3 * This file is part of the GROMACS molecular simulation package.
5 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
6 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
7 * and including many others, as listed in the AUTHORS file in the
8 * top-level source directory and at http://www.gromacs.org.
10 * GROMACS is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public License
12 * as published by the Free Software Foundation; either version 2.1
13 * of the License, or (at your option) any later version.
15 * GROMACS is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with GROMACS; if not, see
22 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
23 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
25 * If you want to redistribute modifications to GROMACS, please
26 * consider that scientific software is very special. Version
27 * control is crucial - bugs must be traceable. We will be happy to
28 * consider code for inclusion in the official distribution, but
29 * derived work must not be called official GROMACS. Details are found
30 * in the README & COPYING files - if they are missing, get the
31 * official version at http://www.gromacs.org.
33 * To help us fund GROMACS development, we humbly ask that you cite
34 * the research papers on the package. Check out http://www.gromacs.org.
36 #error This file must be processed with the Gromacs pre-preprocessor
38 /* #if INCLUDE_HEADER */
45 #include "../nb_kernel.h"
46 #include "types/simple.h"
47 #include "gromacs/math/vec.h"
50 #include "gromacs/simd/math_x86_sse2_double.h"
51 #include "kernelutil_x86_sse2_double.h"
54 /* ## List of variables set by the generating script: */
56 /* ## Setttings that apply to the entire kernel: */
57 /* ## KERNEL_ELEC: String, choice for electrostatic interactions */
58 /* ## KERNEL_VDW: String, choice for van der Waals interactions */
59 /* ## KERNEL_NAME: String, name of this kernel */
60 /* ## KERNEL_VF: String telling if we calculate potential, force, or both */
61 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
63 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops: */
64 /* ## PARTICLES_I[]/ Arrays with lists of i/j particles to use in kernel. It is */
65 /* ## PARTICLES_J[]: just [0] for particle geometry, but can be longer for water */
66 /* ## PARTICLES_ELEC_I[]/ Arrays with lists of i/j particle that have electrostatics */
67 /* ## PARTICLES_ELEC_J[]: interactions that should be calculated in this kernel. */
68 /* ## PARTICLES_VDW_I[]/ Arrays with the list of i/j particle that have VdW */
69 /* ## PARTICLES_VDW_J[]: interactions that should be calculated in this kernel. */
71 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle) */
72 /* ## PAIRS_IJ[]: Array with (i,j) tuples of pairs for which interactions */
73 /* ## should be calculated in this kernel. Zero-charge particles */
74 /* ## do not have interactions with particles without vdw, and */
75 /* ## Vdw-only interactions are not evaluated in a no-vdw-kernel. */
76 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
77 /* ## For each i-j pair, the element [I][J] is a list of strings */
78 /* ## defining properties/flags of this interaction. Examples */
79 /* ## include 'electrostatics'/'vdw' if that type of interaction */
80 /* ## should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values */
81 /* ## are needed, and 'exactcutoff' or 'shift','switch' to */
82 /* ## decide if the force/potential should be modified. This way */
83 /* ## we only calculate values absolutely needed for each case. */
85 /* ## Calculate the size and offset for (merged/interleaved) table data */
88 * Gromacs nonbonded kernel: {KERNEL_NAME}
89 * Electrostatics interaction: {KERNEL_ELEC}
90 * VdW interaction: {KERNEL_VDW}
91 * Geometry: {GEOMETRY_I}-{GEOMETRY_J}
92 * Calculate force/pot: {KERNEL_VF}
96 (t_nblist * gmx_restrict nlist,
97 rvec * gmx_restrict xx,
98 rvec * gmx_restrict ff,
99 t_forcerec * gmx_restrict fr,
100 t_mdatoms * gmx_restrict mdatoms,
101 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
102 t_nrnb * gmx_restrict nrnb)
104 /* ## Not all variables are used for all kernels, but any optimizing compiler fixes that, */
105 /* ## so there is no point in going to extremes to exclude variables that are not needed. */
106 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
107 * just 0 for non-waters.
108 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
109 * jnr indices corresponding to data put in the four positions in the SIMD register.
111 int i_shift_offset,i_coord_offset,outeriter,inneriter;
112 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
114 int j_coord_offsetA,j_coord_offsetB;
115 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
117 real *shiftvec,*fshift,*x,*f;
118 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
119 /* #for I in PARTICLES_I */
121 __m128d ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
123 /* #for J in PARTICLES_J */
124 int vdwjidx{J}A,vdwjidx{J}B;
125 __m128d jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
127 /* #for I,J in PAIRS_IJ */
128 __m128d dx{I}{J},dy{I}{J},dz{I}{J},rsq{I}{J},rinv{I}{J},rinvsq{I}{J},r{I}{J},qq{I}{J},c6_{I}{J},c12_{I}{J};
130 /* #if KERNEL_ELEC != 'None' */
131 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
134 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
136 __m128d vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
137 __m128d minushalf = _mm_set1_pd(-0.5);
138 real *invsqrta,*dvda,*gbtab;
140 /* #if KERNEL_VDW != 'None' */
142 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
145 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
146 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
148 /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
150 __m128i ifour = _mm_set1_epi32(4);
151 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
154 /* #if 'LJEwald' in KERNEL_VDW */
155 /* #for I,J in PAIRS_IJ */
156 __m128d c6grid_{I}{J};
158 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
160 __m128d one_half = _mm_set1_pd(0.5);
161 __m128d minus_one = _mm_set1_pd(-1.0);
163 /* #if 'Ewald' in KERNEL_ELEC */
165 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
168 /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
169 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
170 real rswitch_scalar,d_scalar;
172 __m128d dummy_mask,cutoff_mask;
173 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
174 __m128d one = _mm_set1_pd(1.0);
175 __m128d two = _mm_set1_pd(2.0);
181 jindex = nlist->jindex;
183 shiftidx = nlist->shift;
185 shiftvec = fr->shift_vec[0];
186 fshift = fr->fshift[0];
187 /* #if KERNEL_ELEC != 'None' */
188 facel = _mm_set1_pd(fr->epsfac);
189 charge = mdatoms->chargeA;
190 /* #if 'ReactionField' in KERNEL_ELEC */
191 krf = _mm_set1_pd(fr->ic->k_rf);
192 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
193 crf = _mm_set1_pd(fr->ic->c_rf);
196 /* #if KERNEL_VDW != 'None' */
197 nvdwtype = fr->ntype;
199 vdwtype = mdatoms->typeA;
201 /* #if 'LJEwald' in KERNEL_VDW */
202 vdwgridparam = fr->ljpme_c6grid;
203 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
204 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
205 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
208 /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
209 vftab = kernel_data->table_elec_vdw->data;
210 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
211 /* #elif 'Table' in KERNEL_ELEC */
212 vftab = kernel_data->table_elec->data;
213 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
214 /* #elif 'Table' in KERNEL_VDW */
215 vftab = kernel_data->table_vdw->data;
216 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
219 /* #if 'Ewald' in KERNEL_ELEC */
220 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
221 /* #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
222 ewtab = fr->ic->tabq_coul_F;
223 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
224 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
226 ewtab = fr->ic->tabq_coul_FDV0;
227 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
228 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
232 /* #if KERNEL_ELEC=='GeneralizedBorn' */
233 invsqrta = fr->invsqrta;
235 gbtabscale = _mm_set1_pd(fr->gbtab.scale);
236 gbtab = fr->gbtab.data;
237 gbinvepsdiff = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
240 /* #if 'Water' in GEOMETRY_I */
241 /* Setup water-specific parameters */
242 inr = nlist->iinr[0];
243 /* #for I in PARTICLES_ELEC_I */
244 iq{I} = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+{I}]));
246 /* #for I in PARTICLES_VDW_I */
247 vdwioffset{I} = 2*nvdwtype*vdwtype[inr+{I}];
251 /* #if 'Water' in GEOMETRY_J */
252 /* #for J in PARTICLES_ELEC_J */
253 jq{J} = _mm_set1_pd(charge[inr+{J}]);
255 /* #for J in PARTICLES_VDW_J */
256 vdwjidx{J}A = 2*vdwtype[inr+{J}];
258 /* #for I,J in PAIRS_IJ */
259 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
260 qq{I}{J} = _mm_mul_pd(iq{I},jq{J});
262 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
263 /* #if 'LJEwald' in KERNEL_VDW */
264 c6_{I}{J} = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
265 c12_{I}{J} = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
266 c6grid_{I}{J} = _mm_set1_pd(vdwgridparam[vdwioffset{I}+vdwjidx{J}A]);
268 c6_{I}{J} = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
269 c12_{I}{J} = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
275 /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
276 /* #if KERNEL_ELEC!='None' */
277 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
278 rcutoff_scalar = fr->rcoulomb;
280 rcutoff_scalar = fr->rvdw;
282 rcutoff = _mm_set1_pd(rcutoff_scalar);
283 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
286 /* #if KERNEL_MOD_VDW=='PotentialShift' */
287 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
288 rvdw = _mm_set1_pd(fr->rvdw);
291 /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
292 /* #if KERNEL_MOD_ELEC=='PotentialSwitch' */
293 rswitch_scalar = fr->rcoulomb_switch;
294 rswitch = _mm_set1_pd(rswitch_scalar);
296 rswitch_scalar = fr->rvdw_switch;
297 rswitch = _mm_set1_pd(rswitch_scalar);
299 /* Setup switch parameters */
300 d_scalar = rcutoff_scalar-rswitch_scalar;
301 d = _mm_set1_pd(d_scalar);
302 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
303 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
304 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
305 /* #if 'Force' in KERNEL_VF */
306 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
307 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
308 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
312 /* Avoid stupid compiler warnings */
317 /* ## Keep track of the floating point operations we issue for reporting! */
318 /* #define OUTERFLOPS 0 */
322 /* Start outer loop over neighborlists */
323 for(iidx=0; iidx<nri; iidx++)
325 /* Load shift vector for this list */
326 i_shift_offset = DIM*shiftidx[iidx];
328 /* Load limits for loop over neighbors */
329 j_index_start = jindex[iidx];
330 j_index_end = jindex[iidx+1];
332 /* Get outer coordinate index */
334 i_coord_offset = DIM*inr;
336 /* Load i particle coords and add shift vector */
337 /* #if GEOMETRY_I == 'Particle' */
338 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
339 /* #elif GEOMETRY_I == 'Water3' */
340 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
341 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
342 /* #elif GEOMETRY_I == 'Water4' */
343 /* #if 0 in PARTICLES_I */
344 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
345 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
347 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
348 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
352 /* #if 'Force' in KERNEL_VF */
353 /* #for I in PARTICLES_I */
354 fix{I} = _mm_setzero_pd();
355 fiy{I} = _mm_setzero_pd();
356 fiz{I} = _mm_setzero_pd();
360 /* ## For water we already preloaded parameters at the start of the kernel */
361 /* #if not 'Water' in GEOMETRY_I */
362 /* Load parameters for i particles */
363 /* #for I in PARTICLES_ELEC_I */
364 iq{I} = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+{I}));
365 /* #define OUTERFLOPS OUTERFLOPS+1 */
366 /* #if KERNEL_ELEC=='GeneralizedBorn' */
367 isai{I} = _mm_load1_pd(invsqrta+inr+{I});
370 /* #for I in PARTICLES_VDW_I */
371 vdwioffset{I} = 2*nvdwtype*vdwtype[inr+{I}];
375 /* #if 'Potential' in KERNEL_VF */
376 /* Reset potential sums */
377 /* #if KERNEL_ELEC != 'None' */
378 velecsum = _mm_setzero_pd();
380 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
381 vgbsum = _mm_setzero_pd();
383 /* #if KERNEL_VDW != 'None' */
384 vvdwsum = _mm_setzero_pd();
387 /* #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
388 dvdasum = _mm_setzero_pd();
391 /* #for ROUND in ['Loop','Epilogue'] */
393 /* #if ROUND =='Loop' */
394 /* Start inner kernel loop */
395 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
397 /* ## First round is normal loop (next statement resets indentation) */
404 /* ## Second round is epilogue */
406 /* #define INNERFLOPS 0 */
408 /* #if ROUND =='Loop' */
409 /* Get j neighbor index, and coordinate index */
412 j_coord_offsetA = DIM*jnrA;
413 j_coord_offsetB = DIM*jnrB;
415 /* load j atom coordinates */
416 /* #if GEOMETRY_J == 'Particle' */
417 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
419 /* #elif GEOMETRY_J == 'Water3' */
420 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
421 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
422 /* #elif GEOMETRY_J == 'Water4' */
423 /* #if 0 in PARTICLES_J */
424 gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
425 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
426 &jy2,&jz2,&jx3,&jy3,&jz3);
428 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
429 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
434 j_coord_offsetA = DIM*jnrA;
436 /* load j atom coordinates */
437 /* #if GEOMETRY_J == 'Particle' */
438 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
440 /* #elif GEOMETRY_J == 'Water3' */
441 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
442 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
443 /* #elif GEOMETRY_J == 'Water4' */
444 /* #if 0 in PARTICLES_J */
445 gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
446 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
447 &jy2,&jz2,&jx3,&jy3,&jz3);
449 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
450 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
455 /* Calculate displacement vector */
456 /* #for I,J in PAIRS_IJ */
457 dx{I}{J} = _mm_sub_pd(ix{I},jx{J});
458 dy{I}{J} = _mm_sub_pd(iy{I},jy{J});
459 dz{I}{J} = _mm_sub_pd(iz{I},jz{J});
460 /* #define INNERFLOPS INNERFLOPS+3 */
463 /* Calculate squared distance and things based on it */
464 /* #for I,J in PAIRS_IJ */
465 rsq{I}{J} = gmx_mm_calc_rsq_pd(dx{I}{J},dy{I}{J},dz{I}{J});
466 /* #define INNERFLOPS INNERFLOPS+5 */
469 /* #for I,J in PAIRS_IJ */
470 /* #if 'rinv' in INTERACTION_FLAGS[I][J] */
471 rinv{I}{J} = gmx_mm_invsqrt_pd(rsq{I}{J});
472 /* #define INNERFLOPS INNERFLOPS+5 */
476 /* #for I,J in PAIRS_IJ */
477 /* #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
478 /* # if 'rinv' not in INTERACTION_FLAGS[I][J] */
479 rinvsq{I}{J} = gmx_mm_inv_pd(rsq{I}{J});
480 /* #define INNERFLOPS INNERFLOPS+4 */
482 rinvsq{I}{J} = _mm_mul_pd(rinv{I}{J},rinv{I}{J});
483 /* #define INNERFLOPS INNERFLOPS+1 */
488 /* #if not 'Water' in GEOMETRY_J */
489 /* Load parameters for j particles */
490 /* #for J in PARTICLES_ELEC_J */
491 /* #if ROUND =='Loop' */
492 jq{J} = gmx_mm_load_2real_swizzle_pd(charge+jnrA+{J},charge+jnrB+{J});
494 jq{J} = _mm_load_sd(charge+jnrA+{J});
496 /* #if KERNEL_ELEC=='GeneralizedBorn' */
497 /* #if ROUND =='Loop' */
498 isaj{J} = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+{J},invsqrta+jnrB+{J});
500 isaj{J} = _mm_load_sd(invsqrta+jnrA+{J});
504 /* #for J in PARTICLES_VDW_J */
505 vdwjidx{J}A = 2*vdwtype[jnrA+{J}];
506 /* #if ROUND =='Loop' */
507 vdwjidx{J}B = 2*vdwtype[jnrB+{J}];
512 /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
513 /* #for J in PARTICLES_J */
514 fjx{J} = _mm_setzero_pd();
515 fjy{J} = _mm_setzero_pd();
516 fjz{J} = _mm_setzero_pd();
520 /* #for I,J in PAIRS_IJ */
522 /**************************
523 * CALCULATE INTERACTIONS *
524 **************************/
526 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
527 /* ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
528 if (gmx_mm_any_lt(rsq{I}{J},rcutoff2))
530 /* #if 0 ## this and the next two lines is a hack to maintain auto-indentation in template file */
533 /* #define INNERFLOPS INNERFLOPS+1 */
536 /* #if 'r' in INTERACTION_FLAGS[I][J] */
537 r{I}{J} = _mm_mul_pd(rsq{I}{J},rinv{I}{J});
538 /* #define INNERFLOPS INNERFLOPS+1 */
541 /* ## For water geometries we already loaded parameters at the start of the kernel */
542 /* #if not 'Water' in GEOMETRY_J */
543 /* Compute parameters for interactions between i and j atoms */
544 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
545 qq{I}{J} = _mm_mul_pd(iq{I},jq{J});
546 /* #define INNERFLOPS INNERFLOPS+1 */
548 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
549 /* #if ROUND == 'Loop' */
550 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset{I}+vdwjidx{J}A,
551 vdwparam+vdwioffset{I}+vdwjidx{J}B,&c6_{I}{J},&c12_{I}{J});
553 /* #if 'LJEwald' in KERNEL_VDW */
554 c6grid_{I}{J} = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset{I}+vdwjidx{J}A,
555 vdwgridparam+vdwioffset{I}+vdwjidx{J}B);
558 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset{I}+vdwjidx{J}A,&c6_{I}{J},&c12_{I}{J});
560 /* #if 'LJEwald' in KERNEL_VDW */
561 c6grid_{I}{J} = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset{I}+vdwjidx{J}A);
567 /* #if 'table' in INTERACTION_FLAGS[I][J] */
568 /* Calculate table index by multiplying r with table scale and truncate to integer */
569 rt = _mm_mul_pd(r{I}{J},vftabscale);
570 vfitab = _mm_cvttpd_epi32(rt);
571 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
572 /* #define INNERFLOPS INNERFLOPS+4 */
573 /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
574 /* ## 3 tables, 4 data per point: multiply index by 12 */
575 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
576 /* #elif 'Table' in KERNEL_ELEC */
577 /* ## 1 table, 4 data per point: multiply index by 4 */
578 vfitab = _mm_slli_epi32(vfitab,2);
579 /* #elif 'Table' in KERNEL_VDW */
580 /* ## 2 tables, 4 data per point: multiply index by 8 */
581 vfitab = _mm_slli_epi32(vfitab,3);
585 /* ## ELECTROSTATIC INTERACTIONS */
586 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
588 /* #if KERNEL_ELEC=='Coulomb' */
590 /* COULOMB ELECTROSTATICS */
591 velec = _mm_mul_pd(qq{I}{J},rinv{I}{J});
592 /* #define INNERFLOPS INNERFLOPS+1 */
593 /* #if 'Force' in KERNEL_VF */
594 felec = _mm_mul_pd(velec,rinvsq{I}{J});
595 /* #define INNERFLOPS INNERFLOPS+2 */
598 /* #elif KERNEL_ELEC=='ReactionField' */
600 /* REACTION-FIELD ELECTROSTATICS */
601 /* #if 'Potential' in KERNEL_VF */
602 velec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_add_pd(rinv{I}{J},_mm_mul_pd(krf,rsq{I}{J})),crf));
603 /* #define INNERFLOPS INNERFLOPS+4 */
605 /* #if 'Force' in KERNEL_VF */
606 felec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_mul_pd(rinv{I}{J},rinvsq{I}{J}),krf2));
607 /* #define INNERFLOPS INNERFLOPS+3 */
610 /* #elif KERNEL_ELEC=='GeneralizedBorn' */
612 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
613 isaprod = _mm_mul_pd(isai{I},isaj{J});
614 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq{I}{J},_mm_mul_pd(isaprod,gbinvepsdiff)));
615 gbscale = _mm_mul_pd(isaprod,gbtabscale);
616 /* #define INNERFLOPS INNERFLOPS+5 */
618 /* Calculate generalized born table index - this is a separate table from the normal one,
619 * but we use the same procedure by multiplying r with scale and truncating to integer.
621 rt = _mm_mul_pd(r{I}{J},gbscale);
622 gbitab = _mm_cvttpd_epi32(rt);
623 gbeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
624 gbitab = _mm_slli_epi32(gbitab,2);
626 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
627 /* #if ROUND == 'Loop' */
628 F = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
630 F = _mm_setzero_pd();
632 GMX_MM_TRANSPOSE2_PD(Y,F);
633 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
634 /* #if ROUND == 'Loop' */
635 H = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
637 H = _mm_setzero_pd();
639 GMX_MM_TRANSPOSE2_PD(G,H);
640 Heps = _mm_mul_pd(gbeps,H);
641 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
642 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
643 vgb = _mm_mul_pd(gbqqfactor,VV);
644 /* #define INNERFLOPS INNERFLOPS+10 */
646 /* #if 'Force' in KERNEL_VF */
647 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
648 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
649 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r{I}{J})));
650 /* #if ROUND == 'Epilogue' */
651 dvdatmp = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
653 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
654 /* #if ROUND == 'Loop' */
655 gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj{J},isaj{J})));
657 gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj{J},isaj{J})));
659 /* #define INNERFLOPS INNERFLOPS+13 */
661 velec = _mm_mul_pd(qq{I}{J},rinv{I}{J});
662 /* #define INNERFLOPS INNERFLOPS+1 */
663 /* #if 'Force' in KERNEL_VF */
664 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv{I}{J}),fgb),rinv{I}{J});
665 /* #define INNERFLOPS INNERFLOPS+3 */
668 /* #elif KERNEL_ELEC=='Ewald' */
669 /* EWALD ELECTROSTATICS */
671 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
672 ewrt = _mm_mul_pd(r{I}{J},ewtabscale);
673 ewitab = _mm_cvttpd_epi32(ewrt);
674 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
675 /* #define INNERFLOPS INNERFLOPS+4 */
676 /* #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
677 ewitab = _mm_slli_epi32(ewitab,2);
678 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
679 /* #if ROUND == 'Loop' */
680 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
682 ewtabD = _mm_setzero_pd();
684 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
685 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
686 /* #if ROUND == 'Loop' */
687 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
689 ewtabFn = _mm_setzero_pd();
691 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
692 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
693 /* #define INNERFLOPS INNERFLOPS+2 */
694 /* #if KERNEL_MOD_ELEC=='PotentialShift' */
695 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
696 velec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_sub_pd(rinv{I}{J},sh_ewald),velec));
697 /* #define INNERFLOPS INNERFLOPS+7 */
699 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
700 velec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(rinv{I}{J},velec));
701 /* #define INNERFLOPS INNERFLOPS+6 */
703 /* #if 'Force' in KERNEL_VF */
704 felec = _mm_mul_pd(_mm_mul_pd(qq{I}{J},rinv{I}{J}),_mm_sub_pd(rinvsq{I}{J},felec));
705 /* #define INNERFLOPS INNERFLOPS+3 */
707 /* #elif KERNEL_VF=='Force' */
708 /* #if ROUND == 'Loop' */
709 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
712 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
714 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
715 felec = _mm_mul_pd(_mm_mul_pd(qq{I}{J},rinv{I}{J}),_mm_sub_pd(rinvsq{I}{J},felec));
716 /* #define INNERFLOPS INNERFLOPS+7 */
719 /* #elif KERNEL_ELEC=='CubicSplineTable' */
721 /* CUBIC SPLINE TABLE ELECTROSTATICS */
722 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
723 /* #if ROUND == 'Loop' */
724 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
726 F = _mm_setzero_pd();
728 GMX_MM_TRANSPOSE2_PD(Y,F);
729 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
730 /* #if ROUND == 'Loop' */
731 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
733 H = _mm_setzero_pd();
735 GMX_MM_TRANSPOSE2_PD(G,H);
736 Heps = _mm_mul_pd(vfeps,H);
737 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
738 /* #define INNERFLOPS INNERFLOPS+4 */
739 /* #if 'Potential' in KERNEL_VF */
740 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
741 velec = _mm_mul_pd(qq{I}{J},VV);
742 /* #define INNERFLOPS INNERFLOPS+3 */
744 /* #if 'Force' in KERNEL_VF */
745 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
746 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq{I}{J},FF),_mm_mul_pd(vftabscale,rinv{I}{J})));
747 /* #define INNERFLOPS INNERFLOPS+7 */
750 /* ## End of check for electrostatics interaction forms */
752 /* ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
754 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
756 /* #if KERNEL_VDW=='LennardJones' */
758 /* LENNARD-JONES DISPERSION/REPULSION */
760 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
761 /* #define INNERFLOPS INNERFLOPS+2 */
762 /* #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
763 vvdw6 = _mm_mul_pd(c6_{I}{J},rinvsix);
764 vvdw12 = _mm_mul_pd(c12_{I}{J},_mm_mul_pd(rinvsix,rinvsix));
765 /* #define INNERFLOPS INNERFLOPS+3 */
766 /* #if KERNEL_MOD_VDW=='PotentialShift' */
767 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_{I}{J},_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
768 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
769 /* #define INNERFLOPS INNERFLOPS+8 */
771 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
772 /* #define INNERFLOPS INNERFLOPS+3 */
774 /* ## Check for force inside potential check, i.e. this means we already did the potential part */
775 /* #if 'Force' in KERNEL_VF */
776 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq{I}{J});
777 /* #define INNERFLOPS INNERFLOPS+2 */
779 /* #elif KERNEL_VF=='Force' */
780 /* ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
781 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_{I}{J},rinvsix),c6_{I}{J}),_mm_mul_pd(rinvsix,rinvsq{I}{J}));
782 /* #define INNERFLOPS INNERFLOPS+4 */
785 /* #elif KERNEL_VDW=='CubicSplineTable' */
787 /* CUBIC SPLINE TABLE DISPERSION */
788 /* #if 'Table' in KERNEL_ELEC */
789 vfitab = _mm_add_epi32(vfitab,ifour);
791 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
792 /* #if ROUND == 'Loop' */
793 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
795 F = _mm_setzero_pd();
797 GMX_MM_TRANSPOSE2_PD(Y,F);
798 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
799 /* #if ROUND == 'Loop' */
800 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
802 H = _mm_setzero_pd();
804 GMX_MM_TRANSPOSE2_PD(G,H);
805 Heps = _mm_mul_pd(vfeps,H);
806 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
807 /* #define INNERFLOPS INNERFLOPS+4 */
808 /* #if 'Potential' in KERNEL_VF */
809 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
810 vvdw6 = _mm_mul_pd(c6_{I}{J},VV);
811 /* #define INNERFLOPS INNERFLOPS+3 */
813 /* #if 'Force' in KERNEL_VF */
814 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
815 fvdw6 = _mm_mul_pd(c6_{I}{J},FF);
816 /* #define INNERFLOPS INNERFLOPS+4 */
819 /* CUBIC SPLINE TABLE REPULSION */
820 vfitab = _mm_add_epi32(vfitab,ifour);
821 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
822 /* #if ROUND == 'Loop' */
823 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
825 F = _mm_setzero_pd();
827 GMX_MM_TRANSPOSE2_PD(Y,F);
828 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
829 /* #if ROUND == 'Loop' */
830 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
832 H = _mm_setzero_pd();
834 GMX_MM_TRANSPOSE2_PD(G,H);
835 Heps = _mm_mul_pd(vfeps,H);
836 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
837 /* #define INNERFLOPS INNERFLOPS+4 */
838 /* #if 'Potential' in KERNEL_VF */
839 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
840 vvdw12 = _mm_mul_pd(c12_{I}{J},VV);
841 /* #define INNERFLOPS INNERFLOPS+3 */
843 /* #if 'Force' in KERNEL_VF */
844 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
845 fvdw12 = _mm_mul_pd(c12_{I}{J},FF);
846 /* #define INNERFLOPS INNERFLOPS+5 */
848 /* #if 'Potential' in KERNEL_VF */
849 vvdw = _mm_add_pd(vvdw12,vvdw6);
850 /* #define INNERFLOPS INNERFLOPS+1 */
852 /* #if 'Force' in KERNEL_VF */
853 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv{I}{J})));
854 /* #define INNERFLOPS INNERFLOPS+4 */
857 /* #elif KERNEL_VDW=='LJEwald' */
859 /* Analytical LJ-PME */
860 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
861 ewcljrsq = _mm_mul_pd(ewclj2,rsq{I}{J});
862 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
863 exponent = gmx_simd_exp_d(ewcljrsq);
864 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
865 poly = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
866 /* #define INNERFLOPS INNERFLOPS+11 */
867 /* #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch'*/
868 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
869 vvdw6 = _mm_mul_pd(_mm_sub_pd(c6_{I}{J},_mm_mul_pd(c6grid_{I}{J},_mm_sub_pd(one,poly))),rinvsix);
870 vvdw12 = _mm_mul_pd(c12_{I}{J},_mm_mul_pd(rinvsix,rinvsix));
871 /* #define INNERFLOPS INNERFLOPS+6 */
872 /* #if KERNEL_MOD_VDW=='PotentialShift' */
873 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_{I}{J},_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
874 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_add_pd(_mm_mul_pd(c6_{I}{J},sh_vdw_invrcut6),_mm_mul_pd(c6grid_{I}{J},sh_lj_ewald))),one_sixth));
875 /* #define INNERFLOPS INNERFLOPS+10 */
877 vvdw = _mm_sub_pd(_mm_mul_pd(vvdw12,one_twelfth),_mm_mul_pd(vvdw6,one_sixth));
878 /* #define INNERFLOPS INNERFLOPS+3 */
880 /* ## Check for force inside potential check, i.e. this means we already did the potential part */
881 /* #if 'Force' in KERNEL_VF */
882 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
883 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,_mm_sub_pd(vvdw6,_mm_mul_pd(_mm_mul_pd(c6grid_{I}{J},one_sixth),_mm_mul_pd(exponent,ewclj6)))),rinvsq{I}{J});
884 /* #define INNERFLOPS INNERFLOPS+6 */
886 /* #elif KERNEL_VF=='Force' */
887 /* f6A = 6 * C6grid * (1 - poly) */
888 f6A = _mm_mul_pd(c6grid_{I}{J},_mm_sub_pd(one,poly));
889 /* f6B = C6grid * exponent * beta^6 */
890 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_{I}{J},one_sixth),_mm_mul_pd(exponent,ewclj6));
891 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
892 fvdw = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_{I}{J},rinvsix),_mm_sub_pd(c6_{I}{J},f6A)),rinvsix),f6B),rinvsq{I}{J});
893 /* #define INNERFLOPS INNERFLOPS+11 */
896 /* ## End of check for vdw interaction forms */
898 /* ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
900 /* #if 'switch' in INTERACTION_FLAGS[I][J] */
901 d = _mm_sub_pd(r{I}{J},rswitch);
902 d = _mm_max_pd(d,_mm_setzero_pd());
903 d2 = _mm_mul_pd(d,d);
904 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
905 /* #define INNERFLOPS INNERFLOPS+10 */
907 /* #if 'Force' in KERNEL_VF */
908 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
909 /* #define INNERFLOPS INNERFLOPS+5 */
912 /* Evaluate switch function */
913 /* #if 'Force' in KERNEL_VF */
914 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
915 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
916 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv{I}{J},_mm_mul_pd(velec,dsw)) );
917 /* #define INNERFLOPS INNERFLOPS+4 */
919 /* #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
920 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv{I}{J},_mm_mul_pd(vvdw,dsw)) );
921 /* #define INNERFLOPS INNERFLOPS+4 */
924 /* #if 'Potential' in KERNEL_VF */
925 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
926 velec = _mm_mul_pd(velec,sw);
927 /* #define INNERFLOPS INNERFLOPS+1 */
929 /* #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
930 vvdw = _mm_mul_pd(vvdw,sw);
931 /* #define INNERFLOPS INNERFLOPS+1 */
935 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
936 cutoff_mask = _mm_cmplt_pd(rsq{I}{J},rcutoff2);
937 /* #define INNERFLOPS INNERFLOPS+1 */
940 /* #if 'Potential' in KERNEL_VF */
941 /* Update potential sum for this i atom from the interaction with this j atom. */
942 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
943 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
944 velec = _mm_and_pd(velec,cutoff_mask);
945 /* #define INNERFLOPS INNERFLOPS+1 */
947 /* #if ROUND == 'Epilogue' */
948 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
950 velecsum = _mm_add_pd(velecsum,velec);
951 /* #define INNERFLOPS INNERFLOPS+1 */
952 /* #if KERNEL_ELEC=='GeneralizedBorn' */
953 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
954 vgb = _mm_and_pd(vgb,cutoff_mask);
955 /* #define INNERFLOPS INNERFLOPS+1 */
957 /* #if ROUND == 'Epilogue' */
958 vgb = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
960 vgbsum = _mm_add_pd(vgbsum,vgb);
961 /* #define INNERFLOPS INNERFLOPS+1 */
964 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
965 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
966 vvdw = _mm_and_pd(vvdw,cutoff_mask);
967 /* #define INNERFLOPS INNERFLOPS+1 */
969 /* #if ROUND == 'Epilogue' */
970 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
972 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
973 /* #define INNERFLOPS INNERFLOPS+1 */
977 /* #if 'Force' in KERNEL_VF */
979 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
980 fscal = _mm_add_pd(felec,fvdw);
981 /* #define INNERFLOPS INNERFLOPS+1 */
982 /* #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
984 /* #elif 'vdw' in INTERACTION_FLAGS[I][J] */
988 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
989 fscal = _mm_and_pd(fscal,cutoff_mask);
990 /* #define INNERFLOPS INNERFLOPS+1 */
993 /* #if ROUND == 'Epilogue' */
994 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
997 /* Calculate temporary vectorial force */
998 tx = _mm_mul_pd(fscal,dx{I}{J});
999 ty = _mm_mul_pd(fscal,dy{I}{J});
1000 tz = _mm_mul_pd(fscal,dz{I}{J});
1002 /* Update vectorial force */
1003 fix{I} = _mm_add_pd(fix{I},tx);
1004 fiy{I} = _mm_add_pd(fiy{I},ty);
1005 fiz{I} = _mm_add_pd(fiz{I},tz);
1006 /* #define INNERFLOPS INNERFLOPS+6 */
1008 /* #if GEOMETRY_I == 'Particle' */
1009 /* #if ROUND == 'Loop' */
1010 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
1012 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
1014 /* #define INNERFLOPS INNERFLOPS+3 */
1016 fjx{J} = _mm_add_pd(fjx{J},tx);
1017 fjy{J} = _mm_add_pd(fjy{J},ty);
1018 fjz{J} = _mm_add_pd(fjz{J},tz);
1019 /* #define INNERFLOPS INNERFLOPS+3 */
1024 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
1025 /* #if 0 ## This and next two lines is a hack to maintain indentation in template file */
1030 /* ## End of check for the interaction being outside the cutoff */
1033 /* ## End of loop over i-j interaction pairs */
1035 /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
1036 /* #if ROUND == 'Loop' */
1037 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1039 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1041 /* #define INNERFLOPS INNERFLOPS+3 */
1042 /* #elif GEOMETRY_J == 'Water3' */
1043 /* #if ROUND == 'Loop' */
1044 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1046 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1048 /* #define INNERFLOPS INNERFLOPS+9 */
1049 /* #elif GEOMETRY_J == 'Water4' */
1050 /* #if 0 in PARTICLES_J */
1051 /* #if ROUND == 'Loop' */
1052 gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1054 gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1056 /* #define INNERFLOPS INNERFLOPS+12 */
1058 /* #if ROUND == 'Loop' */
1059 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1061 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1063 /* #define INNERFLOPS INNERFLOPS+9 */
1067 /* Inner loop uses {INNERFLOPS} flops */
1072 /* End of innermost loop */
1074 /* #if 'Force' in KERNEL_VF */
1075 /* #if GEOMETRY_I == 'Particle' */
1076 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
1077 f+i_coord_offset,fshift+i_shift_offset);
1078 /* #define OUTERFLOPS OUTERFLOPS+6 */
1079 /* #elif GEOMETRY_I == 'Water3' */
1080 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1081 f+i_coord_offset,fshift+i_shift_offset);
1082 /* #define OUTERFLOPS OUTERFLOPS+18 */
1083 /* #elif GEOMETRY_I == 'Water4' */
1084 /* #if 0 in PARTICLES_I */
1085 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1086 f+i_coord_offset,fshift+i_shift_offset);
1087 /* #define OUTERFLOPS OUTERFLOPS+24 */
1089 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1090 f+i_coord_offset+DIM,fshift+i_shift_offset);
1091 /* #define OUTERFLOPS OUTERFLOPS+18 */
1096 /* #if 'Potential' in KERNEL_VF */
1098 /* Update potential energies */
1099 /* #if KERNEL_ELEC != 'None' */
1100 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1101 /* #define OUTERFLOPS OUTERFLOPS+1 */
1103 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
1104 gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
1105 /* #define OUTERFLOPS OUTERFLOPS+1 */
1107 /* #if KERNEL_VDW != 'None' */
1108 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1109 /* #define OUTERFLOPS OUTERFLOPS+1 */
1112 /* #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1113 dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai{I},isai{I}));
1114 gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
1117 /* Increment number of inner iterations */
1118 inneriter += j_index_end - j_index_start;
1120 /* Outer loop uses {OUTERFLOPS} flops */
1123 /* Increment number of outer iterations */
1126 /* Update outer/inner flops */
1127 /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1128 /* ## primitive and replaces aggressively even in strings inside these directives, we need to */
1129 /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source. */
1130 /* #if GEOMETRY_I == 'Water3' */
1131 /* #define ISUFFIX '_W3' */
1132 /* #elif GEOMETRY_I == 'Water4' */
1133 /* #define ISUFFIX '_W4' */
1135 /* #define ISUFFIX '' */
1137 /* #if GEOMETRY_J == 'Water3' */
1138 /* #define JSUFFIX 'W3' */
1139 /* #elif GEOMETRY_J == 'Water4' */
1140 /* #define JSUFFIX 'W4' */
1142 /* #define JSUFFIX '' */
1144 /* #if 'PotentialAndForce' in KERNEL_VF */
1145 /* #define VFSUFFIX '_VF' */
1146 /* #elif 'Potential' in KERNEL_VF */
1147 /* #define VFSUFFIX '_V' */
1149 /* #define VFSUFFIX '_F' */
1152 /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1153 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1154 /* #elif KERNEL_ELEC != 'None' */
1155 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1157 inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});