src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwLJ_GeomW4W4_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4W4_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwLJ_GeomW4W4_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  80     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  81     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  82     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  83     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  84     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  85     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  86     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  87     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  88     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  89     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  90     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  91     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  92     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  93     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  94     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  95     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 102     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 103     __m128           dummy_mask,cutoff_mask;
 104     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 105     __m128           one     = _mm_set1_ps(1.0);
 106     __m128           two     = _mm_set1_ps(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_ps(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     krf              = _mm_set1_ps(fr->ic->k_rf);
 121     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 122     crf              = _mm_set1_ps(fr->ic->c_rf);
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 130     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 131     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     jq1              = _mm_set1_ps(charge[inr+1]);
 135     jq2              = _mm_set1_ps(charge[inr+2]);
 136     jq3              = _mm_set1_ps(charge[inr+3]);
 137     vdwjidx0A        = 2*vdwtype[inr+0];
 138     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 139     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 140     qq11             = _mm_mul_ps(iq1,jq1);
 141     qq12             = _mm_mul_ps(iq1,jq2);
 142     qq13             = _mm_mul_ps(iq1,jq3);
 143     qq21             = _mm_mul_ps(iq2,jq1);
 144     qq22             = _mm_mul_ps(iq2,jq2);
 145     qq23             = _mm_mul_ps(iq2,jq3);
 146     qq31             = _mm_mul_ps(iq3,jq1);
 147     qq32             = _mm_mul_ps(iq3,jq2);
 148     qq33             = _mm_mul_ps(iq3,jq3);
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = jnrC = jnrD = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154     j_coord_offsetC = 0;
 155     j_coord_offsetD = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix0             = _mm_setzero_ps();
 184         fiy0             = _mm_setzero_ps();
 185         fiz0             = _mm_setzero_ps();
 186         fix1             = _mm_setzero_ps();
 187         fiy1             = _mm_setzero_ps();
 188         fiz1             = _mm_setzero_ps();
 189         fix2             = _mm_setzero_ps();
 190         fiy2             = _mm_setzero_ps();
 191         fiz2             = _mm_setzero_ps();
 192         fix3             = _mm_setzero_ps();
 193         fiy3             = _mm_setzero_ps();
 194         fiz3             = _mm_setzero_ps();
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm_setzero_ps();
 198         vvdwsum          = _mm_setzero_ps();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             j_coord_offsetA  = DIM*jnrA;
 210             j_coord_offsetB  = DIM*jnrB;
 211             j_coord_offsetC  = DIM*jnrC;
 212             j_coord_offsetD  = DIM*jnrD;
 213
 214             /* load j atom coordinates */
 215             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                               x+j_coord_offsetC,x+j_coord_offsetD,
 217                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 218                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm_sub_ps(ix0,jx0);
 222             dy00             = _mm_sub_ps(iy0,jy0);
 223             dz00             = _mm_sub_ps(iz0,jz0);
 224             dx11             = _mm_sub_ps(ix1,jx1);
 225             dy11             = _mm_sub_ps(iy1,jy1);
 226             dz11             = _mm_sub_ps(iz1,jz1);
 227             dx12             = _mm_sub_ps(ix1,jx2);
 228             dy12             = _mm_sub_ps(iy1,jy2);
 229             dz12             = _mm_sub_ps(iz1,jz2);
 230             dx13             = _mm_sub_ps(ix1,jx3);
 231             dy13             = _mm_sub_ps(iy1,jy3);
 232             dz13             = _mm_sub_ps(iz1,jz3);
 233             dx21             = _mm_sub_ps(ix2,jx1);
 234             dy21             = _mm_sub_ps(iy2,jy1);
 235             dz21             = _mm_sub_ps(iz2,jz1);
 236             dx22             = _mm_sub_ps(ix2,jx2);
 237             dy22             = _mm_sub_ps(iy2,jy2);
 238             dz22             = _mm_sub_ps(iz2,jz2);
 239             dx23             = _mm_sub_ps(ix2,jx3);
 240             dy23             = _mm_sub_ps(iy2,jy3);
 241             dz23             = _mm_sub_ps(iz2,jz3);
 242             dx31             = _mm_sub_ps(ix3,jx1);
 243             dy31             = _mm_sub_ps(iy3,jy1);
 244             dz31             = _mm_sub_ps(iz3,jz1);
 245             dx32             = _mm_sub_ps(ix3,jx2);
 246             dy32             = _mm_sub_ps(iy3,jy2);
 247             dz32             = _mm_sub_ps(iz3,jz2);
 248             dx33             = _mm_sub_ps(ix3,jx3);
 249             dy33             = _mm_sub_ps(iy3,jy3);
 250             dz33             = _mm_sub_ps(iz3,jz3);
 251
 252             /* Calculate squared distance and things based on it */
 253             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 254             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 255             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 256             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 257             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 258             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 259             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 260             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 261             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 262             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 263
 264             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 265             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 266             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 267             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 268             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 269             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 270             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 271             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 272             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 273
 274             rinvsq00         = gmx_mm_inv_ps(rsq00);
 275             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 276             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 277             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 278             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 279             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 280             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 281             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 282             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 283             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 284
 285             fjx0             = _mm_setzero_ps();
 286             fjy0             = _mm_setzero_ps();
 287             fjz0             = _mm_setzero_ps();
 288             fjx1             = _mm_setzero_ps();
 289             fjy1             = _mm_setzero_ps();
 290             fjz1             = _mm_setzero_ps();
 291             fjx2             = _mm_setzero_ps();
 292             fjy2             = _mm_setzero_ps();
 293             fjz2             = _mm_setzero_ps();
 294             fjx3             = _mm_setzero_ps();
 295             fjy3             = _mm_setzero_ps();
 296             fjz3             = _mm_setzero_ps();
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             /* LENNARD-JONES DISPERSION/REPULSION */
 303
 304             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 305             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 306             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 307             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 308             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 312
 313             fscal            = fvdw;
 314
 315              /* Update vectorial force */
 316             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 317             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 318             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 319
 320             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 321             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 322             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             /* REACTION-FIELD ELECTROSTATICS */
 329             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 330             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm_add_ps(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337              /* Update vectorial force */
 338             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 339             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 340             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 341
 342             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 343             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 344             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             /* REACTION-FIELD ELECTROSTATICS */
 351             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 352             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velecsum         = _mm_add_ps(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359              /* Update vectorial force */
 360             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 361             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 362             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 363
 364             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 365             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 366             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             /* REACTION-FIELD ELECTROSTATICS */
 373             velec            = _mm_mul_ps(qq13,_mm_sub_ps(_mm_macc_ps(krf,rsq13,rinv13),crf));
 374             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm_add_ps(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381              /* Update vectorial force */
 382             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 383             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 384             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 385
 386             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 387             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 388             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             /* REACTION-FIELD ELECTROSTATICS */
 395             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 396             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 397
 398             /* Update potential sum for this i atom from the interaction with this j atom. */
 399             velecsum         = _mm_add_ps(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403              /* Update vectorial force */
 404             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 405             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 406             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 407
 408             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 409             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 410             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             /* REACTION-FIELD ELECTROSTATICS */
 417             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 418             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velecsum         = _mm_add_ps(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425              /* Update vectorial force */
 426             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 427             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 428             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 429
 430             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 431             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 432             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 433
 434             /**************************
 435              * CALCULATE INTERACTIONS *
 436              **************************/
 437
 438             /* REACTION-FIELD ELECTROSTATICS */
 439             velec            = _mm_mul_ps(qq23,_mm_sub_ps(_mm_macc_ps(krf,rsq23,rinv23),crf));
 440             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
 441
 442             /* Update potential sum for this i atom from the interaction with this j atom. */
 443             velecsum         = _mm_add_ps(velecsum,velec);
 444
 445             fscal            = felec;
 446
 447              /* Update vectorial force */
 448             fix2             = _mm_macc_ps(dx23,fscal,fix2);
 449             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 450             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 451
 452             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 453             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 454             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             /* REACTION-FIELD ELECTROSTATICS */
 461             velec            = _mm_mul_ps(qq31,_mm_sub_ps(_mm_macc_ps(krf,rsq31,rinv31),crf));
 462             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
 463
 464             /* Update potential sum for this i atom from the interaction with this j atom. */
 465             velecsum         = _mm_add_ps(velecsum,velec);
 466
 467             fscal            = felec;
 468
 469              /* Update vectorial force */
 470             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 471             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 472             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 473
 474             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 475             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 476             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             /* REACTION-FIELD ELECTROSTATICS */
 483             velec            = _mm_mul_ps(qq32,_mm_sub_ps(_mm_macc_ps(krf,rsq32,rinv32),crf));
 484             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             velecsum         = _mm_add_ps(velecsum,velec);
 488
 489             fscal            = felec;
 490
 491              /* Update vectorial force */
 492             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 493             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 494             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 495
 496             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 497             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 498             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             /* REACTION-FIELD ELECTROSTATICS */
 505             velec            = _mm_mul_ps(qq33,_mm_sub_ps(_mm_macc_ps(krf,rsq33,rinv33),crf));
 506             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velecsum         = _mm_add_ps(velecsum,velec);
 510
 511             fscal            = felec;
 512
 513              /* Update vectorial force */
 514             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 515             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 516             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 517
 518             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 519             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 520             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 521
 522             fjptrA             = f+j_coord_offsetA;
 523             fjptrB             = f+j_coord_offsetB;
 524             fjptrC             = f+j_coord_offsetC;
 525             fjptrD             = f+j_coord_offsetD;
 526
 527             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 528                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 529                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 530
 531             /* Inner loop uses 353 flops */
 532         }
 533
 534         if(jidx<j_index_end)
 535         {
 536
 537             /* Get j neighbor index, and coordinate index */
 538             jnrlistA         = jjnr[jidx];
 539             jnrlistB         = jjnr[jidx+1];
 540             jnrlistC         = jjnr[jidx+2];
 541             jnrlistD         = jjnr[jidx+3];
 542             /* Sign of each element will be negative for non-real atoms.
 543              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 544              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 545              */
 546             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 547             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 548             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 549             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 550             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 551             j_coord_offsetA  = DIM*jnrA;
 552             j_coord_offsetB  = DIM*jnrB;
 553             j_coord_offsetC  = DIM*jnrC;
 554             j_coord_offsetD  = DIM*jnrD;
 555
 556             /* load j atom coordinates */
 557             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 558                                               x+j_coord_offsetC,x+j_coord_offsetD,
 559                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 560                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 561
 562             /* Calculate displacement vector */
 563             dx00             = _mm_sub_ps(ix0,jx0);
 564             dy00             = _mm_sub_ps(iy0,jy0);
 565             dz00             = _mm_sub_ps(iz0,jz0);
 566             dx11             = _mm_sub_ps(ix1,jx1);
 567             dy11             = _mm_sub_ps(iy1,jy1);
 568             dz11             = _mm_sub_ps(iz1,jz1);
 569             dx12             = _mm_sub_ps(ix1,jx2);
 570             dy12             = _mm_sub_ps(iy1,jy2);
 571             dz12             = _mm_sub_ps(iz1,jz2);
 572             dx13             = _mm_sub_ps(ix1,jx3);
 573             dy13             = _mm_sub_ps(iy1,jy3);
 574             dz13             = _mm_sub_ps(iz1,jz3);
 575             dx21             = _mm_sub_ps(ix2,jx1);
 576             dy21             = _mm_sub_ps(iy2,jy1);
 577             dz21             = _mm_sub_ps(iz2,jz1);
 578             dx22             = _mm_sub_ps(ix2,jx2);
 579             dy22             = _mm_sub_ps(iy2,jy2);
 580             dz22             = _mm_sub_ps(iz2,jz2);
 581             dx23             = _mm_sub_ps(ix2,jx3);
 582             dy23             = _mm_sub_ps(iy2,jy3);
 583             dz23             = _mm_sub_ps(iz2,jz3);
 584             dx31             = _mm_sub_ps(ix3,jx1);
 585             dy31             = _mm_sub_ps(iy3,jy1);
 586             dz31             = _mm_sub_ps(iz3,jz1);
 587             dx32             = _mm_sub_ps(ix3,jx2);
 588             dy32             = _mm_sub_ps(iy3,jy2);
 589             dz32             = _mm_sub_ps(iz3,jz2);
 590             dx33             = _mm_sub_ps(ix3,jx3);
 591             dy33             = _mm_sub_ps(iy3,jy3);
 592             dz33             = _mm_sub_ps(iz3,jz3);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 596             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 597             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 598             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 599             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 600             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 601             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 602             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 603             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 604             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 605
 606             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 607             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 608             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 609             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 610             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 611             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 612             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 613             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 614             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 615
 616             rinvsq00         = gmx_mm_inv_ps(rsq00);
 617             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 618             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 619             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 620             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 621             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 622             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 623             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 624             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 625             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 626
 627             fjx0             = _mm_setzero_ps();
 628             fjy0             = _mm_setzero_ps();
 629             fjz0             = _mm_setzero_ps();
 630             fjx1             = _mm_setzero_ps();
 631             fjy1             = _mm_setzero_ps();
 632             fjz1             = _mm_setzero_ps();
 633             fjx2             = _mm_setzero_ps();
 634             fjy2             = _mm_setzero_ps();
 635             fjz2             = _mm_setzero_ps();
 636             fjx3             = _mm_setzero_ps();
 637             fjy3             = _mm_setzero_ps();
 638             fjz3             = _mm_setzero_ps();
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             /* LENNARD-JONES DISPERSION/REPULSION */
 645
 646             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 647             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 648             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 649             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 650             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 651
 652             /* Update potential sum for this i atom from the interaction with this j atom. */
 653             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 654             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 655
 656             fscal            = fvdw;
 657
 658             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 659
 660              /* Update vectorial force */
 661             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 662             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 663             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 664
 665             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 666             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 667             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 668
 669             /**************************
 670              * CALCULATE INTERACTIONS *
 671              **************************/
 672
 673             /* REACTION-FIELD ELECTROSTATICS */
 674             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 675             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 676
 677             /* Update potential sum for this i atom from the interaction with this j atom. */
 678             velec            = _mm_andnot_ps(dummy_mask,velec);
 679             velecsum         = _mm_add_ps(velecsum,velec);
 680
 681             fscal            = felec;
 682
 683             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 684
 685              /* Update vectorial force */
 686             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 687             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 688             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 689
 690             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 691             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 692             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 693
 694             /**************************
 695              * CALCULATE INTERACTIONS *
 696              **************************/
 697
 698             /* REACTION-FIELD ELECTROSTATICS */
 699             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 700             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 701
 702             /* Update potential sum for this i atom from the interaction with this j atom. */
 703             velec            = _mm_andnot_ps(dummy_mask,velec);
 704             velecsum         = _mm_add_ps(velecsum,velec);
 705
 706             fscal            = felec;
 707
 708             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 709
 710              /* Update vectorial force */
 711             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 712             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 713             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 714
 715             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 716             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 717             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 718
 719             /**************************
 720              * CALCULATE INTERACTIONS *
 721              **************************/
 722
 723             /* REACTION-FIELD ELECTROSTATICS */
 724             velec            = _mm_mul_ps(qq13,_mm_sub_ps(_mm_macc_ps(krf,rsq13,rinv13),crf));
 725             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
 726
 727             /* Update potential sum for this i atom from the interaction with this j atom. */
 728             velec            = _mm_andnot_ps(dummy_mask,velec);
 729             velecsum         = _mm_add_ps(velecsum,velec);
 730
 731             fscal            = felec;
 732
 733             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 734
 735              /* Update vectorial force */
 736             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 737             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 738             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 739
 740             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 741             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 742             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 743
 744             /**************************
 745              * CALCULATE INTERACTIONS *
 746              **************************/
 747
 748             /* REACTION-FIELD ELECTROSTATICS */
 749             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 750             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 751
 752             /* Update potential sum for this i atom from the interaction with this j atom. */
 753             velec            = _mm_andnot_ps(dummy_mask,velec);
 754             velecsum         = _mm_add_ps(velecsum,velec);
 755
 756             fscal            = felec;
 757
 758             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 759
 760              /* Update vectorial force */
 761             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 762             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 763             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 764
 765             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 766             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 767             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 768
 769             /**************************
 770              * CALCULATE INTERACTIONS *
 771              **************************/
 772
 773             /* REACTION-FIELD ELECTROSTATICS */
 774             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 775             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 776
 777             /* Update potential sum for this i atom from the interaction with this j atom. */
 778             velec            = _mm_andnot_ps(dummy_mask,velec);
 779             velecsum         = _mm_add_ps(velecsum,velec);
 780
 781             fscal            = felec;
 782
 783             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 784
 785              /* Update vectorial force */
 786             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 787             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 788             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 789
 790             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 791             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 792             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 793
 794             /**************************
 795              * CALCULATE INTERACTIONS *
 796              **************************/
 797
 798             /* REACTION-FIELD ELECTROSTATICS */
 799             velec            = _mm_mul_ps(qq23,_mm_sub_ps(_mm_macc_ps(krf,rsq23,rinv23),crf));
 800             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
 801
 802             /* Update potential sum for this i atom from the interaction with this j atom. */
 803             velec            = _mm_andnot_ps(dummy_mask,velec);
 804             velecsum         = _mm_add_ps(velecsum,velec);
 805
 806             fscal            = felec;
 807
 808             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 809
 810              /* Update vectorial force */
 811             fix2             = _mm_macc_ps(dx23,fscal,fix2);
 812             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 813             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 814
 815             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 816             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 817             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             /* REACTION-FIELD ELECTROSTATICS */
 824             velec            = _mm_mul_ps(qq31,_mm_sub_ps(_mm_macc_ps(krf,rsq31,rinv31),crf));
 825             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
 826
 827             /* Update potential sum for this i atom from the interaction with this j atom. */
 828             velec            = _mm_andnot_ps(dummy_mask,velec);
 829             velecsum         = _mm_add_ps(velecsum,velec);
 830
 831             fscal            = felec;
 832
 833             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 834
 835              /* Update vectorial force */
 836             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 837             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 838             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 839
 840             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 841             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 842             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 843
 844             /**************************
 845              * CALCULATE INTERACTIONS *
 846              **************************/
 847
 848             /* REACTION-FIELD ELECTROSTATICS */
 849             velec            = _mm_mul_ps(qq32,_mm_sub_ps(_mm_macc_ps(krf,rsq32,rinv32),crf));
 850             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
 851
 852             /* Update potential sum for this i atom from the interaction with this j atom. */
 853             velec            = _mm_andnot_ps(dummy_mask,velec);
 854             velecsum         = _mm_add_ps(velecsum,velec);
 855
 856             fscal            = felec;
 857
 858             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 859
 860              /* Update vectorial force */
 861             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 862             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 863             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 864
 865             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 866             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 867             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 868
 869             /**************************
 870              * CALCULATE INTERACTIONS *
 871              **************************/
 872
 873             /* REACTION-FIELD ELECTROSTATICS */
 874             velec            = _mm_mul_ps(qq33,_mm_sub_ps(_mm_macc_ps(krf,rsq33,rinv33),crf));
 875             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
 876
 877             /* Update potential sum for this i atom from the interaction with this j atom. */
 878             velec            = _mm_andnot_ps(dummy_mask,velec);
 879             velecsum         = _mm_add_ps(velecsum,velec);
 880
 881             fscal            = felec;
 882
 883             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 884
 885              /* Update vectorial force */
 886             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 887             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 888             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 889
 890             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 891             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 892             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 893
 894             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 895             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 896             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 897             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 898
 899             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 900                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 901                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 902
 903             /* Inner loop uses 353 flops */
 904         }
 905
 906         /* End of innermost loop */
 907
 908         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 909                                               f+i_coord_offset,fshift+i_shift_offset);
 910
 911         ggid                        = gid[iidx];
 912         /* Update potential energies */
 913         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 914         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 915
 916         /* Increment number of inner iterations */
 917         inneriter                  += j_index_end - j_index_start;
 918
 919         /* Outer loop uses 26 flops */
 920     }
 921
 922     /* Increment number of outer iterations */
 923     outeriter        += nri;
 924
 925     /* Update outer/inner flops */
 926
 927     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*353);
 928 }
 929 /*
 930  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4W4_F_avx_128_fma_single
 931  * Electrostatics interaction: ReactionField
 932  * VdW interaction:            LennardJones
 933  * Geometry:                   Water4-Water4
 934  * Calculate force/pot:        Force
 935  */
 936 void
 937 nb_kernel_ElecRF_VdwLJ_GeomW4W4_F_avx_128_fma_single
 938                     (t_nblist * gmx_restrict                nlist,
 939                      rvec * gmx_restrict                    xx,
 940                      rvec * gmx_restrict                    ff,
 941                      t_forcerec * gmx_restrict              fr,
 942                      t_mdatoms * gmx_restrict               mdatoms,
 943                      nb_kernel_data_t * gmx_restrict        kernel_data,
 944                      t_nrnb * gmx_restrict                  nrnb)
 945 {
 946     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 947      * just 0 for non-waters.
 948      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 949      * jnr indices corresponding to data put in the four positions in the SIMD register.
 950      */
 951     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 952     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 953     int              jnrA,jnrB,jnrC,jnrD;
 954     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 955     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 956     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 957     real             rcutoff_scalar;
 958     real             *shiftvec,*fshift,*x,*f;
 959     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 960     real             scratch[4*DIM];
 961     __m128           fscal,rcutoff,rcutoff2,jidxall;
 962     int              vdwioffset0;
 963     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 964     int              vdwioffset1;
 965     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 966     int              vdwioffset2;
 967     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 968     int              vdwioffset3;
 969     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 970     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 971     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 972     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 973     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 974     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 975     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 976     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
 977     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 978     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 979     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 980     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 981     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 982     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 983     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 984     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 985     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 986     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 987     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 988     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 989     real             *charge;
 990     int              nvdwtype;
 991     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 992     int              *vdwtype;
 993     real             *vdwparam;
 994     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 995     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 996     __m128           dummy_mask,cutoff_mask;
 997     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 998     __m128           one     = _mm_set1_ps(1.0);
 999     __m128           two     = _mm_set1_ps(2.0);
1000     x                = xx[0];
1001     f                = ff[0];
1002
1003     nri              = nlist->nri;
1004     iinr             = nlist->iinr;
1005     jindex           = nlist->jindex;
1006     jjnr             = nlist->jjnr;
1007     shiftidx         = nlist->shift;
1008     gid              = nlist->gid;
1009     shiftvec         = fr->shift_vec[0];
1010     fshift           = fr->fshift[0];
1011     facel            = _mm_set1_ps(fr->epsfac);
1012     charge           = mdatoms->chargeA;
1013     krf              = _mm_set1_ps(fr->ic->k_rf);
1014     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
1015     crf              = _mm_set1_ps(fr->ic->c_rf);
1016     nvdwtype         = fr->ntype;
1017     vdwparam         = fr->nbfp;
1018     vdwtype          = mdatoms->typeA;
1019
1020     /* Setup water-specific parameters */
1021     inr              = nlist->iinr[0];
1022     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1023     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1024     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1025     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1026
1027     jq1              = _mm_set1_ps(charge[inr+1]);
1028     jq2              = _mm_set1_ps(charge[inr+2]);
1029     jq3              = _mm_set1_ps(charge[inr+3]);
1030     vdwjidx0A        = 2*vdwtype[inr+0];
1031     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1032     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1033     qq11             = _mm_mul_ps(iq1,jq1);
1034     qq12             = _mm_mul_ps(iq1,jq2);
1035     qq13             = _mm_mul_ps(iq1,jq3);
1036     qq21             = _mm_mul_ps(iq2,jq1);
1037     qq22             = _mm_mul_ps(iq2,jq2);
1038     qq23             = _mm_mul_ps(iq2,jq3);
1039     qq31             = _mm_mul_ps(iq3,jq1);
1040     qq32             = _mm_mul_ps(iq3,jq2);
1041     qq33             = _mm_mul_ps(iq3,jq3);
1042
1043     /* Avoid stupid compiler warnings */
1044     jnrA = jnrB = jnrC = jnrD = 0;
1045     j_coord_offsetA = 0;
1046     j_coord_offsetB = 0;
1047     j_coord_offsetC = 0;
1048     j_coord_offsetD = 0;
1049
1050     outeriter        = 0;
1051     inneriter        = 0;
1052
1053     for(iidx=0;iidx<4*DIM;iidx++)
1054     {
1055         scratch[iidx] = 0.0;
1056     }
1057
1058     /* Start outer loop over neighborlists */
1059     for(iidx=0; iidx<nri; iidx++)
1060     {
1061         /* Load shift vector for this list */
1062         i_shift_offset   = DIM*shiftidx[iidx];
1063
1064         /* Load limits for loop over neighbors */
1065         j_index_start    = jindex[iidx];
1066         j_index_end      = jindex[iidx+1];
1067
1068         /* Get outer coordinate index */
1069         inr              = iinr[iidx];
1070         i_coord_offset   = DIM*inr;
1071
1072         /* Load i particle coords and add shift vector */
1073         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1074                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1075
1076         fix0             = _mm_setzero_ps();
1077         fiy0             = _mm_setzero_ps();
1078         fiz0             = _mm_setzero_ps();
1079         fix1             = _mm_setzero_ps();
1080         fiy1             = _mm_setzero_ps();
1081         fiz1             = _mm_setzero_ps();
1082         fix2             = _mm_setzero_ps();
1083         fiy2             = _mm_setzero_ps();
1084         fiz2             = _mm_setzero_ps();
1085         fix3             = _mm_setzero_ps();
1086         fiy3             = _mm_setzero_ps();
1087         fiz3             = _mm_setzero_ps();
1088
1089         /* Start inner kernel loop */
1090         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1091         {
1092
1093             /* Get j neighbor index, and coordinate index */
1094             jnrA             = jjnr[jidx];
1095             jnrB             = jjnr[jidx+1];
1096             jnrC             = jjnr[jidx+2];
1097             jnrD             = jjnr[jidx+3];
1098             j_coord_offsetA  = DIM*jnrA;
1099             j_coord_offsetB  = DIM*jnrB;
1100             j_coord_offsetC  = DIM*jnrC;
1101             j_coord_offsetD  = DIM*jnrD;
1102
1103             /* load j atom coordinates */
1104             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1105                                               x+j_coord_offsetC,x+j_coord_offsetD,
1106                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1107                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1108
1109             /* Calculate displacement vector */
1110             dx00             = _mm_sub_ps(ix0,jx0);
1111             dy00             = _mm_sub_ps(iy0,jy0);
1112             dz00             = _mm_sub_ps(iz0,jz0);
1113             dx11             = _mm_sub_ps(ix1,jx1);
1114             dy11             = _mm_sub_ps(iy1,jy1);
1115             dz11             = _mm_sub_ps(iz1,jz1);
1116             dx12             = _mm_sub_ps(ix1,jx2);
1117             dy12             = _mm_sub_ps(iy1,jy2);
1118             dz12             = _mm_sub_ps(iz1,jz2);
1119             dx13             = _mm_sub_ps(ix1,jx3);
1120             dy13             = _mm_sub_ps(iy1,jy3);
1121             dz13             = _mm_sub_ps(iz1,jz3);
1122             dx21             = _mm_sub_ps(ix2,jx1);
1123             dy21             = _mm_sub_ps(iy2,jy1);
1124             dz21             = _mm_sub_ps(iz2,jz1);
1125             dx22             = _mm_sub_ps(ix2,jx2);
1126             dy22             = _mm_sub_ps(iy2,jy2);
1127             dz22             = _mm_sub_ps(iz2,jz2);
1128             dx23             = _mm_sub_ps(ix2,jx3);
1129             dy23             = _mm_sub_ps(iy2,jy3);
1130             dz23             = _mm_sub_ps(iz2,jz3);
1131             dx31             = _mm_sub_ps(ix3,jx1);
1132             dy31             = _mm_sub_ps(iy3,jy1);
1133             dz31             = _mm_sub_ps(iz3,jz1);
1134             dx32             = _mm_sub_ps(ix3,jx2);
1135             dy32             = _mm_sub_ps(iy3,jy2);
1136             dz32             = _mm_sub_ps(iz3,jz2);
1137             dx33             = _mm_sub_ps(ix3,jx3);
1138             dy33             = _mm_sub_ps(iy3,jy3);
1139             dz33             = _mm_sub_ps(iz3,jz3);
1140
1141             /* Calculate squared distance and things based on it */
1142             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1143             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1144             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1145             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1146             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1147             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1148             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1149             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1150             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1151             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1152
1153             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1154             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1155             rinv13           = gmx_mm_invsqrt_ps(rsq13);
1156             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1157             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1158             rinv23           = gmx_mm_invsqrt_ps(rsq23);
1159             rinv31           = gmx_mm_invsqrt_ps(rsq31);
1160             rinv32           = gmx_mm_invsqrt_ps(rsq32);
1161             rinv33           = gmx_mm_invsqrt_ps(rsq33);
1162
1163             rinvsq00         = gmx_mm_inv_ps(rsq00);
1164             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1165             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1166             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1167             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1168             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1169             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1170             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1171             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1172             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1173
1174             fjx0             = _mm_setzero_ps();
1175             fjy0             = _mm_setzero_ps();
1176             fjz0             = _mm_setzero_ps();
1177             fjx1             = _mm_setzero_ps();
1178             fjy1             = _mm_setzero_ps();
1179             fjz1             = _mm_setzero_ps();
1180             fjx2             = _mm_setzero_ps();
1181             fjy2             = _mm_setzero_ps();
1182             fjz2             = _mm_setzero_ps();
1183             fjx3             = _mm_setzero_ps();
1184             fjy3             = _mm_setzero_ps();
1185             fjz3             = _mm_setzero_ps();
1186
1187             /**************************
1188              * CALCULATE INTERACTIONS *
1189              **************************/
1190
1191             /* LENNARD-JONES DISPERSION/REPULSION */
1192
1193             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1194             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1195
1196             fscal            = fvdw;
1197
1198              /* Update vectorial force */
1199             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1200             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1201             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1202
1203             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1204             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1205             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1206
1207             /**************************
1208              * CALCULATE INTERACTIONS *
1209              **************************/
1210
1211             /* REACTION-FIELD ELECTROSTATICS */
1212             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1213
1214             fscal            = felec;
1215
1216              /* Update vectorial force */
1217             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1218             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1219             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1220
1221             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1222             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1223             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1224
1225             /**************************
1226              * CALCULATE INTERACTIONS *
1227              **************************/
1228
1229             /* REACTION-FIELD ELECTROSTATICS */
1230             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1231
1232             fscal            = felec;
1233
1234              /* Update vectorial force */
1235             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1236             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1237             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1238
1239             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1240             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1241             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1242
1243             /**************************
1244              * CALCULATE INTERACTIONS *
1245              **************************/
1246
1247             /* REACTION-FIELD ELECTROSTATICS */
1248             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
1249
1250             fscal            = felec;
1251
1252              /* Update vectorial force */
1253             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1254             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1255             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1256
1257             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1258             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1259             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1260
1261             /**************************
1262              * CALCULATE INTERACTIONS *
1263              **************************/
1264
1265             /* REACTION-FIELD ELECTROSTATICS */
1266             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1267
1268             fscal            = felec;
1269
1270              /* Update vectorial force */
1271             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1272             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1273             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1274
1275             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1276             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1277             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1278
1279             /**************************
1280              * CALCULATE INTERACTIONS *
1281              **************************/
1282
1283             /* REACTION-FIELD ELECTROSTATICS */
1284             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1285
1286             fscal            = felec;
1287
1288              /* Update vectorial force */
1289             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1290             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1291             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1292
1293             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1294             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1295             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1296
1297             /**************************
1298              * CALCULATE INTERACTIONS *
1299              **************************/
1300
1301             /* REACTION-FIELD ELECTROSTATICS */
1302             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
1303
1304             fscal            = felec;
1305
1306              /* Update vectorial force */
1307             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1308             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1309             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1310
1311             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1312             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1313             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1314
1315             /**************************
1316              * CALCULATE INTERACTIONS *
1317              **************************/
1318
1319             /* REACTION-FIELD ELECTROSTATICS */
1320             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
1321
1322             fscal            = felec;
1323
1324              /* Update vectorial force */
1325             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1326             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1327             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1328
1329             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1330             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1331             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1332
1333             /**************************
1334              * CALCULATE INTERACTIONS *
1335              **************************/
1336
1337             /* REACTION-FIELD ELECTROSTATICS */
1338             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
1339
1340             fscal            = felec;
1341
1342              /* Update vectorial force */
1343             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1344             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1345             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1346
1347             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1348             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1349             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1350
1351             /**************************
1352              * CALCULATE INTERACTIONS *
1353              **************************/
1354
1355             /* REACTION-FIELD ELECTROSTATICS */
1356             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
1357
1358             fscal            = felec;
1359
1360              /* Update vectorial force */
1361             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1362             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1363             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1364
1365             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1366             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1367             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1368
1369             fjptrA             = f+j_coord_offsetA;
1370             fjptrB             = f+j_coord_offsetB;
1371             fjptrC             = f+j_coord_offsetC;
1372             fjptrD             = f+j_coord_offsetD;
1373
1374             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1375                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1376                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1377
1378             /* Inner loop uses 303 flops */
1379         }
1380
1381         if(jidx<j_index_end)
1382         {
1383
1384             /* Get j neighbor index, and coordinate index */
1385             jnrlistA         = jjnr[jidx];
1386             jnrlistB         = jjnr[jidx+1];
1387             jnrlistC         = jjnr[jidx+2];
1388             jnrlistD         = jjnr[jidx+3];
1389             /* Sign of each element will be negative for non-real atoms.
1390              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1391              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1392              */
1393             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1394             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1395             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1396             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1397             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1398             j_coord_offsetA  = DIM*jnrA;
1399             j_coord_offsetB  = DIM*jnrB;
1400             j_coord_offsetC  = DIM*jnrC;
1401             j_coord_offsetD  = DIM*jnrD;
1402
1403             /* load j atom coordinates */
1404             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1405                                               x+j_coord_offsetC,x+j_coord_offsetD,
1406                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1407                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1408
1409             /* Calculate displacement vector */
1410             dx00             = _mm_sub_ps(ix0,jx0);
1411             dy00             = _mm_sub_ps(iy0,jy0);
1412             dz00             = _mm_sub_ps(iz0,jz0);
1413             dx11             = _mm_sub_ps(ix1,jx1);
1414             dy11             = _mm_sub_ps(iy1,jy1);
1415             dz11             = _mm_sub_ps(iz1,jz1);
1416             dx12             = _mm_sub_ps(ix1,jx2);
1417             dy12             = _mm_sub_ps(iy1,jy2);
1418             dz12             = _mm_sub_ps(iz1,jz2);
1419             dx13             = _mm_sub_ps(ix1,jx3);
1420             dy13             = _mm_sub_ps(iy1,jy3);
1421             dz13             = _mm_sub_ps(iz1,jz3);
1422             dx21             = _mm_sub_ps(ix2,jx1);
1423             dy21             = _mm_sub_ps(iy2,jy1);
1424             dz21             = _mm_sub_ps(iz2,jz1);
1425             dx22             = _mm_sub_ps(ix2,jx2);
1426             dy22             = _mm_sub_ps(iy2,jy2);
1427             dz22             = _mm_sub_ps(iz2,jz2);
1428             dx23             = _mm_sub_ps(ix2,jx3);
1429             dy23             = _mm_sub_ps(iy2,jy3);
1430             dz23             = _mm_sub_ps(iz2,jz3);
1431             dx31             = _mm_sub_ps(ix3,jx1);
1432             dy31             = _mm_sub_ps(iy3,jy1);
1433             dz31             = _mm_sub_ps(iz3,jz1);
1434             dx32             = _mm_sub_ps(ix3,jx2);
1435             dy32             = _mm_sub_ps(iy3,jy2);
1436             dz32             = _mm_sub_ps(iz3,jz2);
1437             dx33             = _mm_sub_ps(ix3,jx3);
1438             dy33             = _mm_sub_ps(iy3,jy3);
1439             dz33             = _mm_sub_ps(iz3,jz3);
1440
1441             /* Calculate squared distance and things based on it */
1442             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1443             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1444             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1445             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1446             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1447             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1448             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1449             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1450             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1451             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1452
1453             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1454             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1455             rinv13           = gmx_mm_invsqrt_ps(rsq13);
1456             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1457             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1458             rinv23           = gmx_mm_invsqrt_ps(rsq23);
1459             rinv31           = gmx_mm_invsqrt_ps(rsq31);
1460             rinv32           = gmx_mm_invsqrt_ps(rsq32);
1461             rinv33           = gmx_mm_invsqrt_ps(rsq33);
1462
1463             rinvsq00         = gmx_mm_inv_ps(rsq00);
1464             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1465             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1466             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1467             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1468             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1469             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1470             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1471             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1472             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1473
1474             fjx0             = _mm_setzero_ps();
1475             fjy0             = _mm_setzero_ps();
1476             fjz0             = _mm_setzero_ps();
1477             fjx1             = _mm_setzero_ps();
1478             fjy1             = _mm_setzero_ps();
1479             fjz1             = _mm_setzero_ps();
1480             fjx2             = _mm_setzero_ps();
1481             fjy2             = _mm_setzero_ps();
1482             fjz2             = _mm_setzero_ps();
1483             fjx3             = _mm_setzero_ps();
1484             fjy3             = _mm_setzero_ps();
1485             fjz3             = _mm_setzero_ps();
1486
1487             /**************************
1488              * CALCULATE INTERACTIONS *
1489              **************************/
1490
1491             /* LENNARD-JONES DISPERSION/REPULSION */
1492
1493             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1494             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1495
1496             fscal            = fvdw;
1497
1498             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1499
1500              /* Update vectorial force */
1501             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1502             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1503             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1504
1505             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1506             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1507             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1508
1509             /**************************
1510              * CALCULATE INTERACTIONS *
1511              **************************/
1512
1513             /* REACTION-FIELD ELECTROSTATICS */
1514             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1515
1516             fscal            = felec;
1517
1518             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1519
1520              /* Update vectorial force */
1521             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1522             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1523             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1524
1525             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1526             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1527             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1528
1529             /**************************
1530              * CALCULATE INTERACTIONS *
1531              **************************/
1532
1533             /* REACTION-FIELD ELECTROSTATICS */
1534             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1535
1536             fscal            = felec;
1537
1538             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1539
1540              /* Update vectorial force */
1541             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1542             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1543             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1544
1545             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1546             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1547             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1548
1549             /**************************
1550              * CALCULATE INTERACTIONS *
1551              **************************/
1552
1553             /* REACTION-FIELD ELECTROSTATICS */
1554             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
1555
1556             fscal            = felec;
1557
1558             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1559
1560              /* Update vectorial force */
1561             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1562             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1563             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1564
1565             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1566             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1567             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1568
1569             /**************************
1570              * CALCULATE INTERACTIONS *
1571              **************************/
1572
1573             /* REACTION-FIELD ELECTROSTATICS */
1574             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1575
1576             fscal            = felec;
1577
1578             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1579
1580              /* Update vectorial force */
1581             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1582             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1583             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1584
1585             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1586             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1587             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1588
1589             /**************************
1590              * CALCULATE INTERACTIONS *
1591              **************************/
1592
1593             /* REACTION-FIELD ELECTROSTATICS */
1594             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1595
1596             fscal            = felec;
1597
1598             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1599
1600              /* Update vectorial force */
1601             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1602             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1603             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1604
1605             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1606             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1607             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1608
1609             /**************************
1610              * CALCULATE INTERACTIONS *
1611              **************************/
1612
1613             /* REACTION-FIELD ELECTROSTATICS */
1614             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
1615
1616             fscal            = felec;
1617
1618             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1619
1620              /* Update vectorial force */
1621             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1622             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1623             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1624
1625             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1626             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1627             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1628
1629             /**************************
1630              * CALCULATE INTERACTIONS *
1631              **************************/
1632
1633             /* REACTION-FIELD ELECTROSTATICS */
1634             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
1635
1636             fscal            = felec;
1637
1638             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1639
1640              /* Update vectorial force */
1641             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1642             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1643             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1644
1645             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1646             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1647             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1648
1649             /**************************
1650              * CALCULATE INTERACTIONS *
1651              **************************/
1652
1653             /* REACTION-FIELD ELECTROSTATICS */
1654             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
1655
1656             fscal            = felec;
1657
1658             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1659
1660              /* Update vectorial force */
1661             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1662             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1663             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1664
1665             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1666             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1667             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1668
1669             /**************************
1670              * CALCULATE INTERACTIONS *
1671              **************************/
1672
1673             /* REACTION-FIELD ELECTROSTATICS */
1674             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
1675
1676             fscal            = felec;
1677
1678             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1679
1680              /* Update vectorial force */
1681             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1682             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1683             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1684
1685             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1686             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1687             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1688
1689             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1690             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1691             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1692             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1693
1694             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1695                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1696                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1697
1698             /* Inner loop uses 303 flops */
1699         }
1700
1701         /* End of innermost loop */
1702
1703         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1704                                               f+i_coord_offset,fshift+i_shift_offset);
1705
1706         /* Increment number of inner iterations */
1707         inneriter                  += j_index_end - j_index_start;
1708
1709         /* Outer loop uses 24 flops */
1710     }
1711
1712     /* Increment number of outer iterations */
1713     outeriter        += nri;
1714
1715     /* Update outer/inner flops */
1716
1717     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*303);
1718 }