2 /* ## This file is part of the GROMACS molecular simulation package. */
4 /* ## Copyright (c) 2012, by the GROMACS development team, led by */
5 /* ## David van der Spoel, Berk Hess, Erik Lindahl, and including many */
6 /* ## others, as listed in the AUTHORS file in the top-level source */
7 /* ## directory and at http://www.gromacs.org. */
9 /* ## GROMACS is free software; you can redistribute it and/or */
10 /* ## modify it under the terms of the GNU Lesser General Public License */
11 /* ## as published by the Free Software Foundation; either version 2.1 */
12 /* ## of the License, or (at your option) any later version. */
14 /* ## GROMACS is distributed in the hope that it will be useful, */
15 /* ## but WITHOUT ANY WARRANTY; without even the implied warranty of */
16 /* ## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU */
17 /* ## Lesser General Public License for more details. */
19 /* ## You should have received a copy of the GNU Lesser General Public */
20 /* ## License along with GROMACS; if not, see */
21 /* ## http://www.gnu.org/licenses, or write to the Free Software Foundation, */
22 /* ## Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */
24 /* ## If you want to redistribute modifications to GROMACS, please */
25 /* ## consider that scientific software is very special. Version */
26 /* ## control is crucial - bugs must be traceable. We will be happy to */
27 /* ## consider code for inclusion in the official distribution, but */
28 /* ## derived work must not be called official GROMACS. Details are found */
29 /* ## in the README & COPYING files - if they are missing, get the */
30 /* ## official version at http://www.gromacs.org. */
32 /* ## To help us fund GROMACS development, we humbly ask that you cite */
33 /* ## the research papers on the package. Check out http://www.gromacs.org. */
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"
50 #include "gmx_math_x86_avx_128_fma_double.h"
51 #include "kernelutil_x86_avx_128_fma_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_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,twogbeps,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,twovfeps;
154 /* #if 'Ewald' in KERNEL_ELEC */
156 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
159 /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
160 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
161 real rswitch_scalar,d_scalar;
163 __m128d dummy_mask,cutoff_mask;
164 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
165 __m128d one = _mm_set1_pd(1.0);
166 __m128d two = _mm_set1_pd(2.0);
172 jindex = nlist->jindex;
174 shiftidx = nlist->shift;
176 shiftvec = fr->shift_vec[0];
177 fshift = fr->fshift[0];
178 /* #if KERNEL_ELEC != 'None' */
179 facel = _mm_set1_pd(fr->epsfac);
180 charge = mdatoms->chargeA;
181 /* #if 'ReactionField' in KERNEL_ELEC */
182 krf = _mm_set1_pd(fr->ic->k_rf);
183 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
184 crf = _mm_set1_pd(fr->ic->c_rf);
187 /* #if KERNEL_VDW != 'None' */
188 nvdwtype = fr->ntype;
190 vdwtype = mdatoms->typeA;
193 /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
194 vftab = kernel_data->table_elec_vdw->data;
195 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
196 /* #elif 'Table' in KERNEL_ELEC */
197 vftab = kernel_data->table_elec->data;
198 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
199 /* #elif 'Table' in KERNEL_VDW */
200 vftab = kernel_data->table_vdw->data;
201 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
204 /* #if 'Ewald' in KERNEL_ELEC */
205 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
206 /* #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
207 ewtab = fr->ic->tabq_coul_F;
208 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
209 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
211 ewtab = fr->ic->tabq_coul_FDV0;
212 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
213 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
217 /* #if KERNEL_ELEC=='GeneralizedBorn' */
218 invsqrta = fr->invsqrta;
220 gbtabscale = _mm_set1_pd(fr->gbtab.scale);
221 gbtab = fr->gbtab.data;
222 gbinvepsdiff = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
225 /* #if 'Water' in GEOMETRY_I */
226 /* Setup water-specific parameters */
227 inr = nlist->iinr[0];
228 /* #for I in PARTICLES_ELEC_I */
229 iq{I} = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+{I}]));
231 /* #for I in PARTICLES_VDW_I */
232 vdwioffset{I} = 2*nvdwtype*vdwtype[inr+{I}];
236 /* #if 'Water' in GEOMETRY_J */
237 /* #for J in PARTICLES_ELEC_J */
238 jq{J} = _mm_set1_pd(charge[inr+{J}]);
240 /* #for J in PARTICLES_VDW_J */
241 vdwjidx{J}A = 2*vdwtype[inr+{J}];
243 /* #for I,J in PAIRS_IJ */
244 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
245 qq{I}{J} = _mm_mul_pd(iq{I},jq{J});
247 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
248 c6_{I}{J} = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
249 c12_{I}{J} = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
254 /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
255 /* #if KERNEL_ELEC!='None' */
256 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
257 rcutoff_scalar = fr->rcoulomb;
259 rcutoff_scalar = fr->rvdw;
261 rcutoff = _mm_set1_pd(rcutoff_scalar);
262 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
265 /* #if KERNEL_MOD_VDW=='PotentialShift' */
266 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
267 rvdw = _mm_set1_pd(fr->rvdw);
270 /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
271 /* #if KERNEL_MOD_ELEC=='PotentialSwitch' */
272 rswitch_scalar = fr->rcoulomb_switch;
273 rswitch = _mm_set1_pd(rswitch_scalar);
275 rswitch_scalar = fr->rvdw_switch;
276 rswitch = _mm_set1_pd(rswitch_scalar);
278 /* Setup switch parameters */
279 d_scalar = rcutoff_scalar-rswitch_scalar;
280 d = _mm_set1_pd(d_scalar);
281 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
282 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
283 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
284 /* #if 'Force' in KERNEL_VF */
285 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
286 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
287 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
291 /* Avoid stupid compiler warnings */
296 /* ## Keep track of the floating point operations we issue for reporting! */
297 /* #define OUTERFLOPS 0 */
301 /* Start outer loop over neighborlists */
302 for(iidx=0; iidx<nri; iidx++)
304 /* Load shift vector for this list */
305 i_shift_offset = DIM*shiftidx[iidx];
307 /* Load limits for loop over neighbors */
308 j_index_start = jindex[iidx];
309 j_index_end = jindex[iidx+1];
311 /* Get outer coordinate index */
313 i_coord_offset = DIM*inr;
315 /* Load i particle coords and add shift vector */
316 /* #if GEOMETRY_I == 'Particle' */
317 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
318 /* #elif GEOMETRY_I == 'Water3' */
319 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
320 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
321 /* #elif GEOMETRY_I == 'Water4' */
322 /* #if 0 in PARTICLES_I */
323 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
324 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
326 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
327 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
331 /* #if 'Force' in KERNEL_VF */
332 /* #for I in PARTICLES_I */
333 fix{I} = _mm_setzero_pd();
334 fiy{I} = _mm_setzero_pd();
335 fiz{I} = _mm_setzero_pd();
339 /* ## For water we already preloaded parameters at the start of the kernel */
340 /* #if not 'Water' in GEOMETRY_I */
341 /* Load parameters for i particles */
342 /* #for I in PARTICLES_ELEC_I */
343 iq{I} = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+{I}));
344 /* #define OUTERFLOPS OUTERFLOPS+1 */
345 /* #if KERNEL_ELEC=='GeneralizedBorn' */
346 isai{I} = _mm_load1_pd(invsqrta+inr+{I});
349 /* #for I in PARTICLES_VDW_I */
350 vdwioffset{I} = 2*nvdwtype*vdwtype[inr+{I}];
354 /* #if 'Potential' in KERNEL_VF */
355 /* Reset potential sums */
356 /* #if KERNEL_ELEC != 'None' */
357 velecsum = _mm_setzero_pd();
359 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
360 vgbsum = _mm_setzero_pd();
362 /* #if KERNEL_VDW != 'None' */
363 vvdwsum = _mm_setzero_pd();
366 /* #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
367 dvdasum = _mm_setzero_pd();
370 /* #for ROUND in ['Loop','Epilogue'] */
372 /* #if ROUND =='Loop' */
373 /* Start inner kernel loop */
374 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
376 /* ## First round is normal loop (next statement resets indentation) */
383 /* ## Second round is epilogue */
385 /* #define INNERFLOPS 0 */
387 /* #if ROUND =='Loop' */
388 /* Get j neighbor index, and coordinate index */
391 j_coord_offsetA = DIM*jnrA;
392 j_coord_offsetB = DIM*jnrB;
394 /* load j atom coordinates */
395 /* #if GEOMETRY_J == 'Particle' */
396 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
398 /* #elif GEOMETRY_J == 'Water3' */
399 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
400 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
401 /* #elif GEOMETRY_J == 'Water4' */
402 /* #if 0 in PARTICLES_J */
403 gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
404 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
405 &jy2,&jz2,&jx3,&jy3,&jz3);
407 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
408 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
413 j_coord_offsetA = DIM*jnrA;
415 /* load j atom coordinates */
416 /* #if GEOMETRY_J == 'Particle' */
417 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
419 /* #elif GEOMETRY_J == 'Water3' */
420 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
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_1ptr_swizzle_pd(x+j_coord_offsetA,
425 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
426 &jy2,&jz2,&jx3,&jy3,&jz3);
428 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
429 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
434 /* Calculate displacement vector */
435 /* #for I,J in PAIRS_IJ */
436 dx{I}{J} = _mm_sub_pd(ix{I},jx{J});
437 dy{I}{J} = _mm_sub_pd(iy{I},jy{J});
438 dz{I}{J} = _mm_sub_pd(iz{I},jz{J});
439 /* #define INNERFLOPS INNERFLOPS+3 */
442 /* Calculate squared distance and things based on it */
443 /* #for I,J in PAIRS_IJ */
444 rsq{I}{J} = gmx_mm_calc_rsq_pd(dx{I}{J},dy{I}{J},dz{I}{J});
445 /* #define INNERFLOPS INNERFLOPS+5 */
448 /* #for I,J in PAIRS_IJ */
449 /* #if 'rinv' in INTERACTION_FLAGS[I][J] */
450 rinv{I}{J} = gmx_mm_invsqrt_pd(rsq{I}{J});
451 /* #define INNERFLOPS INNERFLOPS+5 */
455 /* #for I,J in PAIRS_IJ */
456 /* #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
457 /* # if 'rinv' not in INTERACTION_FLAGS[I][J] */
458 rinvsq{I}{J} = gmx_mm_inv_pd(rsq{I}{J});
459 /* #define INNERFLOPS INNERFLOPS+4 */
461 rinvsq{I}{J} = _mm_mul_pd(rinv{I}{J},rinv{I}{J});
462 /* #define INNERFLOPS INNERFLOPS+1 */
467 /* #if not 'Water' in GEOMETRY_J */
468 /* Load parameters for j particles */
469 /* #for J in PARTICLES_ELEC_J */
470 /* #if ROUND =='Loop' */
471 jq{J} = gmx_mm_load_2real_swizzle_pd(charge+jnrA+{J},charge+jnrB+{J});
473 jq{J} = _mm_load_sd(charge+jnrA+{J});
475 /* #if KERNEL_ELEC=='GeneralizedBorn' */
476 /* #if ROUND =='Loop' */
477 isaj{J} = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+{J},invsqrta+jnrB+{J});
479 isaj{J} = _mm_load_sd(invsqrta+jnrA+{J});
483 /* #for J in PARTICLES_VDW_J */
484 vdwjidx{J}A = 2*vdwtype[jnrA+{J}];
485 /* #if ROUND =='Loop' */
486 vdwjidx{J}B = 2*vdwtype[jnrB+{J}];
491 /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
492 /* #for J in PARTICLES_J */
493 fjx{J} = _mm_setzero_pd();
494 fjy{J} = _mm_setzero_pd();
495 fjz{J} = _mm_setzero_pd();
499 /* #for I,J in PAIRS_IJ */
501 /**************************
502 * CALCULATE INTERACTIONS *
503 **************************/
505 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
506 /* ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
507 if (gmx_mm_any_lt(rsq{I}{J},rcutoff2))
509 /* #if 0 ## this and the next two lines is a hack to maintain auto-indentation in template file */
512 /* #define INNERFLOPS INNERFLOPS+1 */
515 /* #if 'r' in INTERACTION_FLAGS[I][J] */
516 r{I}{J} = _mm_mul_pd(rsq{I}{J},rinv{I}{J});
517 /* #define INNERFLOPS INNERFLOPS+1 */
520 /* ## For water geometries we already loaded parameters at the start of the kernel */
521 /* #if not 'Water' in GEOMETRY_J */
522 /* Compute parameters for interactions between i and j atoms */
523 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
524 qq{I}{J} = _mm_mul_pd(iq{I},jq{J});
525 /* #define INNERFLOPS INNERFLOPS+1 */
527 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
528 /* #if ROUND == 'Loop' */
529 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset{I}+vdwjidx{J}A,
530 vdwparam+vdwioffset{I}+vdwjidx{J}B,&c6_{I}{J},&c12_{I}{J});
532 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset{I}+vdwjidx{J}A,&c6_{I}{J},&c12_{I}{J});
537 /* #if 'table' in INTERACTION_FLAGS[I][J] */
538 /* Calculate table index by multiplying r with table scale and truncate to integer */
539 rt = _mm_mul_pd(r{I}{J},vftabscale);
540 vfitab = _mm_cvttpd_epi32(rt);
542 vfeps = _mm_frcz_pd(rt);
544 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
546 twovfeps = _mm_add_pd(vfeps,vfeps);
547 /* #define INNERFLOPS INNERFLOPS+4 */
548 /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
549 /* ## 3 tables, 4 data per point: multiply index by 12 */
550 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
551 /* #elif 'Table' in KERNEL_ELEC */
552 /* ## 1 table, 4 data per point: multiply index by 4 */
553 vfitab = _mm_slli_epi32(vfitab,2);
554 /* #elif 'Table' in KERNEL_VDW */
555 /* ## 2 tables, 4 data per point: multiply index by 8 */
556 vfitab = _mm_slli_epi32(vfitab,3);
560 /* ## ELECTROSTATIC INTERACTIONS */
561 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
563 /* #if KERNEL_ELEC=='Coulomb' */
565 /* COULOMB ELECTROSTATICS */
566 velec = _mm_mul_pd(qq{I}{J},rinv{I}{J});
567 /* #define INNERFLOPS INNERFLOPS+1 */
568 /* #if 'Force' in KERNEL_VF */
569 felec = _mm_mul_pd(velec,rinvsq{I}{J});
570 /* #define INNERFLOPS INNERFLOPS+2 */
573 /* #elif KERNEL_ELEC=='ReactionField' */
575 /* REACTION-FIELD ELECTROSTATICS */
576 /* #if 'Potential' in KERNEL_VF */
577 velec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_macc_pd(krf,rsq{I}{J},rinv{I}{J}),crf));
578 /* #define INNERFLOPS INNERFLOPS+4 */
580 /* #if 'Force' in KERNEL_VF */
581 felec = _mm_mul_pd(qq{I}{J},_mm_msub_pd(rinv{I}{J},rinvsq{I}{J},krf2));
582 /* #define INNERFLOPS INNERFLOPS+3 */
585 /* #elif KERNEL_ELEC=='GeneralizedBorn' */
587 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
588 isaprod = _mm_mul_pd(isai{I},isaj{J});
589 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq{I}{J},_mm_mul_pd(isaprod,gbinvepsdiff)));
590 gbscale = _mm_mul_pd(isaprod,gbtabscale);
591 /* #define INNERFLOPS INNERFLOPS+5 */
593 /* Calculate generalized born table index - this is a separate table from the normal one,
594 * but we use the same procedure by multiplying r with scale and truncating to integer.
596 rt = _mm_mul_pd(r{I}{J},gbscale);
597 gbitab = _mm_cvttpd_epi32(rt);
599 gbeps = _mm_frcz_pd(rt);
601 gbeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
603 gbitab = _mm_slli_epi32(gbitab,2);
605 Y = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
606 /* #if ROUND == 'Loop' */
607 F = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,1) );
609 F = _mm_setzero_pd();
611 GMX_MM_TRANSPOSE2_PD(Y,F);
612 G = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,0) +2);
613 /* #if ROUND == 'Loop' */
614 H = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,1) +2);
616 H = _mm_setzero_pd();
618 GMX_MM_TRANSPOSE2_PD(G,H);
619 Fp = _mm_macc_pd(gbeps,_mm_macc_pd(gbeps,H,G),F);
620 VV = _mm_macc_pd(gbeps,Fp,Y);
621 vgb = _mm_mul_pd(gbqqfactor,VV);
622 /* #define INNERFLOPS INNERFLOPS+10 */
624 /* #if 'Force' in KERNEL_VF */
625 twogbeps = _mm_add_pd(gbeps,gbeps);
626 FF = _mm_macc_pd(_mm_macc_pd(twogbeps,H,G),gbeps,Fp);
627 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
628 dvdatmp = _mm_mul_pd(minushalf,_mm_macc_pd(fgb,r{I}{J},vgb));
629 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
630 /* #if ROUND == 'Loop' */
631 gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj{J},isaj{J})));
633 gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj{J},isaj{J})));
635 /* #define INNERFLOPS INNERFLOPS+13 */
637 velec = _mm_mul_pd(qq{I}{J},rinv{I}{J});
638 /* #define INNERFLOPS INNERFLOPS+1 */
639 /* #if 'Force' in KERNEL_VF */
640 felec = _mm_mul_pd(_mm_msub_pd(velec,rinv{I}{J},fgb),rinv{I}{J});
641 /* #define INNERFLOPS INNERFLOPS+3 */
644 /* #elif KERNEL_ELEC=='Ewald' */
645 /* EWALD ELECTROSTATICS */
647 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
648 ewrt = _mm_mul_pd(r{I}{J},ewtabscale);
649 ewitab = _mm_cvttpd_epi32(ewrt);
651 eweps = _mm_frcz_pd(ewrt);
653 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
655 twoeweps = _mm_add_pd(eweps,eweps);
656 /* #define INNERFLOPS INNERFLOPS+4 */
657 /* #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
658 ewitab = _mm_slli_epi32(ewitab,2);
659 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
660 /* #if ROUND == 'Loop' */
661 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
663 ewtabD = _mm_setzero_pd();
665 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
666 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
667 /* #if ROUND == 'Loop' */
668 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
670 ewtabFn = _mm_setzero_pd();
672 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
673 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
674 /* #define INNERFLOPS INNERFLOPS+2 */
675 /* #if KERNEL_MOD_ELEC=='PotentialShift' */
676 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
677 velec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_sub_pd(rinv{I}{J},sh_ewald),velec));
678 /* #define INNERFLOPS INNERFLOPS+7 */
680 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
681 velec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(rinv{I}{J},velec));
682 /* #define INNERFLOPS INNERFLOPS+6 */
684 /* #if 'Force' in KERNEL_VF */
685 felec = _mm_mul_pd(_mm_mul_pd(qq{I}{J},rinv{I}{J}),_mm_sub_pd(rinvsq{I}{J},felec));
686 /* #define INNERFLOPS INNERFLOPS+3 */
688 /* #elif KERNEL_VF=='Force' */
689 /* #if ROUND == 'Loop' */
690 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
693 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
695 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
696 felec = _mm_mul_pd(_mm_mul_pd(qq{I}{J},rinv{I}{J}),_mm_sub_pd(rinvsq{I}{J},felec));
697 /* #define INNERFLOPS INNERFLOPS+7 */
700 /* #elif KERNEL_ELEC=='CubicSplineTable' */
702 /* CUBIC SPLINE TABLE ELECTROSTATICS */
703 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
704 /* #if ROUND == 'Loop' */
705 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
707 F = _mm_setzero_pd();
709 GMX_MM_TRANSPOSE2_PD(Y,F);
710 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
711 /* #if ROUND == 'Loop' */
712 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
714 H = _mm_setzero_pd();
716 GMX_MM_TRANSPOSE2_PD(G,H);
717 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
718 /* #define INNERFLOPS INNERFLOPS+4 */
719 /* #if 'Potential' in KERNEL_VF */
720 VV = _mm_macc_pd(vfeps,Fp,Y);
721 velec = _mm_mul_pd(qq{I}{J},VV);
722 /* #define INNERFLOPS INNERFLOPS+3 */
724 /* #if 'Force' in KERNEL_VF */
725 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
726 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq{I}{J},FF),_mm_mul_pd(vftabscale,rinv{I}{J})));
727 /* #define INNERFLOPS INNERFLOPS+7 */
730 /* ## End of check for electrostatics interaction forms */
732 /* ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
734 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
736 /* #if KERNEL_VDW=='LennardJones' */
738 /* LENNARD-JONES DISPERSION/REPULSION */
740 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
741 /* #define INNERFLOPS INNERFLOPS+2 */
742 /* #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
743 vvdw6 = _mm_mul_pd(c6_{I}{J},rinvsix);
744 vvdw12 = _mm_mul_pd(c12_{I}{J},_mm_mul_pd(rinvsix,rinvsix));
745 /* #define INNERFLOPS INNERFLOPS+3 */
746 /* #if KERNEL_MOD_VDW=='PotentialShift' */
747 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_{I}{J},_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
748 _mm_mul_pd(_mm_nmacc_pd( c6_{I}{J},sh_vdw_invrcut6,vvdw6),one_sixth));
749 /* #define INNERFLOPS INNERFLOPS+8 */
751 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
752 /* #define INNERFLOPS INNERFLOPS+3 */
754 /* ## Check for force inside potential check, i.e. this means we already did the potential part */
755 /* #if 'Force' in KERNEL_VF */
756 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq{I}{J});
757 /* #define INNERFLOPS INNERFLOPS+2 */
759 /* #elif KERNEL_VF=='Force' */
760 /* ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
761 fvdw = _mm_mul_pd(_mm_msub_pd(c12_{I}{J},rinvsix,c6_{I}{J}),_mm_mul_pd(rinvsix,rinvsq{I}{J}));
762 /* #define INNERFLOPS INNERFLOPS+4 */
765 /* #elif KERNEL_VDW=='CubicSplineTable' */
767 /* CUBIC SPLINE TABLE DISPERSION */
768 /* #if 'Table' in KERNEL_ELEC */
769 vfitab = _mm_add_epi32(vfitab,ifour);
771 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
772 /* #if ROUND == 'Loop' */
773 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
775 F = _mm_setzero_pd();
777 GMX_MM_TRANSPOSE2_PD(Y,F);
778 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
779 /* #if ROUND == 'Loop' */
780 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
782 H = _mm_setzero_pd();
784 GMX_MM_TRANSPOSE2_PD(G,H);
785 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
786 /* #define INNERFLOPS INNERFLOPS+4 */
787 /* #if 'Potential' in KERNEL_VF */
788 VV = _mm_macc_pd(vfeps,Fp,Y);
789 vvdw6 = _mm_mul_pd(c6_{I}{J},VV);
790 /* #define INNERFLOPS INNERFLOPS+3 */
792 /* #if 'Force' in KERNEL_VF */
793 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
794 fvdw6 = _mm_mul_pd(c6_{I}{J},FF);
795 /* #define INNERFLOPS INNERFLOPS+4 */
798 /* CUBIC SPLINE TABLE REPULSION */
799 vfitab = _mm_add_epi32(vfitab,ifour);
800 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
801 /* #if ROUND == 'Loop' */
802 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
804 F = _mm_setzero_pd();
806 GMX_MM_TRANSPOSE2_PD(Y,F);
807 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
808 /* #if ROUND == 'Loop' */
809 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
811 H = _mm_setzero_pd();
813 GMX_MM_TRANSPOSE2_PD(G,H);
814 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
815 /* #define INNERFLOPS INNERFLOPS+4 */
816 /* #if 'Potential' in KERNEL_VF */
817 VV = _mm_macc_pd(vfeps,Fp,Y);
818 vvdw12 = _mm_mul_pd(c12_{I}{J},VV);
819 /* #define INNERFLOPS INNERFLOPS+3 */
821 /* #if 'Force' in KERNEL_VF */
822 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
823 fvdw12 = _mm_mul_pd(c12_{I}{J},FF);
824 /* #define INNERFLOPS INNERFLOPS+5 */
826 /* #if 'Potential' in KERNEL_VF */
827 vvdw = _mm_add_pd(vvdw12,vvdw6);
828 /* #define INNERFLOPS INNERFLOPS+1 */
830 /* #if 'Force' in KERNEL_VF */
831 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv{I}{J})));
832 /* #define INNERFLOPS INNERFLOPS+4 */
835 /* ## End of check for vdw interaction forms */
837 /* ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
839 /* #if 'switch' in INTERACTION_FLAGS[I][J] */
840 d = _mm_sub_pd(r{I}{J},rswitch);
841 d = _mm_max_pd(d,_mm_setzero_pd());
842 d2 = _mm_mul_pd(d,d);
843 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
844 /* #define INNERFLOPS INNERFLOPS+10 */
846 /* #if 'Force' in KERNEL_VF */
847 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
848 /* #define INNERFLOPS INNERFLOPS+5 */
851 /* Evaluate switch function */
852 /* #if 'Force' in KERNEL_VF */
853 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
854 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
855 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv{I}{J},_mm_mul_pd(velec,dsw)) );
856 /* #define INNERFLOPS INNERFLOPS+4 */
858 /* #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
859 fvdw = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv{I}{J},_mm_mul_pd(vvdw,dsw)) );
860 /* #define INNERFLOPS INNERFLOPS+4 */
863 /* #if 'Potential' in KERNEL_VF */
864 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
865 velec = _mm_mul_pd(velec,sw);
866 /* #define INNERFLOPS INNERFLOPS+1 */
868 /* #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
869 vvdw = _mm_mul_pd(vvdw,sw);
870 /* #define INNERFLOPS INNERFLOPS+1 */
874 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
875 cutoff_mask = _mm_cmplt_pd(rsq{I}{J},rcutoff2);
876 /* #define INNERFLOPS INNERFLOPS+1 */
879 /* #if 'Potential' in KERNEL_VF */
880 /* Update potential sum for this i atom from the interaction with this j atom. */
881 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
882 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
883 velec = _mm_and_pd(velec,cutoff_mask);
884 /* #define INNERFLOPS INNERFLOPS+1 */
886 /* #if ROUND == 'Epilogue' */
887 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
889 velecsum = _mm_add_pd(velecsum,velec);
890 /* #define INNERFLOPS INNERFLOPS+1 */
891 /* #if KERNEL_ELEC=='GeneralizedBorn' */
892 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
893 vgb = _mm_and_pd(vgb,cutoff_mask);
894 /* #define INNERFLOPS INNERFLOPS+1 */
896 /* #if ROUND == 'Epilogue' */
897 vgb = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
899 vgbsum = _mm_add_pd(vgbsum,vgb);
900 /* #define INNERFLOPS INNERFLOPS+1 */
903 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
904 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
905 vvdw = _mm_and_pd(vvdw,cutoff_mask);
906 /* #define INNERFLOPS INNERFLOPS+1 */
908 /* #if ROUND == 'Epilogue' */
909 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
911 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
912 /* #define INNERFLOPS INNERFLOPS+1 */
916 /* #if 'Force' in KERNEL_VF */
918 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
919 fscal = _mm_add_pd(felec,fvdw);
920 /* #define INNERFLOPS INNERFLOPS+1 */
921 /* #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
923 /* #elif 'vdw' in INTERACTION_FLAGS[I][J] */
927 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
928 fscal = _mm_and_pd(fscal,cutoff_mask);
929 /* #define INNERFLOPS INNERFLOPS+1 */
932 /* #if ROUND == 'Epilogue' */
933 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
936 /* ## Construction of vectorial force built into FMA instructions now */
937 /* #define INNERFLOPS INNERFLOPS+3 */
939 /* Update vectorial force */
940 fix{I} = _mm_macc_pd(dx{I}{J},fscal,fix{I});
941 fiy{I} = _mm_macc_pd(dy{I}{J},fscal,fiy{I});
942 fiz{I} = _mm_macc_pd(dz{I}{J},fscal,fiz{I});
943 /* #define INNERFLOPS INNERFLOPS+6 */
945 /* #if GEOMETRY_I == 'Particle' */
946 /* #if ROUND == 'Loop' */
947 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
948 _mm_mul_pd(dx{I}{J},fscal),
949 _mm_mul_pd(dy{I}{J},fscal),
950 _mm_mul_pd(dz{I}{J},fscal));
952 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
953 _mm_mul_pd(dx{I}{J},fscal),
954 _mm_mul_pd(dy{I}{J},fscal),
955 _mm_mul_pd(dz{I}{J},fscal));
957 /* #define INNERFLOPS INNERFLOPS+3 */
959 fjx{J} = _mm_macc_pd(dx{I}{J},fscal,fjx{J});
960 fjy{J} = _mm_macc_pd(dy{I}{J},fscal,fjy{J});
961 fjz{J} = _mm_macc_pd(dz{I}{J},fscal,fjz{J});
962 /* #define INNERFLOPS INNERFLOPS+3 */
967 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
968 /* #if 0 ## This and next two lines is a hack to maintain indentation in template file */
973 /* ## End of check for the interaction being outside the cutoff */
976 /* ## End of loop over i-j interaction pairs */
978 /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
979 /* #if ROUND == 'Loop' */
980 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
982 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
984 /* #define INNERFLOPS INNERFLOPS+3 */
985 /* #elif GEOMETRY_J == 'Water3' */
986 /* #if ROUND == 'Loop' */
987 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
989 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
991 /* #define INNERFLOPS INNERFLOPS+9 */
992 /* #elif GEOMETRY_J == 'Water4' */
993 /* #if 0 in PARTICLES_J */
994 /* #if ROUND == 'Loop' */
995 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);
997 gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
999 /* #define INNERFLOPS INNERFLOPS+12 */
1001 /* #if ROUND == 'Loop' */
1002 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);
1004 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1006 /* #define INNERFLOPS INNERFLOPS+9 */
1010 /* Inner loop uses {INNERFLOPS} flops */
1015 /* End of innermost loop */
1017 /* #if 'Force' in KERNEL_VF */
1018 /* #if GEOMETRY_I == 'Particle' */
1019 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
1020 f+i_coord_offset,fshift+i_shift_offset);
1021 /* #define OUTERFLOPS OUTERFLOPS+6 */
1022 /* #elif GEOMETRY_I == 'Water3' */
1023 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1024 f+i_coord_offset,fshift+i_shift_offset);
1025 /* #define OUTERFLOPS OUTERFLOPS+18 */
1026 /* #elif GEOMETRY_I == 'Water4' */
1027 /* #if 0 in PARTICLES_I */
1028 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1029 f+i_coord_offset,fshift+i_shift_offset);
1030 /* #define OUTERFLOPS OUTERFLOPS+24 */
1032 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1033 f+i_coord_offset+DIM,fshift+i_shift_offset);
1034 /* #define OUTERFLOPS OUTERFLOPS+18 */
1039 /* #if 'Potential' in KERNEL_VF */
1041 /* Update potential energies */
1042 /* #if KERNEL_ELEC != 'None' */
1043 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1044 /* #define OUTERFLOPS OUTERFLOPS+1 */
1046 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
1047 gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
1048 /* #define OUTERFLOPS OUTERFLOPS+1 */
1050 /* #if KERNEL_VDW != 'None' */
1051 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1052 /* #define OUTERFLOPS OUTERFLOPS+1 */
1055 /* #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1056 dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai{I},isai{I}));
1057 gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
1060 /* Increment number of inner iterations */
1061 inneriter += j_index_end - j_index_start;
1063 /* Outer loop uses {OUTERFLOPS} flops */
1066 /* Increment number of outer iterations */
1069 /* Update outer/inner flops */
1070 /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1071 /* ## primitive and replaces aggressively even in strings inside these directives, we need to */
1072 /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source. */
1073 /* #if GEOMETRY_I == 'Water3' */
1074 /* #define ISUFFIX '_W3' */
1075 /* #elif GEOMETRY_I == 'Water4' */
1076 /* #define ISUFFIX '_W4' */
1078 /* #define ISUFFIX '' */
1080 /* #if GEOMETRY_J == 'Water3' */
1081 /* #define JSUFFIX 'W3' */
1082 /* #elif GEOMETRY_J == 'Water4' */
1083 /* #define JSUFFIX 'W4' */
1085 /* #define JSUFFIX '' */
1087 /* #if 'PotentialAndForce' in KERNEL_VF */
1088 /* #define VFSUFFIX '_VF' */
1089 /* #elif 'Potential' in KERNEL_VF */
1090 /* #define VFSUFFIX '_V' */
1092 /* #define VFSUFFIX '_F' */
1095 /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1096 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1097 /* #elif KERNEL_ELEC != 'None' */
1098 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1100 inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});