICS */
1718             velec            = _mm_mul_ps(qq22,rinv22);
1719             felec            = _mm_mul_ps(velec,rinvsq22);
1720
1721             fscal            = felec;
1722
1723             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1724
1725              /* Update vectorial force */
1726             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1727             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1728             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1729
1730             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1731             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1732             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1733
1734             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1735             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1736             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1737             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1738
1739             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1740                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1741
1742             /* Inner loop uses 298 flops */
1743         }
1744
1745         /* End of innermost loop */
1746
1747         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1748                                               f+i_coord_offset,fshift+i_shift_offset);
1749
1750         /* Increment number of inner iterations */
1751         inneriter                  += j_index_end - j_index_start;
1752
1753         /* Outer loop uses 18 flops */
1754     }
1755
1756     /* Increment number of outer iterations */
1757     outeriter        += nri;
1758
1759     /* Update outer/inner flops */
1760
1761     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*298);
1762 }