?  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_ElecRF_VdwNone_GeomP1P1_avx_128_fma_double.cpp
blobd0e4fd88084b898ac713bfa70e7ba5f3d849f033
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_ElecRF_VdwNone_GeomP1P1_VF_avx_128_fma_double
51 * Electrostatics interaction: ReactionField
52 * VdW interaction: None
53 * Geometry: Particle-Particle
54 * Calculate force/pot: PotentialAndForce
55 */
56 void
57 nb_kernel_ElecRF_VdwNone_GeomP1P1_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 vdwjidx0A,vdwjidx0B;
82 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
83 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
84 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
85 real *charge;
86 __m128d dummy_mask,cutoff_mask;
87 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
88 __m128d one = _mm_set1_pd(1.0);
89 __m128d two = _mm_set1_pd(2.0);
90 x = xx[0];
91 f = ff[0];
92
93 nri = nlist->nri;
94 iinr = nlist->iinr;
95 jindex = nlist->jindex;
96 jjnr = nlist->jjnr;
97 shiftidx = nlist->shift;
98 gid = nlist->gid;
99 shiftvec = fr->shift_vec[0];
100 fshift = fr->fshift[0];
101 facel = _mm_set1_pd(fr->ic->epsfac);
102 charge = mdatoms->chargeA;
103 krf = _mm_set1_pd(fr->ic->k_rf);
104 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
105 crf = _mm_set1_pd(fr->ic->c_rf);
106
107 /* Avoid stupid compiler warnings */
108 jnrA = jnrB = 0;
109 j_coord_offsetA = 0;
110 j_coord_offsetB = 0;
111
112 outeriter = 0;
113 inneriter = 0;
114
115 /* Start outer loop over neighborlists */
116 for(iidx=0; iidx<nri; iidx++)
117 {
118 /* Load shift vector for this list */
119 i_shift_offset = DIM*shiftidx[iidx];
120
121 /* Load limits for loop over neighbors */
122 j_index_start = jindex[iidx];
123 j_index_end = jindex[iidx+1];
124
125 /* Get outer coordinate index */
126 inr = iinr[iidx];
127 i_coord_offset = DIM*inr;
128
129 /* Load i particle coords and add shift vector */
130 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
131
132 fix0 = _mm_setzero_pd();
133 fiy0 = _mm_setzero_pd();
134 fiz0 = _mm_setzero_pd();
135
136 /* Load parameters for i particles */
137 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
138
139 /* Reset potential sums */
140 velecsum = _mm_setzero_pd();
141
142 /* Start inner kernel loop */
143 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
144 {
145
146 /* Get j neighbor index, and coordinate index */
147 jnrA = jjnr[jidx];
148 jnrB = jjnr[jidx+1];
149 j_coord_offsetA = DIM*jnrA;
150 j_coord_offsetB = DIM*jnrB;
151
152 /* load j atom coordinates */
153 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
154 &jx0,&jy0,&jz0);
155
156 /* Calculate displacement vector */
157 dx00 = _mm_sub_pd(ix0,jx0);
158 dy00 = _mm_sub_pd(iy0,jy0);
159 dz00 = _mm_sub_pd(iz0,jz0);
160
161 /* Calculate squared distance and things based on it */
162 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
163
164 rinv00 = avx128fma_invsqrt_d(rsq00);
165
166 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
167
168 /* Load parameters for j particles */
169 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
170
171 /**************************
172 * CALCULATE INTERACTIONS *
173 **************************/
174
175 /* Compute parameters for interactions between i and j atoms */
176 qq00 = _mm_mul_pd(iq0,jq0);
177
178 /* REACTION-FIELD ELECTROSTATICS */
179 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
180 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
181
182 /* Update potential sum for this i atom from the interaction with this j atom. */
183 velecsum = _mm_add_pd(velecsum,velec);
184
185 fscal = felec;
186
187 /* Update vectorial force */
188 fix0 = _mm_macc_pd(dx00,fscal,fix0);
189 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
190 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
191
192 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
193 _mm_mul_pd(dx00,fscal),
194 _mm_mul_pd(dy00,fscal),
195 _mm_mul_pd(dz00,fscal));
196
197 /* Inner loop uses 35 flops */
198 }
199
200 if(jidx<j_index_end)
201 {
202
203 jnrA = jjnr[jidx];
204 j_coord_offsetA = DIM*jnrA;
205
206 /* load j atom coordinates */
207 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
208 &jx0,&jy0,&jz0);
209
210 /* Calculate displacement vector */
211 dx00 = _mm_sub_pd(ix0,jx0);
212 dy00 = _mm_sub_pd(iy0,jy0);
213 dz00 = _mm_sub_pd(iz0,jz0);
214
215 /* Calculate squared distance and things based on it */
216 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
217
218 rinv00 = avx128fma_invsqrt_d(rsq00);
219
220 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
221
222 /* Load parameters for j particles */
223 jq0 = _mm_load_sd(charge+jnrA+0);
224
225 /**************************
226 * CALCULATE INTERACTIONS *
227 **************************/
228
229 /* Compute parameters for interactions between i and j atoms */
230 qq00 = _mm_mul_pd(iq0,jq0);
231
232 /* REACTION-FIELD ELECTROSTATICS */
233 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
234 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
235
236 /* Update potential sum for this i atom from the interaction with this j atom. */
237 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
238 velecsum = _mm_add_pd(velecsum,velec);
239
240 fscal = felec;
241
242 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
243
244 /* Update vectorial force */
245 fix0 = _mm_macc_pd(dx00,fscal,fix0);
246 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
247 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
248
249 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
250 _mm_mul_pd(dx00,fscal),
251 _mm_mul_pd(dy00,fscal),
252 _mm_mul_pd(dz00,fscal));
253
254 /* Inner loop uses 35 flops */
255 }
256
257 /* End of innermost loop */
258
259 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
260 f+i_coord_offset,fshift+i_shift_offset);
261
262 ggid = gid[iidx];
263 /* Update potential energies */
264 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
265
266 /* Increment number of inner iterations */
267 inneriter += j_index_end - j_index_start;
268
269 /* Outer loop uses 8 flops */
270 }
271
272 /* Increment number of outer iterations */
273 outeriter += nri;
274
275 /* Update outer/inner flops */
276
277 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*35);
278 }
279 /*
280 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_128_fma_double
281 * Electrostatics interaction: ReactionField
282 * VdW interaction: None
283 * Geometry: Particle-Particle
284 * Calculate force/pot: Force
285 */
286 void
287 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_128_fma_double
288 (t_nblist * gmx_restrict nlist,
289 rvec * gmx_restrict xx,
290 rvec * gmx_restrict ff,
291 struct t_forcerec * gmx_restrict fr,
292 t_mdatoms * gmx_restrict mdatoms,
293 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
294 t_nrnb * gmx_restrict nrnb)
295 {
296 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
297 * just 0 for non-waters.
298 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
299 * jnr indices corresponding to data put in the four positions in the SIMD register.
300 */
301 int i_shift_offset,i_coord_offset,outeriter,inneriter;
302 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
303 int jnrA,jnrB;
304 int j_coord_offsetA,j_coord_offsetB;
305 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
306 real rcutoff_scalar;
307 real *shiftvec,*fshift,*x,*f;
308 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
309 int vdwioffset0;
310 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
311 int vdwjidx0A,vdwjidx0B;
312 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
313 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
314 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
315 real *charge;
316 __m128d dummy_mask,cutoff_mask;
317 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
318 __m128d one = _mm_set1_pd(1.0);
319 __m128d two = _mm_set1_pd(2.0);
320 x = xx[0];
321 f = ff[0];
322
323 nri = nlist->nri;
324 iinr = nlist->iinr;
325 jindex = nlist->jindex;
326 jjnr = nlist->jjnr;
327 shiftidx = nlist->shift;
328 gid = nlist->gid;
329 shiftvec = fr->shift_vec[0];
330 fshift = fr->fshift[0];
331 facel = _mm_set1_pd(fr->ic->epsfac);
332 charge = mdatoms->chargeA;
333 krf = _mm_set1_pd(fr->ic->k_rf);
334 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
335 crf = _mm_set1_pd(fr->ic->c_rf);
336
337 /* Avoid stupid compiler warnings */
338 jnrA = jnrB = 0;
339 j_coord_offsetA = 0;
340 j_coord_offsetB = 0;
341
342 outeriter = 0;
343 inneriter = 0;
344
345 /* Start outer loop over neighborlists */
346 for(iidx=0; iidx<nri; iidx++)
347 {
348 /* Load shift vector for this list */
349 i_shift_offset = DIM*shiftidx[iidx];
350
351 /* Load limits for loop over neighbors */
352 j_index_start = jindex[iidx];
353 j_index_end = jindex[iidx+1];
354
355 /* Get outer coordinate index */
356 inr = iinr[iidx];
357 i_coord_offset = DIM*inr;
358
359 /* Load i particle coords and add shift vector */
360 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
361
362 fix0 = _mm_setzero_pd();
363 fiy0 = _mm_setzero_pd();
364 fiz0 = _mm_setzero_pd();
365
366 /* Load parameters for i particles */
367 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
368
369 /* Start inner kernel loop */
370 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
371 {
372
373 /* Get j neighbor index, and coordinate index */
374 jnrA = jjnr[jidx];
375 jnrB = jjnr[jidx+1];
376 j_coord_offsetA = DIM*jnrA;
377 j_coord_offsetB = DIM*jnrB;
378
379 /* load j atom coordinates */
380 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
381 &jx0,&jy0,&jz0);
382
383 /* Calculate displacement vector */
384 dx00 = _mm_sub_pd(ix0,jx0);
385 dy00 = _mm_sub_pd(iy0,jy0);
386 dz00 = _mm_sub_pd(iz0,jz0);
387
388 /* Calculate squared distance and things based on it */
389 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
390
391 rinv00 = avx128fma_invsqrt_d(rsq00);
392
393 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
394
395 /* Load parameters for j particles */
396 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
397
398 /**************************
399 * CALCULATE INTERACTIONS *
400 **************************/
401
402 /* Compute parameters for interactions between i and j atoms */
403 qq00 = _mm_mul_pd(iq0,jq0);
404
405 /* REACTION-FIELD ELECTROSTATICS */
406 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
407
408 fscal = felec;
409
410 /* Update vectorial force */
411 fix0 = _mm_macc_pd(dx00,fscal,fix0);
412 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
413 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
414
415 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
416 _mm_mul_pd(dx00,fscal),
417 _mm_mul_pd(dy00,fscal),
418 _mm_mul_pd(dz00,fscal));
419
420 /* Inner loop uses 30 flops */
421 }
422
423 if(jidx<j_index_end)
424 {
425
426 jnrA = jjnr[jidx];
427 j_coord_offsetA = DIM*jnrA;
428
429 /* load j atom coordinates */
430 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
431 &jx0,&jy0,&jz0);
432
433 /* Calculate displacement vector */
434 dx00 = _mm_sub_pd(ix0,jx0);
435 dy00 = _mm_sub_pd(iy0,jy0);
436 dz00 = _mm_sub_pd(iz0,jz0);
437
438 /* Calculate squared distance and things based on it */
439 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
440
441 rinv00 = avx128fma_invsqrt_d(rsq00);
442
443 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
444
445 /* Load parameters for j particles */
446 jq0 = _mm_load_sd(charge+jnrA+0);
447
448 /**************************
449 * CALCULATE INTERACTIONS *
450 **************************/
451
452 /* Compute parameters for interactions between i and j atoms */
453 qq00 = _mm_mul_pd(iq0,jq0);
454
455 /* REACTION-FIELD ELECTROSTATICS */
456 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
457
458 fscal = felec;
459
460 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
461
462 /* Update vectorial force */
463 fix0 = _mm_macc_pd(dx00,fscal,fix0);
464 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
465 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
466
467 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
468 _mm_mul_pd(dx00,fscal),
469 _mm_mul_pd(dy00,fscal),
470 _mm_mul_pd(dz00,fscal));
471
472 /* Inner loop uses 30 flops */
473 }
474
475 /* End of innermost loop */
476
477 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
478 f+i_coord_offset,fshift+i_shift_offset);
479
480 /* Increment number of inner iterations */
481 inneriter += j_index_end - j_index_start;
482
483 /* Outer loop uses 7 flops */
484 }
485
486 /* Increment number of outer iterations */
487 outeriter += nri;
488
489 /* Update outer/inner flops */
490
491 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
492 }
The ultimate molecular dynamics simulation package