2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
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 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
50 #include "kernelutil_x86_avx_128_fma_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_avx_128_fma_double
54 * Electrostatics interaction: Ewald
55 * VdW interaction: None
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_avx_128_fma_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 int vdwjidx0A,vdwjidx0B;
89 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
91 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
92 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
93 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
96 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
98 __m128d dummy_mask,cutoff_mask;
99 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
100 __m128d one = _mm_set1_pd(1.0);
101 __m128d two = _mm_set1_pd(2.0);
107 jindex = nlist->jindex;
109 shiftidx = nlist->shift;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm_set1_pd(fr->epsfac);
114 charge = mdatoms->chargeA;
116 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
117 ewtab = fr->ic->tabq_coul_FDV0;
118 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
119 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
121 /* Setup water-specific parameters */
122 inr = nlist->iinr[0];
123 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
124 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
125 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
127 /* Avoid stupid compiler warnings */
135 /* Start outer loop over neighborlists */
136 for(iidx=0; iidx<nri; iidx++)
138 /* Load shift vector for this list */
139 i_shift_offset = DIM*shiftidx[iidx];
141 /* Load limits for loop over neighbors */
142 j_index_start = jindex[iidx];
143 j_index_end = jindex[iidx+1];
145 /* Get outer coordinate index */
147 i_coord_offset = DIM*inr;
149 /* Load i particle coords and add shift vector */
150 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
151 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
153 fix0 = _mm_setzero_pd();
154 fiy0 = _mm_setzero_pd();
155 fiz0 = _mm_setzero_pd();
156 fix1 = _mm_setzero_pd();
157 fiy1 = _mm_setzero_pd();
158 fiz1 = _mm_setzero_pd();
159 fix2 = _mm_setzero_pd();
160 fiy2 = _mm_setzero_pd();
161 fiz2 = _mm_setzero_pd();
163 /* Reset potential sums */
164 velecsum = _mm_setzero_pd();
166 /* Start inner kernel loop */
167 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
170 /* Get j neighbor index, and coordinate index */
173 j_coord_offsetA = DIM*jnrA;
174 j_coord_offsetB = DIM*jnrB;
176 /* load j atom coordinates */
177 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
180 /* Calculate displacement vector */
181 dx00 = _mm_sub_pd(ix0,jx0);
182 dy00 = _mm_sub_pd(iy0,jy0);
183 dz00 = _mm_sub_pd(iz0,jz0);
184 dx10 = _mm_sub_pd(ix1,jx0);
185 dy10 = _mm_sub_pd(iy1,jy0);
186 dz10 = _mm_sub_pd(iz1,jz0);
187 dx20 = _mm_sub_pd(ix2,jx0);
188 dy20 = _mm_sub_pd(iy2,jy0);
189 dz20 = _mm_sub_pd(iz2,jz0);
191 /* Calculate squared distance and things based on it */
192 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
193 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
194 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
196 rinv00 = gmx_mm_invsqrt_pd(rsq00);
197 rinv10 = gmx_mm_invsqrt_pd(rsq10);
198 rinv20 = gmx_mm_invsqrt_pd(rsq20);
200 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
201 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
202 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
204 /* Load parameters for j particles */
205 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
207 fjx0 = _mm_setzero_pd();
208 fjy0 = _mm_setzero_pd();
209 fjz0 = _mm_setzero_pd();
211 /**************************
212 * CALCULATE INTERACTIONS *
213 **************************/
215 r00 = _mm_mul_pd(rsq00,rinv00);
217 /* Compute parameters for interactions between i and j atoms */
218 qq00 = _mm_mul_pd(iq0,jq0);
220 /* EWALD ELECTROSTATICS */
222 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
223 ewrt = _mm_mul_pd(r00,ewtabscale);
224 ewitab = _mm_cvttpd_epi32(ewrt);
226 eweps = _mm_frcz_pd(ewrt);
228 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
230 twoeweps = _mm_add_pd(eweps,eweps);
231 ewitab = _mm_slli_epi32(ewitab,2);
232 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
233 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
234 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
235 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
236 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
237 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
238 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
239 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
240 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
241 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
243 /* Update potential sum for this i atom from the interaction with this j atom. */
244 velecsum = _mm_add_pd(velecsum,velec);
248 /* Update vectorial force */
249 fix0 = _mm_macc_pd(dx00,fscal,fix0);
250 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
251 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
253 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
254 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
255 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
257 /**************************
258 * CALCULATE INTERACTIONS *
259 **************************/
261 r10 = _mm_mul_pd(rsq10,rinv10);
263 /* Compute parameters for interactions between i and j atoms */
264 qq10 = _mm_mul_pd(iq1,jq0);
266 /* EWALD ELECTROSTATICS */
268 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
269 ewrt = _mm_mul_pd(r10,ewtabscale);
270 ewitab = _mm_cvttpd_epi32(ewrt);
272 eweps = _mm_frcz_pd(ewrt);
274 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
276 twoeweps = _mm_add_pd(eweps,eweps);
277 ewitab = _mm_slli_epi32(ewitab,2);
278 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
279 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
280 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
281 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
282 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
283 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
284 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
285 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
286 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
287 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
289 /* Update potential sum for this i atom from the interaction with this j atom. */
290 velecsum = _mm_add_pd(velecsum,velec);
294 /* Update vectorial force */
295 fix1 = _mm_macc_pd(dx10,fscal,fix1);
296 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
297 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
299 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
300 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
301 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
303 /**************************
304 * CALCULATE INTERACTIONS *
305 **************************/
307 r20 = _mm_mul_pd(rsq20,rinv20);
309 /* Compute parameters for interactions between i and j atoms */
310 qq20 = _mm_mul_pd(iq2,jq0);
312 /* EWALD ELECTROSTATICS */
314 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
315 ewrt = _mm_mul_pd(r20,ewtabscale);
316 ewitab = _mm_cvttpd_epi32(ewrt);
318 eweps = _mm_frcz_pd(ewrt);
320 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
322 twoeweps = _mm_add_pd(eweps,eweps);
323 ewitab = _mm_slli_epi32(ewitab,2);
324 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
325 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
326 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
327 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
328 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
329 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
330 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
331 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
332 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
333 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
335 /* Update potential sum for this i atom from the interaction with this j atom. */
336 velecsum = _mm_add_pd(velecsum,velec);
340 /* Update vectorial force */
341 fix2 = _mm_macc_pd(dx20,fscal,fix2);
342 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
343 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
345 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
346 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
347 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
349 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
351 /* Inner loop uses 135 flops */
358 j_coord_offsetA = DIM*jnrA;
360 /* load j atom coordinates */
361 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
364 /* Calculate displacement vector */
365 dx00 = _mm_sub_pd(ix0,jx0);
366 dy00 = _mm_sub_pd(iy0,jy0);
367 dz00 = _mm_sub_pd(iz0,jz0);
368 dx10 = _mm_sub_pd(ix1,jx0);
369 dy10 = _mm_sub_pd(iy1,jy0);
370 dz10 = _mm_sub_pd(iz1,jz0);
371 dx20 = _mm_sub_pd(ix2,jx0);
372 dy20 = _mm_sub_pd(iy2,jy0);
373 dz20 = _mm_sub_pd(iz2,jz0);
375 /* Calculate squared distance and things based on it */
376 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
377 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
378 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
380 rinv00 = gmx_mm_invsqrt_pd(rsq00);
381 rinv10 = gmx_mm_invsqrt_pd(rsq10);
382 rinv20 = gmx_mm_invsqrt_pd(rsq20);
384 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
385 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
386 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
388 /* Load parameters for j particles */
389 jq0 = _mm_load_sd(charge+jnrA+0);
391 fjx0 = _mm_setzero_pd();
392 fjy0 = _mm_setzero_pd();
393 fjz0 = _mm_setzero_pd();
395 /**************************
396 * CALCULATE INTERACTIONS *
397 **************************/
399 r00 = _mm_mul_pd(rsq00,rinv00);
401 /* Compute parameters for interactions between i and j atoms */
402 qq00 = _mm_mul_pd(iq0,jq0);
404 /* EWALD ELECTROSTATICS */
406 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
407 ewrt = _mm_mul_pd(r00,ewtabscale);
408 ewitab = _mm_cvttpd_epi32(ewrt);
410 eweps = _mm_frcz_pd(ewrt);
412 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
414 twoeweps = _mm_add_pd(eweps,eweps);
415 ewitab = _mm_slli_epi32(ewitab,2);
416 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
417 ewtabD = _mm_setzero_pd();
418 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
419 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
420 ewtabFn = _mm_setzero_pd();
421 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
422 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
423 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
424 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
425 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
427 /* Update potential sum for this i atom from the interaction with this j atom. */
428 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
429 velecsum = _mm_add_pd(velecsum,velec);
433 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
435 /* Update vectorial force */
436 fix0 = _mm_macc_pd(dx00,fscal,fix0);
437 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
438 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
440 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
441 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
442 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
444 /**************************
445 * CALCULATE INTERACTIONS *
446 **************************/
448 r10 = _mm_mul_pd(rsq10,rinv10);
450 /* Compute parameters for interactions between i and j atoms */
451 qq10 = _mm_mul_pd(iq1,jq0);
453 /* EWALD ELECTROSTATICS */
455 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
456 ewrt = _mm_mul_pd(r10,ewtabscale);
457 ewitab = _mm_cvttpd_epi32(ewrt);
459 eweps = _mm_frcz_pd(ewrt);
461 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
463 twoeweps = _mm_add_pd(eweps,eweps);
464 ewitab = _mm_slli_epi32(ewitab,2);
465 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
466 ewtabD = _mm_setzero_pd();
467 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
468 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
469 ewtabFn = _mm_setzero_pd();
470 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
471 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
472 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
473 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
474 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
476 /* Update potential sum for this i atom from the interaction with this j atom. */
477 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
478 velecsum = _mm_add_pd(velecsum,velec);
482 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
484 /* Update vectorial force */
485 fix1 = _mm_macc_pd(dx10,fscal,fix1);
486 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
487 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
489 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
490 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
491 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
493 /**************************
494 * CALCULATE INTERACTIONS *
495 **************************/
497 r20 = _mm_mul_pd(rsq20,rinv20);
499 /* Compute parameters for interactions between i and j atoms */
500 qq20 = _mm_mul_pd(iq2,jq0);
502 /* EWALD ELECTROSTATICS */
504 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
505 ewrt = _mm_mul_pd(r20,ewtabscale);
506 ewitab = _mm_cvttpd_epi32(ewrt);
508 eweps = _mm_frcz_pd(ewrt);
510 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
512 twoeweps = _mm_add_pd(eweps,eweps);
513 ewitab = _mm_slli_epi32(ewitab,2);
514 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
515 ewtabD = _mm_setzero_pd();
516 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
517 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
518 ewtabFn = _mm_setzero_pd();
519 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
520 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
521 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
522 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
523 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
525 /* Update potential sum for this i atom from the interaction with this j atom. */
526 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
527 velecsum = _mm_add_pd(velecsum,velec);
531 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
533 /* Update vectorial force */
534 fix2 = _mm_macc_pd(dx20,fscal,fix2);
535 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
536 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
538 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
539 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
540 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
542 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
544 /* Inner loop uses 135 flops */
547 /* End of innermost loop */
549 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
550 f+i_coord_offset,fshift+i_shift_offset);
553 /* Update potential energies */
554 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
556 /* Increment number of inner iterations */
557 inneriter += j_index_end - j_index_start;
559 /* Outer loop uses 19 flops */
562 /* Increment number of outer iterations */
565 /* Update outer/inner flops */
567 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*135);
570 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_128_fma_double
571 * Electrostatics interaction: Ewald
572 * VdW interaction: None
573 * Geometry: Water3-Particle
574 * Calculate force/pot: Force
577 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_128_fma_double
578 (t_nblist * gmx_restrict nlist,
579 rvec * gmx_restrict xx,
580 rvec * gmx_restrict ff,
581 t_forcerec * gmx_restrict fr,
582 t_mdatoms * gmx_restrict mdatoms,
583 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
584 t_nrnb * gmx_restrict nrnb)
586 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
587 * just 0 for non-waters.
588 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
589 * jnr indices corresponding to data put in the four positions in the SIMD register.
591 int i_shift_offset,i_coord_offset,outeriter,inneriter;
592 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
594 int j_coord_offsetA,j_coord_offsetB;
595 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
597 real *shiftvec,*fshift,*x,*f;
598 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
600 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
602 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
604 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
605 int vdwjidx0A,vdwjidx0B;
606 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
607 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
608 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
609 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
610 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
613 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
615 __m128d dummy_mask,cutoff_mask;
616 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
617 __m128d one = _mm_set1_pd(1.0);
618 __m128d two = _mm_set1_pd(2.0);
624 jindex = nlist->jindex;
626 shiftidx = nlist->shift;
628 shiftvec = fr->shift_vec[0];
629 fshift = fr->fshift[0];
630 facel = _mm_set1_pd(fr->epsfac);
631 charge = mdatoms->chargeA;
633 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
634 ewtab = fr->ic->tabq_coul_F;
635 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
636 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
638 /* Setup water-specific parameters */
639 inr = nlist->iinr[0];
640 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
641 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
642 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
644 /* Avoid stupid compiler warnings */
652 /* Start outer loop over neighborlists */
653 for(iidx=0; iidx<nri; iidx++)
655 /* Load shift vector for this list */
656 i_shift_offset = DIM*shiftidx[iidx];
658 /* Load limits for loop over neighbors */
659 j_index_start = jindex[iidx];
660 j_index_end = jindex[iidx+1];
662 /* Get outer coordinate index */
664 i_coord_offset = DIM*inr;
666 /* Load i particle coords and add shift vector */
667 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
668 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
670 fix0 = _mm_setzero_pd();
671 fiy0 = _mm_setzero_pd();
672 fiz0 = _mm_setzero_pd();
673 fix1 = _mm_setzero_pd();
674 fiy1 = _mm_setzero_pd();
675 fiz1 = _mm_setzero_pd();
676 fix2 = _mm_setzero_pd();
677 fiy2 = _mm_setzero_pd();
678 fiz2 = _mm_setzero_pd();
680 /* Start inner kernel loop */
681 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
684 /* Get j neighbor index, and coordinate index */
687 j_coord_offsetA = DIM*jnrA;
688 j_coord_offsetB = DIM*jnrB;
690 /* load j atom coordinates */
691 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
694 /* Calculate displacement vector */
695 dx00 = _mm_sub_pd(ix0,jx0);
696 dy00 = _mm_sub_pd(iy0,jy0);
697 dz00 = _mm_sub_pd(iz0,jz0);
698 dx10 = _mm_sub_pd(ix1,jx0);
699 dy10 = _mm_sub_pd(iy1,jy0);
700 dz10 = _mm_sub_pd(iz1,jz0);
701 dx20 = _mm_sub_pd(ix2,jx0);
702 dy20 = _mm_sub_pd(iy2,jy0);
703 dz20 = _mm_sub_pd(iz2,jz0);
705 /* Calculate squared distance and things based on it */
706 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
707 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
708 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
710 rinv00 = gmx_mm_invsqrt_pd(rsq00);
711 rinv10 = gmx_mm_invsqrt_pd(rsq10);
712 rinv20 = gmx_mm_invsqrt_pd(rsq20);
714 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
715 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
716 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
718 /* Load parameters for j particles */
719 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
721 fjx0 = _mm_setzero_pd();
722 fjy0 = _mm_setzero_pd();
723 fjz0 = _mm_setzero_pd();
725 /**************************
726 * CALCULATE INTERACTIONS *
727 **************************/
729 r00 = _mm_mul_pd(rsq00,rinv00);
731 /* Compute parameters for interactions between i and j atoms */
732 qq00 = _mm_mul_pd(iq0,jq0);
734 /* EWALD ELECTROSTATICS */
736 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
737 ewrt = _mm_mul_pd(r00,ewtabscale);
738 ewitab = _mm_cvttpd_epi32(ewrt);
740 eweps = _mm_frcz_pd(ewrt);
742 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
744 twoeweps = _mm_add_pd(eweps,eweps);
745 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
747 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
748 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
752 /* Update vectorial force */
753 fix0 = _mm_macc_pd(dx00,fscal,fix0);
754 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
755 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
757 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
758 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
759 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
761 /**************************
762 * CALCULATE INTERACTIONS *
763 **************************/
765 r10 = _mm_mul_pd(rsq10,rinv10);
767 /* Compute parameters for interactions between i and j atoms */
768 qq10 = _mm_mul_pd(iq1,jq0);
770 /* EWALD ELECTROSTATICS */
772 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
773 ewrt = _mm_mul_pd(r10,ewtabscale);
774 ewitab = _mm_cvttpd_epi32(ewrt);
776 eweps = _mm_frcz_pd(ewrt);
778 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
780 twoeweps = _mm_add_pd(eweps,eweps);
781 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
783 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
784 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
788 /* Update vectorial force */
789 fix1 = _mm_macc_pd(dx10,fscal,fix1);
790 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
791 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
793 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
794 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
795 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
797 /**************************
798 * CALCULATE INTERACTIONS *
799 **************************/
801 r20 = _mm_mul_pd(rsq20,rinv20);
803 /* Compute parameters for interactions between i and j atoms */
804 qq20 = _mm_mul_pd(iq2,jq0);
806 /* EWALD ELECTROSTATICS */
808 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
809 ewrt = _mm_mul_pd(r20,ewtabscale);
810 ewitab = _mm_cvttpd_epi32(ewrt);
812 eweps = _mm_frcz_pd(ewrt);
814 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
816 twoeweps = _mm_add_pd(eweps,eweps);
817 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
819 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
820 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
824 /* Update vectorial force */
825 fix2 = _mm_macc_pd(dx20,fscal,fix2);
826 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
827 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
829 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
830 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
831 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
833 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
835 /* Inner loop uses 120 flops */
842 j_coord_offsetA = DIM*jnrA;
844 /* load j atom coordinates */
845 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
848 /* Calculate displacement vector */
849 dx00 = _mm_sub_pd(ix0,jx0);
850 dy00 = _mm_sub_pd(iy0,jy0);
851 dz00 = _mm_sub_pd(iz0,jz0);
852 dx10 = _mm_sub_pd(ix1,jx0);
853 dy10 = _mm_sub_pd(iy1,jy0);
854 dz10 = _mm_sub_pd(iz1,jz0);
855 dx20 = _mm_sub_pd(ix2,jx0);
856 dy20 = _mm_sub_pd(iy2,jy0);
857 dz20 = _mm_sub_pd(iz2,jz0);
859 /* Calculate squared distance and things based on it */
860 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
861 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
862 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
864 rinv00 = gmx_mm_invsqrt_pd(rsq00);
865 rinv10 = gmx_mm_invsqrt_pd(rsq10);
866 rinv20 = gmx_mm_invsqrt_pd(rsq20);
868 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
869 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
870 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
872 /* Load parameters for j particles */
873 jq0 = _mm_load_sd(charge+jnrA+0);
875 fjx0 = _mm_setzero_pd();
876 fjy0 = _mm_setzero_pd();
877 fjz0 = _mm_setzero_pd();
879 /**************************
880 * CALCULATE INTERACTIONS *
881 **************************/
883 r00 = _mm_mul_pd(rsq00,rinv00);
885 /* Compute parameters for interactions between i and j atoms */
886 qq00 = _mm_mul_pd(iq0,jq0);
888 /* EWALD ELECTROSTATICS */
890 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
891 ewrt = _mm_mul_pd(r00,ewtabscale);
892 ewitab = _mm_cvttpd_epi32(ewrt);
894 eweps = _mm_frcz_pd(ewrt);
896 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
898 twoeweps = _mm_add_pd(eweps,eweps);
899 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
900 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
901 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
905 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
907 /* Update vectorial force */
908 fix0 = _mm_macc_pd(dx00,fscal,fix0);
909 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
910 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
912 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
913 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
914 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
916 /**************************
917 * CALCULATE INTERACTIONS *
918 **************************/
920 r10 = _mm_mul_pd(rsq10,rinv10);
922 /* Compute parameters for interactions between i and j atoms */
923 qq10 = _mm_mul_pd(iq1,jq0);
925 /* EWALD ELECTROSTATICS */
927 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
928 ewrt = _mm_mul_pd(r10,ewtabscale);
929 ewitab = _mm_cvttpd_epi32(ewrt);
931 eweps = _mm_frcz_pd(ewrt);
933 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
935 twoeweps = _mm_add_pd(eweps,eweps);
936 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
937 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
938 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
942 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
944 /* Update vectorial force */
945 fix1 = _mm_macc_pd(dx10,fscal,fix1);
946 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
947 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
949 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
950 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
951 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
953 /**************************
954 * CALCULATE INTERACTIONS *
955 **************************/
957 r20 = _mm_mul_pd(rsq20,rinv20);
959 /* Compute parameters for interactions between i and j atoms */
960 qq20 = _mm_mul_pd(iq2,jq0);
962 /* EWALD ELECTROSTATICS */
964 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
965 ewrt = _mm_mul_pd(r20,ewtabscale);
966 ewitab = _mm_cvttpd_epi32(ewrt);
968 eweps = _mm_frcz_pd(ewrt);
970 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
972 twoeweps = _mm_add_pd(eweps,eweps);
973 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
974 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
975 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
979 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
981 /* Update vectorial force */
982 fix2 = _mm_macc_pd(dx20,fscal,fix2);
983 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
984 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
986 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
987 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
988 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
990 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
992 /* Inner loop uses 120 flops */
995 /* End of innermost loop */
997 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
998 f+i_coord_offset,fshift+i_shift_offset);
1000 /* Increment number of inner iterations */
1001 inneriter += j_index_end - j_index_start;
1003 /* Outer loop uses 18 flops */
1006 /* Increment number of outer iterations */
1009 /* Update outer/inner flops */
1011 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*120);