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_single.h"
51 #include "kernelutil_x86_avx_128_fma_single.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,C,D refer to j loop unrolling done with AVX_128, e.g. for the four 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;
113 int jnrA,jnrB,jnrC,jnrD;
114 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
115 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
116 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
118 real *shiftvec,*fshift,*x,*f;
119 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
121 __m128 fscal,rcutoff,rcutoff2,jidxall;
122 /* #for I in PARTICLES_I */
124 __m128 ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
126 /* #for J in PARTICLES_J */
127 int vdwjidx{J}A,vdwjidx{J}B,vdwjidx{J}C,vdwjidx{J}D;
128 __m128 jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
130 /* #for I,J in PAIRS_IJ */
131 __m128 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};
133 /* #if KERNEL_ELEC != 'None' */
134 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
137 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
139 __m128 vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,twogbeps,dvdatmp;
140 __m128 minushalf = _mm_set1_ps(-0.5);
141 real *invsqrta,*dvda,*gbtab;
143 /* #if KERNEL_VDW != 'None' */
145 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
148 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
149 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
151 /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
153 __m128i ifour = _mm_set1_epi32(4);
154 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
157 /* #if 'Ewald' in KERNEL_ELEC */
159 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
160 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
163 /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
164 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
165 real rswitch_scalar,d_scalar;
167 __m128 dummy_mask,cutoff_mask;
168 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
169 __m128 one = _mm_set1_ps(1.0);
170 __m128 two = _mm_set1_ps(2.0);
176 jindex = nlist->jindex;
178 shiftidx = nlist->shift;
180 shiftvec = fr->shift_vec[0];
181 fshift = fr->fshift[0];
182 /* #if KERNEL_ELEC != 'None' */
183 facel = _mm_set1_ps(fr->epsfac);
184 charge = mdatoms->chargeA;
185 /* #if 'ReactionField' in KERNEL_ELEC */
186 krf = _mm_set1_ps(fr->ic->k_rf);
187 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
188 crf = _mm_set1_ps(fr->ic->c_rf);
191 /* #if KERNEL_VDW != 'None' */
192 nvdwtype = fr->ntype;
194 vdwtype = mdatoms->typeA;
197 /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
198 vftab = kernel_data->table_elec_vdw->data;
199 vftabscale = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
200 /* #elif 'Table' in KERNEL_ELEC */
201 vftab = kernel_data->table_elec->data;
202 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
203 /* #elif 'Table' in KERNEL_VDW */
204 vftab = kernel_data->table_vdw->data;
205 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
208 /* #if 'Ewald' in KERNEL_ELEC */
209 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
210 beta = _mm_set1_ps(fr->ic->ewaldcoeff);
211 beta2 = _mm_mul_ps(beta,beta);
212 beta3 = _mm_mul_ps(beta,beta2);
213 /* #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
214 ewtab = fr->ic->tabq_coul_F;
215 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
216 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
218 ewtab = fr->ic->tabq_coul_FDV0;
219 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
220 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
224 /* #if KERNEL_ELEC=='GeneralizedBorn' */
225 invsqrta = fr->invsqrta;
227 gbtabscale = _mm_set1_ps(fr->gbtab.scale);
228 gbtab = fr->gbtab.data;
229 gbinvepsdiff = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
232 /* #if 'Water' in GEOMETRY_I */
233 /* Setup water-specific parameters */
234 inr = nlist->iinr[0];
235 /* #for I in PARTICLES_ELEC_I */
236 iq{I} = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+{I}]));
238 /* #for I in PARTICLES_VDW_I */
239 vdwioffset{I} = 2*nvdwtype*vdwtype[inr+{I}];
243 /* #if 'Water' in GEOMETRY_J */
244 /* #for J in PARTICLES_ELEC_J */
245 jq{J} = _mm_set1_ps(charge[inr+{J}]);
247 /* #for J in PARTICLES_VDW_J */
248 vdwjidx{J}A = 2*vdwtype[inr+{J}];
250 /* #for I,J in PAIRS_IJ */
251 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
252 qq{I}{J} = _mm_mul_ps(iq{I},jq{J});
254 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
255 c6_{I}{J} = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
256 c12_{I}{J} = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
261 /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
262 /* #if KERNEL_ELEC!='None' */
263 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
264 rcutoff_scalar = fr->rcoulomb;
266 rcutoff_scalar = fr->rvdw;
268 rcutoff = _mm_set1_ps(rcutoff_scalar);
269 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
272 /* #if KERNEL_MOD_VDW=='PotentialShift' */
273 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
274 rvdw = _mm_set1_ps(fr->rvdw);
277 /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
278 /* #if KERNEL_MOD_ELEC=='PotentialSwitch' */
279 rswitch_scalar = fr->rcoulomb_switch;
280 rswitch = _mm_set1_ps(rswitch_scalar);
282 rswitch_scalar = fr->rvdw_switch;
283 rswitch = _mm_set1_ps(rswitch_scalar);
285 /* Setup switch parameters */
286 d_scalar = rcutoff_scalar-rswitch_scalar;
287 d = _mm_set1_ps(d_scalar);
288 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
289 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
290 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
291 /* #if 'Force' in KERNEL_VF */
292 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
293 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
294 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
298 /* Avoid stupid compiler warnings */
299 jnrA = jnrB = jnrC = jnrD = 0;
305 /* ## Keep track of the floating point operations we issue for reporting! */
306 /* #define OUTERFLOPS 0 */
310 for(iidx=0;iidx<4*DIM;iidx++)
315 /* Start outer loop over neighborlists */
316 for(iidx=0; iidx<nri; iidx++)
318 /* Load shift vector for this list */
319 i_shift_offset = DIM*shiftidx[iidx];
321 /* Load limits for loop over neighbors */
322 j_index_start = jindex[iidx];
323 j_index_end = jindex[iidx+1];
325 /* Get outer coordinate index */
327 i_coord_offset = DIM*inr;
329 /* Load i particle coords and add shift vector */
330 /* #if GEOMETRY_I == 'Particle' */
331 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
332 /* #elif GEOMETRY_I == 'Water3' */
333 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
334 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
335 /* #elif GEOMETRY_I == 'Water4' */
336 /* #if 0 in PARTICLES_I */
337 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
338 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
340 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
341 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
345 /* #if 'Force' in KERNEL_VF */
346 /* #for I in PARTICLES_I */
347 fix{I} = _mm_setzero_ps();
348 fiy{I} = _mm_setzero_ps();
349 fiz{I} = _mm_setzero_ps();
353 /* ## For water we already preloaded parameters at the start of the kernel */
354 /* #if not 'Water' in GEOMETRY_I */
355 /* Load parameters for i particles */
356 /* #for I in PARTICLES_ELEC_I */
357 iq{I} = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+{I}));
358 /* #define OUTERFLOPS OUTERFLOPS+1 */
359 /* #if KERNEL_ELEC=='GeneralizedBorn' */
360 isai{I} = _mm_load1_ps(invsqrta+inr+{I});
363 /* #for I in PARTICLES_VDW_I */
364 vdwioffset{I} = 2*nvdwtype*vdwtype[inr+{I}];
368 /* #if 'Potential' in KERNEL_VF */
369 /* Reset potential sums */
370 /* #if KERNEL_ELEC != 'None' */
371 velecsum = _mm_setzero_ps();
373 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
374 vgbsum = _mm_setzero_ps();
376 /* #if KERNEL_VDW != 'None' */
377 vvdwsum = _mm_setzero_ps();
380 /* #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
381 dvdasum = _mm_setzero_ps();
384 /* #for ROUND in ['Loop','Epilogue'] */
386 /* #if ROUND =='Loop' */
387 /* Start inner kernel loop */
388 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
390 /* ## First round is normal loop (next statement resets indentation) */
397 /* ## Second round is epilogue */
399 /* #define INNERFLOPS 0 */
401 /* Get j neighbor index, and coordinate index */
402 /* #if ROUND =='Loop' */
408 jnrlistA = jjnr[jidx];
409 jnrlistB = jjnr[jidx+1];
410 jnrlistC = jjnr[jidx+2];
411 jnrlistD = jjnr[jidx+3];
412 /* Sign of each element will be negative for non-real atoms.
413 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
414 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
416 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
417 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
418 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
419 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
420 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
422 j_coord_offsetA = DIM*jnrA;
423 j_coord_offsetB = DIM*jnrB;
424 j_coord_offsetC = DIM*jnrC;
425 j_coord_offsetD = DIM*jnrD;
427 /* load j atom coordinates */
428 /* #if GEOMETRY_J == 'Particle' */
429 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
430 x+j_coord_offsetC,x+j_coord_offsetD,
432 /* #elif GEOMETRY_J == 'Water3' */
433 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
434 x+j_coord_offsetC,x+j_coord_offsetD,
435 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
436 /* #elif GEOMETRY_J == 'Water4' */
437 /* #if 0 in PARTICLES_J */
438 gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
439 x+j_coord_offsetC,x+j_coord_offsetD,
440 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
441 &jy2,&jz2,&jx3,&jy3,&jz3);
443 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
444 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
445 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
449 /* Calculate displacement vector */
450 /* #for I,J in PAIRS_IJ */
451 dx{I}{J} = _mm_sub_ps(ix{I},jx{J});
452 dy{I}{J} = _mm_sub_ps(iy{I},jy{J});
453 dz{I}{J} = _mm_sub_ps(iz{I},jz{J});
454 /* #define INNERFLOPS INNERFLOPS+3 */
457 /* Calculate squared distance and things based on it */
458 /* #for I,J in PAIRS_IJ */
459 rsq{I}{J} = gmx_mm_calc_rsq_ps(dx{I}{J},dy{I}{J},dz{I}{J});
460 /* #define INNERFLOPS INNERFLOPS+5 */
463 /* #for I,J in PAIRS_IJ */
464 /* #if 'rinv' in INTERACTION_FLAGS[I][J] */
465 rinv{I}{J} = gmx_mm_invsqrt_ps(rsq{I}{J});
466 /* #define INNERFLOPS INNERFLOPS+5 */
470 /* #for I,J in PAIRS_IJ */
471 /* #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
472 /* # if 'rinv' not in INTERACTION_FLAGS[I][J] */
473 rinvsq{I}{J} = gmx_mm_inv_ps(rsq{I}{J});
474 /* #define INNERFLOPS INNERFLOPS+4 */
476 rinvsq{I}{J} = _mm_mul_ps(rinv{I}{J},rinv{I}{J});
477 /* #define INNERFLOPS INNERFLOPS+1 */
482 /* #if not 'Water' in GEOMETRY_J */
483 /* Load parameters for j particles */
484 /* #for J in PARTICLES_ELEC_J */
485 jq{J} = gmx_mm_load_4real_swizzle_ps(charge+jnrA+{J},charge+jnrB+{J},
486 charge+jnrC+{J},charge+jnrD+{J});
487 /* #if KERNEL_ELEC=='GeneralizedBorn' */
488 isaj{J} = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+{J},invsqrta+jnrB+{J},
489 invsqrta+jnrC+{J},invsqrta+jnrD+{J});
492 /* #for J in PARTICLES_VDW_J */
493 vdwjidx{J}A = 2*vdwtype[jnrA+{J}];
494 vdwjidx{J}B = 2*vdwtype[jnrB+{J}];
495 vdwjidx{J}C = 2*vdwtype[jnrC+{J}];
496 vdwjidx{J}D = 2*vdwtype[jnrD+{J}];
500 /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
501 /* #for J in PARTICLES_J */
502 fjx{J} = _mm_setzero_ps();
503 fjy{J} = _mm_setzero_ps();
504 fjz{J} = _mm_setzero_ps();
508 /* #for I,J in PAIRS_IJ */
510 /**************************
511 * CALCULATE INTERACTIONS *
512 **************************/
514 /* ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
515 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
516 /* ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
517 if (gmx_mm_any_lt(rsq{I}{J},rcutoff2))
519 /* #if 0 ## this and the next two lines is a hack to maintain auto-indentation in template file */
522 /* #define INNERFLOPS INNERFLOPS+1 */
525 /* #if 'r' in INTERACTION_FLAGS[I][J] */
526 r{I}{J} = _mm_mul_ps(rsq{I}{J},rinv{I}{J});
527 /* #if ROUND == 'Epilogue' */
528 r{I}{J} = _mm_andnot_ps(dummy_mask,r{I}{J});
529 /* #define INNERFLOPS INNERFLOPS+1 */
531 /* #define INNERFLOPS INNERFLOPS+1 */
534 /* ## For water geometries we already loaded parameters at the start of the kernel */
535 /* #if not 'Water' in GEOMETRY_J */
536 /* Compute parameters for interactions between i and j atoms */
537 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
538 qq{I}{J} = _mm_mul_ps(iq{I},jq{J});
539 /* #define INNERFLOPS INNERFLOPS+1 */
541 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
542 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset{I}+vdwjidx{J}A,
543 vdwparam+vdwioffset{I}+vdwjidx{J}B,
544 vdwparam+vdwioffset{I}+vdwjidx{J}C,
545 vdwparam+vdwioffset{I}+vdwjidx{J}D,
546 &c6_{I}{J},&c12_{I}{J});
550 /* #if 'table' in INTERACTION_FLAGS[I][J] */
551 /* Calculate table index by multiplying r with table scale and truncate to integer */
552 rt = _mm_mul_ps(r{I}{J},vftabscale);
553 vfitab = _mm_cvttps_epi32(rt);
555 vfeps = _mm_frcz_ps(rt);
557 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
559 twovfeps = _mm_add_ps(vfeps,vfeps);
560 /* #define INNERFLOPS INNERFLOPS+4 */
561 /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
562 /* ## 3 tables, 4 bytes per point: multiply index by 12 */
563 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
564 /* #elif 'Table' in KERNEL_ELEC */
565 /* ## 1 table, 4 bytes per point: multiply index by 4 */
566 vfitab = _mm_slli_epi32(vfitab,2);
567 /* #elif 'Table' in KERNEL_VDW */
568 /* ## 2 tables, 4 bytes per point: multiply index by 8 */
569 vfitab = _mm_slli_epi32(vfitab,3);
573 /* ## ELECTROSTATIC INTERACTIONS */
574 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
576 /* #if KERNEL_ELEC=='Coulomb' */
578 /* COULOMB ELECTROSTATICS */
579 velec = _mm_mul_ps(qq{I}{J},rinv{I}{J});
580 /* #define INNERFLOPS INNERFLOPS+1 */
581 /* #if 'Force' in KERNEL_VF */
582 felec = _mm_mul_ps(velec,rinvsq{I}{J});
583 /* #define INNERFLOPS INNERFLOPS+2 */
586 /* #elif KERNEL_ELEC=='ReactionField' */
588 /* REACTION-FIELD ELECTROSTATICS */
589 /* #if 'Potential' in KERNEL_VF */
590 velec = _mm_mul_ps(qq{I}{J},_mm_sub_ps(_mm_macc_ps(krf,rsq{I}{J},rinv{I}{J}),crf));
591 /* #define INNERFLOPS INNERFLOPS+4 */
593 /* #if 'Force' in KERNEL_VF */
594 felec = _mm_mul_ps(qq{I}{J},_mm_msub_ps(rinv{I}{J},rinvsq{I}{J},krf2));
595 /* #define INNERFLOPS INNERFLOPS+3 */
598 /* #elif KERNEL_ELEC=='GeneralizedBorn' */
600 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
601 isaprod = _mm_mul_ps(isai{I},isaj{J});
602 gbqqfactor = _mm_xor_ps(signbit,_mm_mul_ps(qq{I}{J},_mm_mul_ps(isaprod,gbinvepsdiff)));
603 gbscale = _mm_mul_ps(isaprod,gbtabscale);
604 /* #define INNERFLOPS INNERFLOPS+5 */
606 /* Calculate generalized born table index - this is a separate table from the normal one,
607 * but we use the same procedure by multiplying r with scale and truncating to integer.
609 rt = _mm_mul_ps(r{I}{J},gbscale);
610 gbitab = _mm_cvttps_epi32(rt);
612 gbeps = _mm_frcz_ps(rt);
614 gbeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
616 gbitab = _mm_slli_epi32(gbitab,2);
618 Y = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,0) );
619 F = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,1) );
620 G = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,2) );
621 H = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,3) );
622 _MM_TRANSPOSE4_PS(Y,F,G,H);
623 Fp = _mm_macc_ps(gbeps,_mm_macc_ps(gbeps,H,G),F);
624 VV = _mm_macc_ps(gbeps,Fp,Y);
625 vgb = _mm_mul_ps(gbqqfactor,VV);
626 /* #define INNERFLOPS INNERFLOPS+10 */
628 /* #if 'Force' in KERNEL_VF */
629 twogbeps = _mm_add_ps(gbeps,gbeps);
630 FF = _mm_macc_ps(_mm_macc_ps(twogbeps,H,G),gbeps,Fp);
631 fgb = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
632 dvdatmp = _mm_mul_ps(minushalf,_mm_macc_ps(fgb,r{I}{J},vgb));
633 dvdasum = _mm_add_ps(dvdasum,dvdatmp);
634 /* #if ROUND == 'Loop' */
640 /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
641 fjptrA = (jnrlistA>=0) ? dvda+jnrA : scratch;
642 fjptrB = (jnrlistB>=0) ? dvda+jnrB : scratch;
643 fjptrC = (jnrlistC>=0) ? dvda+jnrC : scratch;
644 fjptrD = (jnrlistD>=0) ? dvda+jnrD : scratch;
646 gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj{J},isaj{J})));
647 /* #define INNERFLOPS INNERFLOPS+13 */
649 velec = _mm_mul_ps(qq{I}{J},rinv{I}{J});
650 /* #define INNERFLOPS INNERFLOPS+1 */
651 /* #if 'Force' in KERNEL_VF */
652 felec = _mm_mul_ps(_mm_msub_ps(velec,rinv{I}{J},fgb),rinv{I}{J});
653 /* #define INNERFLOPS INNERFLOPS+3 */
656 /* #elif KERNEL_ELEC=='Ewald' */
657 /* EWALD ELECTROSTATICS */
659 /* Analytical PME correction */
660 zeta2 = _mm_mul_ps(beta2,rsq{I}{J});
661 /* #if 'Force' in KERNEL_VF */
662 rinv3 = _mm_mul_ps(rinvsq{I}{J},rinv{I}{J});
663 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
664 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
665 felec = _mm_mul_ps(qq{I}{J},felec);
667 /* #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
668 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
669 /* #if KERNEL_MOD_ELEC=='PotentialShift' */
670 velec = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv{I}{J},sh_ewald));
672 velec = _mm_nmacc_ps(pmecorrV,beta,rinv{I}{J});
674 velec = _mm_mul_ps(qq{I}{J},velec);
677 /* #elif KERNEL_ELEC=='CubicSplineTable' */
679 /* CUBIC SPLINE TABLE ELECTROSTATICS */
680 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
681 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
682 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
683 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
684 _MM_TRANSPOSE4_PS(Y,F,G,H);
685 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
686 /* #define INNERFLOPS INNERFLOPS+4 */
687 /* #if 'Potential' in KERNEL_VF */
688 VV = _mm_macc_ps(vfeps,Fp,Y);
689 velec = _mm_mul_ps(qq{I}{J},VV);
690 /* #define INNERFLOPS INNERFLOPS+3 */
692 /* #if 'Force' in KERNEL_VF */
693 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
694 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq{I}{J},FF),_mm_mul_ps(vftabscale,rinv{I}{J})));
695 /* #define INNERFLOPS INNERFLOPS+7 */
698 /* ## End of check for electrostatics interaction forms */
700 /* ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
702 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
704 /* #if KERNEL_VDW=='LennardJones' */
706 /* LENNARD-JONES DISPERSION/REPULSION */
708 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
709 /* #define INNERFLOPS INNERFLOPS+2 */
710 /* #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
711 vvdw6 = _mm_mul_ps(c6_{I}{J},rinvsix);
712 vvdw12 = _mm_mul_ps(c12_{I}{J},_mm_mul_ps(rinvsix,rinvsix));
713 /* #define INNERFLOPS INNERFLOPS+3 */
714 /* #if KERNEL_MOD_VDW=='PotentialShift' */
715 vvdw = _mm_msub_ps(_mm_nmacc_ps(c12_{I}{J},_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
716 _mm_mul_ps( _mm_nmacc_ps(c6_{I}{J},sh_vdw_invrcut6,vvdw6),one_sixth));
717 /* #define INNERFLOPS INNERFLOPS+8 */
719 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
720 /* #define INNERFLOPS INNERFLOPS+3 */
722 /* ## Check for force inside potential check, i.e. this means we already did the potential part */
723 /* #if 'Force' in KERNEL_VF */
724 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq{I}{J});
725 /* #define INNERFLOPS INNERFLOPS+2 */
727 /* #elif KERNEL_VF=='Force' */
728 /* ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
729 fvdw = _mm_mul_ps(_mm_msub_ps(c12_{I}{J},rinvsix,c6_{I}{J}),_mm_mul_ps(rinvsix,rinvsq{I}{J}));
730 /* #define INNERFLOPS INNERFLOPS+4 */
733 /* #elif KERNEL_VDW=='CubicSplineTable' */
735 /* CUBIC SPLINE TABLE DISPERSION */
736 /* #if 'Table' in KERNEL_ELEC */
737 vfitab = _mm_add_epi32(vfitab,ifour);
739 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
740 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
741 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
742 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
743 _MM_TRANSPOSE4_PS(Y,F,G,H);
744 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
745 /* #define INNERFLOPS INNERFLOPS+4 */
746 /* #if 'Potential' in KERNEL_VF */
747 VV = _mm_macc_ps(vfeps,Fp,Y);
748 vvdw6 = _mm_mul_ps(c6_{I}{J},VV);
749 /* #define INNERFLOPS INNERFLOPS+3 */
751 /* #if 'Force' in KERNEL_VF */
752 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
753 fvdw6 = _mm_mul_ps(c6_{I}{J},FF);
754 /* #define INNERFLOPS INNERFLOPS+4 */
757 /* CUBIC SPLINE TABLE REPULSION */
758 vfitab = _mm_add_epi32(vfitab,ifour);
759 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
760 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
761 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
762 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
763 _MM_TRANSPOSE4_PS(Y,F,G,H);
764 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
765 /* #define INNERFLOPS INNERFLOPS+4 */
766 /* #if 'Potential' in KERNEL_VF */
767 VV = _mm_macc_ps(vfeps,Fp,Y);
768 vvdw12 = _mm_mul_ps(c12_{I}{J},VV);
769 /* #define INNERFLOPS INNERFLOPS+3 */
771 /* #if 'Force' in KERNEL_VF */
772 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
773 fvdw12 = _mm_mul_ps(c12_{I}{J},FF);
774 /* #define INNERFLOPS INNERFLOPS+5 */
776 /* #if 'Potential' in KERNEL_VF */
777 vvdw = _mm_add_ps(vvdw12,vvdw6);
778 /* #define INNERFLOPS INNERFLOPS+1 */
780 /* #if 'Force' in KERNEL_VF */
781 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv{I}{J})));
782 /* #define INNERFLOPS INNERFLOPS+4 */
785 /* ## End of check for vdw interaction forms */
787 /* ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
789 /* #if 'switch' in INTERACTION_FLAGS[I][J] */
790 d = _mm_sub_ps(r{I}{J},rswitch);
791 d = _mm_max_ps(d,_mm_setzero_ps());
792 d2 = _mm_mul_ps(d,d);
793 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
794 /* #define INNERFLOPS INNERFLOPS+10 */
796 /* #if 'Force' in KERNEL_VF */
797 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
798 /* #define INNERFLOPS INNERFLOPS+5 */
801 /* Evaluate switch function */
802 /* #if 'Force' in KERNEL_VF */
803 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
804 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
805 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv{I}{J},_mm_mul_ps(velec,dsw)) );
806 /* #define INNERFLOPS INNERFLOPS+4 */
808 /* #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
809 fvdw = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv{I}{J},_mm_mul_ps(vvdw,dsw)) );
810 /* #define INNERFLOPS INNERFLOPS+4 */
813 /* #if 'Potential' in KERNEL_VF */
814 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
815 velec = _mm_mul_ps(velec,sw);
816 /* #define INNERFLOPS INNERFLOPS+1 */
818 /* #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
819 vvdw = _mm_mul_ps(vvdw,sw);
820 /* #define INNERFLOPS INNERFLOPS+1 */
824 /* ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
825 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
826 cutoff_mask = _mm_cmplt_ps(rsq{I}{J},rcutoff2);
827 /* #define INNERFLOPS INNERFLOPS+1 */
830 /* #if 'Potential' in KERNEL_VF */
831 /* Update potential sum for this i atom from the interaction with this j atom. */
832 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
833 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
834 velec = _mm_and_ps(velec,cutoff_mask);
835 /* #define INNERFLOPS INNERFLOPS+1 */
837 /* #if ROUND == 'Epilogue' */
838 velec = _mm_andnot_ps(dummy_mask,velec);
840 velecsum = _mm_add_ps(velecsum,velec);
841 /* #define INNERFLOPS INNERFLOPS+1 */
842 /* #if KERNEL_ELEC=='GeneralizedBorn' */
843 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
844 vgb = _mm_and_ps(vgb,cutoff_mask);
845 /* #define INNERFLOPS INNERFLOPS+1 */
847 /* #if ROUND == 'Epilogue' */
848 vgb = _mm_andnot_ps(dummy_mask,vgb);
850 vgbsum = _mm_add_ps(vgbsum,vgb);
851 /* #define INNERFLOPS INNERFLOPS+1 */
854 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
855 /* ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
856 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
857 vvdw = _mm_and_ps(vvdw,cutoff_mask);
858 /* #define INNERFLOPS INNERFLOPS+1 */
860 /* #if ROUND == 'Epilogue' */
861 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
863 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
864 /* #define INNERFLOPS INNERFLOPS+1 */
868 /* #if 'Force' in KERNEL_VF */
870 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
871 fscal = _mm_add_ps(felec,fvdw);
872 /* #define INNERFLOPS INNERFLOPS+1 */
873 /* #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
875 /* #elif 'vdw' in INTERACTION_FLAGS[I][J] */
879 /* ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
880 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
881 fscal = _mm_and_ps(fscal,cutoff_mask);
882 /* #define INNERFLOPS INNERFLOPS+1 */
885 /* #if ROUND == 'Epilogue' */
886 fscal = _mm_andnot_ps(dummy_mask,fscal);
889 /* ## Construction of vectorial force built into FMA instructions now */
890 /* #define INNERFLOPS INNERFLOPS+3 */
892 /* Update vectorial force */
893 fix{I} = _mm_macc_ps(dx{I}{J},fscal,fix{I});
894 fiy{I} = _mm_macc_ps(dy{I}{J},fscal,fiy{I});
895 fiz{I} = _mm_macc_ps(dz{I}{J},fscal,fiz{I});
896 /* #define INNERFLOPS INNERFLOPS+6 */
898 /* #if GEOMETRY_I == 'Particle' */
899 /* #if ROUND == 'Loop' */
900 fjptrA = f+j_coord_offsetA;
901 fjptrB = f+j_coord_offsetB;
902 fjptrC = f+j_coord_offsetC;
903 fjptrD = f+j_coord_offsetD;
905 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
906 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
907 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
908 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
910 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
911 _mm_mul_ps(dx{I}{J},fscal),
912 _mm_mul_ps(dy{I}{J},fscal),
913 _mm_mul_ps(dz{I}{J},fscal));
914 /* #define INNERFLOPS INNERFLOPS+3 */
916 fjx{J} = _mm_macc_ps(dx{I}{J},fscal,fjx{J});
917 fjy{J} = _mm_macc_ps(dy{I}{J},fscal,fjy{J});
918 fjz{J} = _mm_macc_ps(dz{I}{J},fscal,fjz{J});
919 /* #define INNERFLOPS INNERFLOPS+3 */
924 /* ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
925 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
926 /* #if 0 ## This and next two lines is a hack to maintain indentation in template file */
931 /* ## End of check for the interaction being outside the cutoff */
934 /* ## End of loop over i-j interaction pairs */
936 /* #if GEOMETRY_I != 'Particle' */
937 /* #if ROUND == 'Loop' */
938 fjptrA = f+j_coord_offsetA;
939 fjptrB = f+j_coord_offsetB;
940 fjptrC = f+j_coord_offsetC;
941 fjptrD = f+j_coord_offsetD;
943 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
944 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
945 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
946 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
950 /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
951 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
952 /* #elif GEOMETRY_J == 'Water3' */
953 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
954 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
955 /* #define INNERFLOPS INNERFLOPS+9 */
956 /* #elif GEOMETRY_J == 'Water4' */
957 /* #if 0 in PARTICLES_J */
958 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
959 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
960 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
961 /* #define INNERFLOPS INNERFLOPS+12 */
963 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
964 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
965 /* #define INNERFLOPS INNERFLOPS+9 */
969 /* Inner loop uses {INNERFLOPS} flops */
974 /* End of innermost loop */
976 /* #if 'Force' in KERNEL_VF */
977 /* #if GEOMETRY_I == 'Particle' */
978 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
979 f+i_coord_offset,fshift+i_shift_offset);
980 /* #define OUTERFLOPS OUTERFLOPS+6 */
981 /* #elif GEOMETRY_I == 'Water3' */
982 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
983 f+i_coord_offset,fshift+i_shift_offset);
984 /* #define OUTERFLOPS OUTERFLOPS+18 */
985 /* #elif GEOMETRY_I == 'Water4' */
986 /* #if 0 in PARTICLES_I */
987 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
988 f+i_coord_offset,fshift+i_shift_offset);
989 /* #define OUTERFLOPS OUTERFLOPS+24 */
991 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
992 f+i_coord_offset+DIM,fshift+i_shift_offset);
993 /* #define OUTERFLOPS OUTERFLOPS+18 */
998 /* #if 'Potential' in KERNEL_VF */
1000 /* Update potential energies */
1001 /* #if KERNEL_ELEC != 'None' */
1002 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1003 /* #define OUTERFLOPS OUTERFLOPS+1 */
1005 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
1006 gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
1007 /* #define OUTERFLOPS OUTERFLOPS+1 */
1009 /* #if KERNEL_VDW != 'None' */
1010 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1011 /* #define OUTERFLOPS OUTERFLOPS+1 */
1014 /* #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1015 dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai{I},isai{I}));
1016 gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
1019 /* Increment number of inner iterations */
1020 inneriter += j_index_end - j_index_start;
1022 /* Outer loop uses {OUTERFLOPS} flops */
1025 /* Increment number of outer iterations */
1028 /* Update outer/inner flops */
1029 /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1030 /* ## primitive and replaces aggressively even in strings inside these directives, we need to */
1031 /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source. */
1032 /* #if GEOMETRY_I == 'Water3' */
1033 /* #define ISUFFIX '_W3' */
1034 /* #elif GEOMETRY_I == 'Water4' */
1035 /* #define ISUFFIX '_W4' */
1037 /* #define ISUFFIX '' */
1039 /* #if GEOMETRY_J == 'Water3' */
1040 /* #define JSUFFIX 'W3' */
1041 /* #elif GEOMETRY_J == 'Water4' */
1042 /* #define JSUFFIX 'W4' */
1044 /* #define JSUFFIX '' */
1046 /* #if 'PotentialAndForce' in KERNEL_VF */
1047 /* #define VFSUFFIX '_VF' */
1048 /* #elif 'Potential' in KERNEL_VF */
1049 /* #define VFSUFFIX '_V' */
1051 /* #define VFSUFFIX '_F' */
1054 /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1055 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1056 /* #elif KERNEL_ELEC != 'None' */
1057 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1059 inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});