?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Fix segmentation fault in minimize
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_128_fma_double / nb_kernel_ElecEw_VdwNone_GeomW3P1_avx_128_fma_double.cpp
blobee7bc60af7528339f82970674f19dbfc07a58ab9
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015,2017,2018, 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.
8 *
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.
13 *
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.
18 *
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.
23 *
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.
31 *
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.
34 */
35 /*
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
37 */
38 #include "gmxpre.h"
39
40 #include "config.h"
41
42 #include <math.h>
43
44 #include "../nb_kernel.h"
45 #include "gromacs/gmxlib/nrnb.h"
46
47 #include "kernelutil_x86_avx_128_fma_double.h"
48
49 /*
50 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_avx_128_fma_double
51 * Electrostatics interaction: Ewald
52 * VdW interaction: None
53 * Geometry: Water3-Particle
54 * Calculate force/pot: PotentialAndForce
55 */
56 void
57 nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_avx_128_fma_double
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
65 {
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
70 */
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
73 int jnrA,jnrB;
74 int j_coord_offsetA,j_coord_offsetB;
75 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
76 real rcutoff_scalar;
77 real *shiftvec,*fshift,*x,*f;
78 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
79 int vdwioffset0;
80 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
81 int vdwioffset1;
82 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
83 int vdwioffset2;
84 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
85 int vdwjidx0A,vdwjidx0B;
86 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
87 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
88 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
89 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
90 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
91 real *charge;
92 __m128i ewitab;
93 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
94 real *ewtab;
95 __m128d dummy_mask,cutoff_mask;
96 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
97 __m128d one = _mm_set1_pd(1.0);
98 __m128d two = _mm_set1_pd(2.0);
99 x = xx[0];
100 f = ff[0];
101
102 nri = nlist->nri;
103 iinr = nlist->iinr;
104 jindex = nlist->jindex;
105 jjnr = nlist->jjnr;
106 shiftidx = nlist->shift;
107 gid = nlist->gid;
108 shiftvec = fr->shift_vec[0];
109 fshift = fr->fshift[0];
110 facel = _mm_set1_pd(fr->ic->epsfac);
111 charge = mdatoms->chargeA;
112
113 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
114 ewtab = fr->ic->tabq_coul_FDV0;
115 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
116 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
117
118 /* Setup water-specific parameters */
119 inr = nlist->iinr[0];
120 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
121 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
122 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
123
124 /* Avoid stupid compiler warnings */
125 jnrA = jnrB = 0;
126 j_coord_offsetA = 0;
127 j_coord_offsetB = 0;
128
129 outeriter = 0;
130 inneriter = 0;
131
132 /* Start outer loop over neighborlists */
133 for(iidx=0; iidx<nri; iidx++)
134 {
135 /* Load shift vector for this list */
136 i_shift_offset = DIM*shiftidx[iidx];
137
138 /* Load limits for loop over neighbors */
139 j_index_start = jindex[iidx];
140 j_index_end = jindex[iidx+1];
141
142 /* Get outer coordinate index */
143 inr = iinr[iidx];
144 i_coord_offset = DIM*inr;
145
146 /* Load i particle coords and add shift vector */
147 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
148 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
149
150 fix0 = _mm_setzero_pd();
151 fiy0 = _mm_setzero_pd();
152 fiz0 = _mm_setzero_pd();
153 fix1 = _mm_setzero_pd();
154 fiy1 = _mm_setzero_pd();
155 fiz1 = _mm_setzero_pd();
156 fix2 = _mm_setzero_pd();
157 fiy2 = _mm_setzero_pd();
158 fiz2 = _mm_setzero_pd();
159
160 /* Reset potential sums */
161 velecsum = _mm_setzero_pd();
162
163 /* Start inner kernel loop */
164 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
165 {
166
167 /* Get j neighbor index, and coordinate index */
168 jnrA = jjnr[jidx];
169 jnrB = jjnr[jidx+1];
170 j_coord_offsetA = DIM*jnrA;
171 j_coord_offsetB = DIM*jnrB;
172
173 /* load j atom coordinates */
174 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
175 &jx0,&jy0,&jz0);
176
177 /* Calculate displacement vector */
178 dx00 = _mm_sub_pd(ix0,jx0);
179 dy00 = _mm_sub_pd(iy0,jy0);
180 dz00 = _mm_sub_pd(iz0,jz0);
181 dx10 = _mm_sub_pd(ix1,jx0);
182 dy10 = _mm_sub_pd(iy1,jy0);
183 dz10 = _mm_sub_pd(iz1,jz0);
184 dx20 = _mm_sub_pd(ix2,jx0);
185 dy20 = _mm_sub_pd(iy2,jy0);
186 dz20 = _mm_sub_pd(iz2,jz0);
187
188 /* Calculate squared distance and things based on it */
189 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
190 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
191 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
192
193 rinv00 = avx128fma_invsqrt_d(rsq00);
194 rinv10 = avx128fma_invsqrt_d(rsq10);
195 rinv20 = avx128fma_invsqrt_d(rsq20);
196
197 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
198 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
199 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
200
201 /* Load parameters for j particles */
202 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
203
204 fjx0 = _mm_setzero_pd();
205 fjy0 = _mm_setzero_pd();
206 fjz0 = _mm_setzero_pd();
207
208 /**************************
209 * CALCULATE INTERACTIONS *
210 **************************/
211
212 r00 = _mm_mul_pd(rsq00,rinv00);
213
214 /* Compute parameters for interactions between i and j atoms */
215 qq00 = _mm_mul_pd(iq0,jq0);
216
217 /* EWALD ELECTROSTATICS */
218
219 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
220 ewrt = _mm_mul_pd(r00,ewtabscale);
221 ewitab = _mm_cvttpd_epi32(ewrt);
222 #ifdef __XOP__
223 eweps = _mm_frcz_pd(ewrt);
224 #else
225 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
226 #endif
227 twoeweps = _mm_add_pd(eweps,eweps);
228 ewitab = _mm_slli_epi32(ewitab,2);
229 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
230 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
231 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
232 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
233 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
234 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
235 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
236 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
237 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
238 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
239
240 /* Update potential sum for this i atom from the interaction with this j atom. */
241 velecsum = _mm_add_pd(velecsum,velec);
242
243 fscal = felec;
244
245 /* Update vectorial force */
246 fix0 = _mm_macc_pd(dx00,fscal,fix0);
247 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
248 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
249
250 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
251 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
252 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
253
254 /**************************
255 * CALCULATE INTERACTIONS *
256 **************************/
257
258 r10 = _mm_mul_pd(rsq10,rinv10);
259
260 /* Compute parameters for interactions between i and j atoms */
261 qq10 = _mm_mul_pd(iq1,jq0);
262
263 /* EWALD ELECTROSTATICS */
264
265 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
266 ewrt = _mm_mul_pd(r10,ewtabscale);
267 ewitab = _mm_cvttpd_epi32(ewrt);
268 #ifdef __XOP__
269 eweps = _mm_frcz_pd(ewrt);
270 #else
271 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
272 #endif
273 twoeweps = _mm_add_pd(eweps,eweps);
274 ewitab = _mm_slli_epi32(ewitab,2);
275 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
276 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
277 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
278 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
279 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
280 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
281 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
282 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
283 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
284 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
285
286 /* Update potential sum for this i atom from the interaction with this j atom. */
287 velecsum = _mm_add_pd(velecsum,velec);
288
289 fscal = felec;
290
291 /* Update vectorial force */
292 fix1 = _mm_macc_pd(dx10,fscal,fix1);
293 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
294 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
295
296 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
297 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
298 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
299
300 /**************************
301 * CALCULATE INTERACTIONS *
302 **************************/
303
304 r20 = _mm_mul_pd(rsq20,rinv20);
305
306 /* Compute parameters for interactions between i and j atoms */
307 qq20 = _mm_mul_pd(iq2,jq0);
308
309 /* EWALD ELECTROSTATICS */
310
311 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
312 ewrt = _mm_mul_pd(r20,ewtabscale);
313 ewitab = _mm_cvttpd_epi32(ewrt);
314 #ifdef __XOP__
315 eweps = _mm_frcz_pd(ewrt);
316 #else
317 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
318 #endif
319 twoeweps = _mm_add_pd(eweps,eweps);
320 ewitab = _mm_slli_epi32(ewitab,2);
321 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
322 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
323 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
324 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
325 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
326 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
327 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
328 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
329 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
330 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
331
332 /* Update potential sum for this i atom from the interaction with this j atom. */
333 velecsum = _mm_add_pd(velecsum,velec);
334
335 fscal = felec;
336
337 /* Update vectorial force */
338 fix2 = _mm_macc_pd(dx20,fscal,fix2);
339 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
340 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
341
342 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
343 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
344 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
345
346 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
347
348 /* Inner loop uses 135 flops */
349 }
350
351 if(jidx<j_index_end)
352 {
353
354 jnrA = jjnr[jidx];
355 j_coord_offsetA = DIM*jnrA;
356
357 /* load j atom coordinates */
358 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
359 &jx0,&jy0,&jz0);
360
361 /* Calculate displacement vector */
362 dx00 = _mm_sub_pd(ix0,jx0);
363 dy00 = _mm_sub_pd(iy0,jy0);
364 dz00 = _mm_sub_pd(iz0,jz0);
365 dx10 = _mm_sub_pd(ix1,jx0);
366 dy10 = _mm_sub_pd(iy1,jy0);
367 dz10 = _mm_sub_pd(iz1,jz0);
368 dx20 = _mm_sub_pd(ix2,jx0);
369 dy20 = _mm_sub_pd(iy2,jy0);
370 dz20 = _mm_sub_pd(iz2,jz0);
371
372 /* Calculate squared distance and things based on it */
373 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
374 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
375 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
376
377 rinv00 = avx128fma_invsqrt_d(rsq00);
378 rinv10 = avx128fma_invsqrt_d(rsq10);
379 rinv20 = avx128fma_invsqrt_d(rsq20);
380
381 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
382 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
383 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
384
385 /* Load parameters for j particles */
386 jq0 = _mm_load_sd(charge+jnrA+0);
387
388 fjx0 = _mm_setzero_pd();
389 fjy0 = _mm_setzero_pd();
390 fjz0 = _mm_setzero_pd();
391
392 /**************************
393 * CALCULATE INTERACTIONS *
394 **************************/
395
396 r00 = _mm_mul_pd(rsq00,rinv00);
397
398 /* Compute parameters for interactions between i and j atoms */
399 qq00 = _mm_mul_pd(iq0,jq0);
400
401 /* EWALD ELECTROSTATICS */
402
403 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
404 ewrt = _mm_mul_pd(r00,ewtabscale);
405 ewitab = _mm_cvttpd_epi32(ewrt);
406 #ifdef __XOP__
407 eweps = _mm_frcz_pd(ewrt);
408 #else
409 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
410 #endif
411 twoeweps = _mm_add_pd(eweps,eweps);
412 ewitab = _mm_slli_epi32(ewitab,2);
413 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
414 ewtabD = _mm_setzero_pd();
415 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
416 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
417 ewtabFn = _mm_setzero_pd();
418 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
419 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
420 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
421 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
422 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
423
424 /* Update potential sum for this i atom from the interaction with this j atom. */
425 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
426 velecsum = _mm_add_pd(velecsum,velec);
427
428 fscal = felec;
429
430 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
431
432 /* Update vectorial force */
433 fix0 = _mm_macc_pd(dx00,fscal,fix0);
434 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
435 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
436
437 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
438 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
439 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
440
441 /**************************
442 * CALCULATE INTERACTIONS *
443 **************************/
444
445 r10 = _mm_mul_pd(rsq10,rinv10);
446
447 /* Compute parameters for interactions between i and j atoms */
448 qq10 = _mm_mul_pd(iq1,jq0);
449
450 /* EWALD ELECTROSTATICS */
451
452 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
453 ewrt = _mm_mul_pd(r10,ewtabscale);
454 ewitab = _mm_cvttpd_epi32(ewrt);
455 #ifdef __XOP__
456 eweps = _mm_frcz_pd(ewrt);
457 #else
458 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
459 #endif
460 twoeweps = _mm_add_pd(eweps,eweps);
461 ewitab = _mm_slli_epi32(ewitab,2);
462 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
463 ewtabD = _mm_setzero_pd();
464 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
465 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
466 ewtabFn = _mm_setzero_pd();
467 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
468 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
469 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
470 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
471 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
472
473 /* Update potential sum for this i atom from the interaction with this j atom. */
474 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
475 velecsum = _mm_add_pd(velecsum,velec);
476
477 fscal = felec;
478
479 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
480
481 /* Update vectorial force */
482 fix1 = _mm_macc_pd(dx10,fscal,fix1);
483 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
484 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
485
486 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
487 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
488 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
489
490 /**************************
491 * CALCULATE INTERACTIONS *
492 **************************/
493
494 r20 = _mm_mul_pd(rsq20,rinv20);
495
496 /* Compute parameters for interactions between i and j atoms */
497 qq20 = _mm_mul_pd(iq2,jq0);
498
499 /* EWALD ELECTROSTATICS */
500
501 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
502 ewrt = _mm_mul_pd(r20,ewtabscale);
503 ewitab = _mm_cvttpd_epi32(ewrt);
504 #ifdef __XOP__
505 eweps = _mm_frcz_pd(ewrt);
506 #else
507 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
508 #endif
509 twoeweps = _mm_add_pd(eweps,eweps);
510 ewitab = _mm_slli_epi32(ewitab,2);
511 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
512 ewtabD = _mm_setzero_pd();
513 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
514 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
515 ewtabFn = _mm_setzero_pd();
516 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
517 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
518 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
519 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
520 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
521
522 /* Update potential sum for this i atom from the interaction with this j atom. */
523 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
524 velecsum = _mm_add_pd(velecsum,velec);
525
526 fscal = felec;
527
528 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
529
530 /* Update vectorial force */
531 fix2 = _mm_macc_pd(dx20,fscal,fix2);
532 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
533 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
534
535 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
536 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
537 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
538
539 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
540
541 /* Inner loop uses 135 flops */
542 }
543
544 /* End of innermost loop */
545
546 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
547 f+i_coord_offset,fshift+i_shift_offset);
548
549 ggid = gid[iidx];
550 /* Update potential energies */
551 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
552
553 /* Increment number of inner iterations */
554 inneriter += j_index_end - j_index_start;
555
556 /* Outer loop uses 19 flops */
557 }
558
559 /* Increment number of outer iterations */
560 outeriter += nri;
561
562 /* Update outer/inner flops */
563
564 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*135);
565 }
566 /*
567 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_128_fma_double
568 * Electrostatics interaction: Ewald
569 * VdW interaction: None
570 * Geometry: Water3-Particle
571 * Calculate force/pot: Force
572 */
573 void
574 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_128_fma_double
575 (t_nblist * gmx_restrict nlist,
576 rvec * gmx_restrict xx,
577 rvec * gmx_restrict ff,
578 struct t_forcerec * gmx_restrict fr,
579 t_mdatoms * gmx_restrict mdatoms,
580 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
581 t_nrnb * gmx_restrict nrnb)
582 {
583 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
584 * just 0 for non-waters.
585 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
586 * jnr indices corresponding to data put in the four positions in the SIMD register.
587 */
588 int i_shift_offset,i_coord_offset,outeriter,inneriter;
589 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
590 int jnrA,jnrB;
591 int j_coord_offsetA,j_coord_offsetB;
592 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
593 real rcutoff_scalar;
594 real *shiftvec,*fshift,*x,*f;
595 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
596 int vdwioffset0;
597 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
598 int vdwioffset1;
599 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
600 int vdwioffset2;
601 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
602 int vdwjidx0A,vdwjidx0B;
603 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
604 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
605 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
606 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
607 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
608 real *charge;
609 __m128i ewitab;
610 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
611 real *ewtab;
612 __m128d dummy_mask,cutoff_mask;
613 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
614 __m128d one = _mm_set1_pd(1.0);
615 __m128d two = _mm_set1_pd(2.0);
616 x = xx[0];
617 f = ff[0];
618
619 nri = nlist->nri;
620 iinr = nlist->iinr;
621 jindex = nlist->jindex;
622 jjnr = nlist->jjnr;
623 shiftidx = nlist->shift;
624 gid = nlist->gid;
625 shiftvec = fr->shift_vec[0];
626 fshift = fr->fshift[0];
627 facel = _mm_set1_pd(fr->ic->epsfac);
628 charge = mdatoms->chargeA;
629
630 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
631 ewtab = fr->ic->tabq_coul_F;
632 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
633 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
634
635 /* Setup water-specific parameters */
636 inr = nlist->iinr[0];
637 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
638 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
639 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
640
641 /* Avoid stupid compiler warnings */
642 jnrA = jnrB = 0;
643 j_coord_offsetA = 0;
644 j_coord_offsetB = 0;
645
646 outeriter = 0;
647 inneriter = 0;
648
649 /* Start outer loop over neighborlists */
650 for(iidx=0; iidx<nri; iidx++)
651 {
652 /* Load shift vector for this list */
653 i_shift_offset = DIM*shiftidx[iidx];
654
655 /* Load limits for loop over neighbors */
656 j_index_start = jindex[iidx];
657 j_index_end = jindex[iidx+1];
658
659 /* Get outer coordinate index */
660 inr = iinr[iidx];
661 i_coord_offset = DIM*inr;
662
663 /* Load i particle coords and add shift vector */
664 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
665 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
666
667 fix0 = _mm_setzero_pd();
668 fiy0 = _mm_setzero_pd();
669 fiz0 = _mm_setzero_pd();
670 fix1 = _mm_setzero_pd();
671 fiy1 = _mm_setzero_pd();
672 fiz1 = _mm_setzero_pd();
673 fix2 = _mm_setzero_pd();
674 fiy2 = _mm_setzero_pd();
675 fiz2 = _mm_setzero_pd();
676
677 /* Start inner kernel loop */
678 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
679 {
680
681 /* Get j neighbor index, and coordinate index */
682 jnrA = jjnr[jidx];
683 jnrB = jjnr[jidx+1];
684 j_coord_offsetA = DIM*jnrA;
685 j_coord_offsetB = DIM*jnrB;
686
687 /* load j atom coordinates */
688 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
689 &jx0,&jy0,&jz0);
690
691 /* Calculate displacement vector */
692 dx00 = _mm_sub_pd(ix0,jx0);
693 dy00 = _mm_sub_pd(iy0,jy0);
694 dz00 = _mm_sub_pd(iz0,jz0);
695 dx10 = _mm_sub_pd(ix1,jx0);
696 dy10 = _mm_sub_pd(iy1,jy0);
697 dz10 = _mm_sub_pd(iz1,jz0);
698 dx20 = _mm_sub_pd(ix2,jx0);
699 dy20 = _mm_sub_pd(iy2,jy0);
700 dz20 = _mm_sub_pd(iz2,jz0);
701
702 /* Calculate squared distance and things based on it */
703 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
704 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
705 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
706
707 rinv00 = avx128fma_invsqrt_d(rsq00);
708 rinv10 = avx128fma_invsqrt_d(rsq10);
709 rinv20 = avx128fma_invsqrt_d(rsq20);
710
711 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
712 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
713 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
714
715 /* Load parameters for j particles */
716 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
717
718 fjx0 = _mm_setzero_pd();
719 fjy0 = _mm_setzero_pd();
720 fjz0 = _mm_setzero_pd();
721
722 /**************************
723 * CALCULATE INTERACTIONS *
724 **************************/
725
726 r00 = _mm_mul_pd(rsq00,rinv00);
727
728 /* Compute parameters for interactions between i and j atoms */
729 qq00 = _mm_mul_pd(iq0,jq0);
730
731 /* EWALD ELECTROSTATICS */
732
733 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
734 ewrt = _mm_mul_pd(r00,ewtabscale);
735 ewitab = _mm_cvttpd_epi32(ewrt);
736 #ifdef __XOP__
737 eweps = _mm_frcz_pd(ewrt);
738 #else
739 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
740 #endif
741 twoeweps = _mm_add_pd(eweps,eweps);
742 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
743 &ewtabF,&ewtabFn);
744 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
745 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
746
747 fscal = felec;
748
749 /* Update vectorial force */
750 fix0 = _mm_macc_pd(dx00,fscal,fix0);
751 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
752 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
753
754 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
755 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
756 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
757
758 /**************************
759 * CALCULATE INTERACTIONS *
760 **************************/
761
762 r10 = _mm_mul_pd(rsq10,rinv10);
763
764 /* Compute parameters for interactions between i and j atoms */
765 qq10 = _mm_mul_pd(iq1,jq0);
766
767 /* EWALD ELECTROSTATICS */
768
769 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
770 ewrt = _mm_mul_pd(r10,ewtabscale);
771 ewitab = _mm_cvttpd_epi32(ewrt);
772 #ifdef __XOP__
773 eweps = _mm_frcz_pd(ewrt);
774 #else
775 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
776 #endif
777 twoeweps = _mm_add_pd(eweps,eweps);
778 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
779 &ewtabF,&ewtabFn);
780 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
781 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
782
783 fscal = felec;
784
785 /* Update vectorial force */
786 fix1 = _mm_macc_pd(dx10,fscal,fix1);
787 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
788 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
789
790 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
791 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
792 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
793
794 /**************************
795 * CALCULATE INTERACTIONS *
796 **************************/
797
798 r20 = _mm_mul_pd(rsq20,rinv20);
799
800 /* Compute parameters for interactions between i and j atoms */
801 qq20 = _mm_mul_pd(iq2,jq0);
802
803 /* EWALD ELECTROSTATICS */
804
805 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
806 ewrt = _mm_mul_pd(r20,ewtabscale);
807 ewitab = _mm_cvttpd_epi32(ewrt);
808 #ifdef __XOP__
809 eweps = _mm_frcz_pd(ewrt);
810 #else
811 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
812 #endif
813 twoeweps = _mm_add_pd(eweps,eweps);
814 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
815 &ewtabF,&ewtabFn);
816 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
817 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
818
819 fscal = felec;
820
821 /* Update vectorial force */
822 fix2 = _mm_macc_pd(dx20,fscal,fix2);
823 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
824 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
825
826 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
827 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
828 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
829
830 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
831
832 /* Inner loop uses 120 flops */
833 }
834
835 if(jidx<j_index_end)
836 {
837
838 jnrA = jjnr[jidx];
839 j_coord_offsetA = DIM*jnrA;
840
841 /* load j atom coordinates */
842 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
843 &jx0,&jy0,&jz0);
844
845 /* Calculate displacement vector */
846 dx00 = _mm_sub_pd(ix0,jx0);
847 dy00 = _mm_sub_pd(iy0,jy0);
848 dz00 = _mm_sub_pd(iz0,jz0);
849 dx10 = _mm_sub_pd(ix1,jx0);
850 dy10 = _mm_sub_pd(iy1,jy0);
851 dz10 = _mm_sub_pd(iz1,jz0);
852 dx20 = _mm_sub_pd(ix2,jx0);
853 dy20 = _mm_sub_pd(iy2,jy0);
854 dz20 = _mm_sub_pd(iz2,jz0);
855
856 /* Calculate squared distance and things based on it */
857 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
858 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
859 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
860
861 rinv00 = avx128fma_invsqrt_d(rsq00);
862 rinv10 = avx128fma_invsqrt_d(rsq10);
863 rinv20 = avx128fma_invsqrt_d(rsq20);
864
865 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
866 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
867 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
868
869 /* Load parameters for j particles */
870 jq0 = _mm_load_sd(charge+jnrA+0);
871
872 fjx0 = _mm_setzero_pd();
873 fjy0 = _mm_setzero_pd();
874 fjz0 = _mm_setzero_pd();
875
876 /**************************
877 * CALCULATE INTERACTIONS *
878 **************************/
879
880 r00 = _mm_mul_pd(rsq00,rinv00);
881
882 /* Compute parameters for interactions between i and j atoms */
883 qq00 = _mm_mul_pd(iq0,jq0);
884
885 /* EWALD ELECTROSTATICS */
886
887 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
888 ewrt = _mm_mul_pd(r00,ewtabscale);
889 ewitab = _mm_cvttpd_epi32(ewrt);
890 #ifdef __XOP__
891 eweps = _mm_frcz_pd(ewrt);
892 #else
893 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
894 #endif
895 twoeweps = _mm_add_pd(eweps,eweps);
896 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
897 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
898 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
899
900 fscal = felec;
901
902 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
903
904 /* Update vectorial force */
905 fix0 = _mm_macc_pd(dx00,fscal,fix0);
906 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
907 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
908
909 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
910 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
911 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
912
913 /**************************
914 * CALCULATE INTERACTIONS *
915 **************************/
916
917 r10 = _mm_mul_pd(rsq10,rinv10);
918
919 /* Compute parameters for interactions between i and j atoms */
920 qq10 = _mm_mul_pd(iq1,jq0);
921
922 /* EWALD ELECTROSTATICS */
923
924 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
925 ewrt = _mm_mul_pd(r10,ewtabscale);
926 ewitab = _mm_cvttpd_epi32(ewrt);
927 #ifdef __XOP__
928 eweps = _mm_frcz_pd(ewrt);
929 #else
930 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
931 #endif
932 twoeweps = _mm_add_pd(eweps,eweps);
933 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
934 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
935 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
936
937 fscal = felec;
938
939 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
940
941 /* Update vectorial force */
942 fix1 = _mm_macc_pd(dx10,fscal,fix1);
943 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
944 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
945
946 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
947 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
948 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
949
950 /**************************
951 * CALCULATE INTERACTIONS *
952 **************************/
953
954 r20 = _mm_mul_pd(rsq20,rinv20);
955
956 /* Compute parameters for interactions between i and j atoms */
957 qq20 = _mm_mul_pd(iq2,jq0);
958
959 /* EWALD ELECTROSTATICS */
960
961 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
962 ewrt = _mm_mul_pd(r20,ewtabscale);
963 ewitab = _mm_cvttpd_epi32(ewrt);
964 #ifdef __XOP__
965 eweps = _mm_frcz_pd(ewrt);
966 #else
967 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
968 #endif
969 twoeweps = _mm_add_pd(eweps,eweps);
970 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
971 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
972 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
973
974 fscal = felec;
975
976 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
977
978 /* Update vectorial force */
979 fix2 = _mm_macc_pd(dx20,fscal,fix2);
980 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
981 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
982
983 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
984 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
985 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
986
987 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
988
989 /* Inner loop uses 120 flops */
990 }
991
992 /* End of innermost loop */
993
994 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
995 f+i_coord_offset,fshift+i_shift_offset);
996
997 /* Increment number of inner iterations */
998 inneriter += j_index_end - j_index_start;
999
1000 /* Outer loop uses 18 flops */
1001 }
1002
1003 /* Increment number of outer iterations */
1004 outeriter += nri;
1005
1006 /* Update outer/inner flops */
1007
1008 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*120);
1009 }
The ultimate molecular dynamics simulation package