?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Remove nb-parameters from t_forcerec
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_256_single / nb_kernel_ElecEw_VdwNone_GeomP1P1_avx_256_single.c
blobc4c3aca8be23d299808a6d2d43ad916383b983b4
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015,2017, 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_256_single 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_256_single.h"
48
49 /*
50 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_256_single
51 * Electrostatics interaction: Ewald
52 * VdW interaction: None
53 * Geometry: Particle-Particle
54 * Calculate force/pot: PotentialAndForce
55 */
56 void
57 nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_256_single
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,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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,jnrC,jnrD;
74 int jnrE,jnrF,jnrG,jnrH;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real rcutoff_scalar;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
83 real scratch[4*DIM];
84 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
85 real * vdwioffsetptr0;
86 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
87 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
88 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
89 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
90 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
91 real *charge;
92 __m256i ewitab;
93 __m128i ewitab_lo,ewitab_hi;
94 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
95 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
96 real *ewtab;
97 __m256 dummy_mask,cutoff_mask;
98 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
99 __m256 one = _mm256_set1_ps(1.0);
100 __m256 two = _mm256_set1_ps(2.0);
101 x = xx[0];
102 f = ff[0];
103
104 nri = nlist->nri;
105 iinr = nlist->iinr;
106 jindex = nlist->jindex;
107 jjnr = nlist->jjnr;
108 shiftidx = nlist->shift;
109 gid = nlist->gid;
110 shiftvec = fr->shift_vec[0];
111 fshift = fr->fshift[0];
112 facel = _mm256_set1_ps(fr->ic->epsfac);
113 charge = mdatoms->chargeA;
114
115 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
116 beta = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
117 beta2 = _mm256_mul_ps(beta,beta);
118 beta3 = _mm256_mul_ps(beta,beta2);
119
120 ewtab = fr->ic->tabq_coul_FDV0;
121 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
122 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
123
124 /* Avoid stupid compiler warnings */
125 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
126 j_coord_offsetA = 0;
127 j_coord_offsetB = 0;
128 j_coord_offsetC = 0;
129 j_coord_offsetD = 0;
130 j_coord_offsetE = 0;
131 j_coord_offsetF = 0;
132 j_coord_offsetG = 0;
133 j_coord_offsetH = 0;
134
135 outeriter = 0;
136 inneriter = 0;
137
138 for(iidx=0;iidx<4*DIM;iidx++)
139 {
140 scratch[iidx] = 0.0;
141 }
142
143 /* Start outer loop over neighborlists */
144 for(iidx=0; iidx<nri; iidx++)
145 {
146 /* Load shift vector for this list */
147 i_shift_offset = DIM*shiftidx[iidx];
148
149 /* Load limits for loop over neighbors */
150 j_index_start = jindex[iidx];
151 j_index_end = jindex[iidx+1];
152
153 /* Get outer coordinate index */
154 inr = iinr[iidx];
155 i_coord_offset = DIM*inr;
156
157 /* Load i particle coords and add shift vector */
158 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
159
160 fix0 = _mm256_setzero_ps();
161 fiy0 = _mm256_setzero_ps();
162 fiz0 = _mm256_setzero_ps();
163
164 /* Load parameters for i particles */
165 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
166
167 /* Reset potential sums */
168 velecsum = _mm256_setzero_ps();
169
170 /* Start inner kernel loop */
171 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
172 {
173
174 /* Get j neighbor index, and coordinate index */
175 jnrA = jjnr[jidx];
176 jnrB = jjnr[jidx+1];
177 jnrC = jjnr[jidx+2];
178 jnrD = jjnr[jidx+3];
179 jnrE = jjnr[jidx+4];
180 jnrF = jjnr[jidx+5];
181 jnrG = jjnr[jidx+6];
182 jnrH = jjnr[jidx+7];
183 j_coord_offsetA = DIM*jnrA;
184 j_coord_offsetB = DIM*jnrB;
185 j_coord_offsetC = DIM*jnrC;
186 j_coord_offsetD = DIM*jnrD;
187 j_coord_offsetE = DIM*jnrE;
188 j_coord_offsetF = DIM*jnrF;
189 j_coord_offsetG = DIM*jnrG;
190 j_coord_offsetH = DIM*jnrH;
191
192 /* load j atom coordinates */
193 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
194 x+j_coord_offsetC,x+j_coord_offsetD,
195 x+j_coord_offsetE,x+j_coord_offsetF,
196 x+j_coord_offsetG,x+j_coord_offsetH,
197 &jx0,&jy0,&jz0);
198
199 /* Calculate displacement vector */
200 dx00 = _mm256_sub_ps(ix0,jx0);
201 dy00 = _mm256_sub_ps(iy0,jy0);
202 dz00 = _mm256_sub_ps(iz0,jz0);
203
204 /* Calculate squared distance and things based on it */
205 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
206
207 rinv00 = avx256_invsqrt_f(rsq00);
208
209 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
210
211 /* Load parameters for j particles */
212 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
213 charge+jnrC+0,charge+jnrD+0,
214 charge+jnrE+0,charge+jnrF+0,
215 charge+jnrG+0,charge+jnrH+0);
216
217 /**************************
218 * CALCULATE INTERACTIONS *
219 **************************/
220
221 r00 = _mm256_mul_ps(rsq00,rinv00);
222
223 /* Compute parameters for interactions between i and j atoms */
224 qq00 = _mm256_mul_ps(iq0,jq0);
225
226 /* EWALD ELECTROSTATICS */
227
228 /* Analytical PME correction */
229 zeta2 = _mm256_mul_ps(beta2,rsq00);
230 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
231 pmecorrF = avx256_pmecorrF_f(zeta2);
232 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
233 felec = _mm256_mul_ps(qq00,felec);
234 pmecorrV = avx256_pmecorrV_f(zeta2);
235 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
236 velec = _mm256_sub_ps(rinv00,pmecorrV);
237 velec = _mm256_mul_ps(qq00,velec);
238
239 /* Update potential sum for this i atom from the interaction with this j atom. */
240 velecsum = _mm256_add_ps(velecsum,velec);
241
242 fscal = felec;
243
244 /* Calculate temporary vectorial force */
245 tx = _mm256_mul_ps(fscal,dx00);
246 ty = _mm256_mul_ps(fscal,dy00);
247 tz = _mm256_mul_ps(fscal,dz00);
248
249 /* Update vectorial force */
250 fix0 = _mm256_add_ps(fix0,tx);
251 fiy0 = _mm256_add_ps(fiy0,ty);
252 fiz0 = _mm256_add_ps(fiz0,tz);
253
254 fjptrA = f+j_coord_offsetA;
255 fjptrB = f+j_coord_offsetB;
256 fjptrC = f+j_coord_offsetC;
257 fjptrD = f+j_coord_offsetD;
258 fjptrE = f+j_coord_offsetE;
259 fjptrF = f+j_coord_offsetF;
260 fjptrG = f+j_coord_offsetG;
261 fjptrH = f+j_coord_offsetH;
262 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
263
264 /* Inner loop uses 84 flops */
265 }
266
267 if(jidx<j_index_end)
268 {
269
270 /* Get j neighbor index, and coordinate index */
271 jnrlistA = jjnr[jidx];
272 jnrlistB = jjnr[jidx+1];
273 jnrlistC = jjnr[jidx+2];
274 jnrlistD = jjnr[jidx+3];
275 jnrlistE = jjnr[jidx+4];
276 jnrlistF = jjnr[jidx+5];
277 jnrlistG = jjnr[jidx+6];
278 jnrlistH = jjnr[jidx+7];
279 /* Sign of each element will be negative for non-real atoms.
280 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
281 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
282 */
283 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
284 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
285
286 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
287 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
288 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
289 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
290 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
291 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
292 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
293 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
294 j_coord_offsetA = DIM*jnrA;
295 j_coord_offsetB = DIM*jnrB;
296 j_coord_offsetC = DIM*jnrC;
297 j_coord_offsetD = DIM*jnrD;
298 j_coord_offsetE = DIM*jnrE;
299 j_coord_offsetF = DIM*jnrF;
300 j_coord_offsetG = DIM*jnrG;
301 j_coord_offsetH = DIM*jnrH;
302
303 /* load j atom coordinates */
304 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
305 x+j_coord_offsetC,x+j_coord_offsetD,
306 x+j_coord_offsetE,x+j_coord_offsetF,
307 x+j_coord_offsetG,x+j_coord_offsetH,
308 &jx0,&jy0,&jz0);
309
310 /* Calculate displacement vector */
311 dx00 = _mm256_sub_ps(ix0,jx0);
312 dy00 = _mm256_sub_ps(iy0,jy0);
313 dz00 = _mm256_sub_ps(iz0,jz0);
314
315 /* Calculate squared distance and things based on it */
316 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
317
318 rinv00 = avx256_invsqrt_f(rsq00);
319
320 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
321
322 /* Load parameters for j particles */
323 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
324 charge+jnrC+0,charge+jnrD+0,
325 charge+jnrE+0,charge+jnrF+0,
326 charge+jnrG+0,charge+jnrH+0);
327
328 /**************************
329 * CALCULATE INTERACTIONS *
330 **************************/
331
332 r00 = _mm256_mul_ps(rsq00,rinv00);
333 r00 = _mm256_andnot_ps(dummy_mask,r00);
334
335 /* Compute parameters for interactions between i and j atoms */
336 qq00 = _mm256_mul_ps(iq0,jq0);
337
338 /* EWALD ELECTROSTATICS */
339
340 /* Analytical PME correction */
341 zeta2 = _mm256_mul_ps(beta2,rsq00);
342 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
343 pmecorrF = avx256_pmecorrF_f(zeta2);
344 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
345 felec = _mm256_mul_ps(qq00,felec);
346 pmecorrV = avx256_pmecorrV_f(zeta2);
347 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
348 velec = _mm256_sub_ps(rinv00,pmecorrV);
349 velec = _mm256_mul_ps(qq00,velec);
350
351 /* Update potential sum for this i atom from the interaction with this j atom. */
352 velec = _mm256_andnot_ps(dummy_mask,velec);
353 velecsum = _mm256_add_ps(velecsum,velec);
354
355 fscal = felec;
356
357 fscal = _mm256_andnot_ps(dummy_mask,fscal);
358
359 /* Calculate temporary vectorial force */
360 tx = _mm256_mul_ps(fscal,dx00);
361 ty = _mm256_mul_ps(fscal,dy00);
362 tz = _mm256_mul_ps(fscal,dz00);
363
364 /* Update vectorial force */
365 fix0 = _mm256_add_ps(fix0,tx);
366 fiy0 = _mm256_add_ps(fiy0,ty);
367 fiz0 = _mm256_add_ps(fiz0,tz);
368
369 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
370 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
371 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
372 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
373 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
374 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
375 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
376 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
377 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
378
379 /* Inner loop uses 85 flops */
380 }
381
382 /* End of innermost loop */
383
384 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
385 f+i_coord_offset,fshift+i_shift_offset);
386
387 ggid = gid[iidx];
388 /* Update potential energies */
389 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
390
391 /* Increment number of inner iterations */
392 inneriter += j_index_end - j_index_start;
393
394 /* Outer loop uses 8 flops */
395 }
396
397 /* Increment number of outer iterations */
398 outeriter += nri;
399
400 /* Update outer/inner flops */
401
402 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*85);
403 }
404 /*
405 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_256_single
406 * Electrostatics interaction: Ewald
407 * VdW interaction: None
408 * Geometry: Particle-Particle
409 * Calculate force/pot: Force
410 */
411 void
412 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_256_single
413 (t_nblist * gmx_restrict nlist,
414 rvec * gmx_restrict xx,
415 rvec * gmx_restrict ff,
416 struct t_forcerec * gmx_restrict fr,
417 t_mdatoms * gmx_restrict mdatoms,
418 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
419 t_nrnb * gmx_restrict nrnb)
420 {
421 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
422 * just 0 for non-waters.
423 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
424 * jnr indices corresponding to data put in the four positions in the SIMD register.
425 */
426 int i_shift_offset,i_coord_offset,outeriter,inneriter;
427 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
428 int jnrA,jnrB,jnrC,jnrD;
429 int jnrE,jnrF,jnrG,jnrH;
430 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
431 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
432 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
433 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
434 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
435 real rcutoff_scalar;
436 real *shiftvec,*fshift,*x,*f;
437 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
438 real scratch[4*DIM];
439 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
440 real * vdwioffsetptr0;
441 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
442 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
443 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
444 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
445 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
446 real *charge;
447 __m256i ewitab;
448 __m128i ewitab_lo,ewitab_hi;
449 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
450 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
451 real *ewtab;
452 __m256 dummy_mask,cutoff_mask;
453 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
454 __m256 one = _mm256_set1_ps(1.0);
455 __m256 two = _mm256_set1_ps(2.0);
456 x = xx[0];
457 f = ff[0];
458
459 nri = nlist->nri;
460 iinr = nlist->iinr;
461 jindex = nlist->jindex;
462 jjnr = nlist->jjnr;
463 shiftidx = nlist->shift;
464 gid = nlist->gid;
465 shiftvec = fr->shift_vec[0];
466 fshift = fr->fshift[0];
467 facel = _mm256_set1_ps(fr->ic->epsfac);
468 charge = mdatoms->chargeA;
469
470 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
471 beta = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
472 beta2 = _mm256_mul_ps(beta,beta);
473 beta3 = _mm256_mul_ps(beta,beta2);
474
475 ewtab = fr->ic->tabq_coul_F;
476 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
477 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
478
479 /* Avoid stupid compiler warnings */
480 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
481 j_coord_offsetA = 0;
482 j_coord_offsetB = 0;
483 j_coord_offsetC = 0;
484 j_coord_offsetD = 0;
485 j_coord_offsetE = 0;
486 j_coord_offsetF = 0;
487 j_coord_offsetG = 0;
488 j_coord_offsetH = 0;
489
490 outeriter = 0;
491 inneriter = 0;
492
493 for(iidx=0;iidx<4*DIM;iidx++)
494 {
495 scratch[iidx] = 0.0;
496 }
497
498 /* Start outer loop over neighborlists */
499 for(iidx=0; iidx<nri; iidx++)
500 {
501 /* Load shift vector for this list */
502 i_shift_offset = DIM*shiftidx[iidx];
503
504 /* Load limits for loop over neighbors */
505 j_index_start = jindex[iidx];
506 j_index_end = jindex[iidx+1];
507
508 /* Get outer coordinate index */
509 inr = iinr[iidx];
510 i_coord_offset = DIM*inr;
511
512 /* Load i particle coords and add shift vector */
513 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
514
515 fix0 = _mm256_setzero_ps();
516 fiy0 = _mm256_setzero_ps();
517 fiz0 = _mm256_setzero_ps();
518
519 /* Load parameters for i particles */
520 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
521
522 /* Start inner kernel loop */
523 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
524 {
525
526 /* Get j neighbor index, and coordinate index */
527 jnrA = jjnr[jidx];
528 jnrB = jjnr[jidx+1];
529 jnrC = jjnr[jidx+2];
530 jnrD = jjnr[jidx+3];
531 jnrE = jjnr[jidx+4];
532 jnrF = jjnr[jidx+5];
533 jnrG = jjnr[jidx+6];
534 jnrH = jjnr[jidx+7];
535 j_coord_offsetA = DIM*jnrA;
536 j_coord_offsetB = DIM*jnrB;
537 j_coord_offsetC = DIM*jnrC;
538 j_coord_offsetD = DIM*jnrD;
539 j_coord_offsetE = DIM*jnrE;
540 j_coord_offsetF = DIM*jnrF;
541 j_coord_offsetG = DIM*jnrG;
542 j_coord_offsetH = DIM*jnrH;
543
544 /* load j atom coordinates */
545 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
546 x+j_coord_offsetC,x+j_coord_offsetD,
547 x+j_coord_offsetE,x+j_coord_offsetF,
548 x+j_coord_offsetG,x+j_coord_offsetH,
549 &jx0,&jy0,&jz0);
550
551 /* Calculate displacement vector */
552 dx00 = _mm256_sub_ps(ix0,jx0);
553 dy00 = _mm256_sub_ps(iy0,jy0);
554 dz00 = _mm256_sub_ps(iz0,jz0);
555
556 /* Calculate squared distance and things based on it */
557 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
558
559 rinv00 = avx256_invsqrt_f(rsq00);
560
561 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
562
563 /* Load parameters for j particles */
564 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
565 charge+jnrC+0,charge+jnrD+0,
566 charge+jnrE+0,charge+jnrF+0,
567 charge+jnrG+0,charge+jnrH+0);
568
569 /**************************
570 * CALCULATE INTERACTIONS *
571 **************************/
572
573 r00 = _mm256_mul_ps(rsq00,rinv00);
574
575 /* Compute parameters for interactions between i and j atoms */
576 qq00 = _mm256_mul_ps(iq0,jq0);
577
578 /* EWALD ELECTROSTATICS */
579
580 /* Analytical PME correction */
581 zeta2 = _mm256_mul_ps(beta2,rsq00);
582 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
583 pmecorrF = avx256_pmecorrF_f(zeta2);
584 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
585 felec = _mm256_mul_ps(qq00,felec);
586
587 fscal = felec;
588
589 /* Calculate temporary vectorial force */
590 tx = _mm256_mul_ps(fscal,dx00);
591 ty = _mm256_mul_ps(fscal,dy00);
592 tz = _mm256_mul_ps(fscal,dz00);
593
594 /* Update vectorial force */
595 fix0 = _mm256_add_ps(fix0,tx);
596 fiy0 = _mm256_add_ps(fiy0,ty);
597 fiz0 = _mm256_add_ps(fiz0,tz);
598
599 fjptrA = f+j_coord_offsetA;
600 fjptrB = f+j_coord_offsetB;
601 fjptrC = f+j_coord_offsetC;
602 fjptrD = f+j_coord_offsetD;
603 fjptrE = f+j_coord_offsetE;
604 fjptrF = f+j_coord_offsetF;
605 fjptrG = f+j_coord_offsetG;
606 fjptrH = f+j_coord_offsetH;
607 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
608
609 /* Inner loop uses 56 flops */
610 }
611
612 if(jidx<j_index_end)
613 {
614
615 /* Get j neighbor index, and coordinate index */
616 jnrlistA = jjnr[jidx];
617 jnrlistB = jjnr[jidx+1];
618 jnrlistC = jjnr[jidx+2];
619 jnrlistD = jjnr[jidx+3];
620 jnrlistE = jjnr[jidx+4];
621 jnrlistF = jjnr[jidx+5];
622 jnrlistG = jjnr[jidx+6];
623 jnrlistH = jjnr[jidx+7];
624 /* Sign of each element will be negative for non-real atoms.
625 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
626 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
627 */
628 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
629 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
630
631 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
632 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
633 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
634 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
635 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
636 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
637 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
638 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
639 j_coord_offsetA = DIM*jnrA;
640 j_coord_offsetB = DIM*jnrB;
641 j_coord_offsetC = DIM*jnrC;
642 j_coord_offsetD = DIM*jnrD;
643 j_coord_offsetE = DIM*jnrE;
644 j_coord_offsetF = DIM*jnrF;
645 j_coord_offsetG = DIM*jnrG;
646 j_coord_offsetH = DIM*jnrH;
647
648 /* load j atom coordinates */
649 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
650 x+j_coord_offsetC,x+j_coord_offsetD,
651 x+j_coord_offsetE,x+j_coord_offsetF,
652 x+j_coord_offsetG,x+j_coord_offsetH,
653 &jx0,&jy0,&jz0);
654
655 /* Calculate displacement vector */
656 dx00 = _mm256_sub_ps(ix0,jx0);
657 dy00 = _mm256_sub_ps(iy0,jy0);
658 dz00 = _mm256_sub_ps(iz0,jz0);
659
660 /* Calculate squared distance and things based on it */
661 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
662
663 rinv00 = avx256_invsqrt_f(rsq00);
664
665 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
666
667 /* Load parameters for j particles */
668 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
669 charge+jnrC+0,charge+jnrD+0,
670 charge+jnrE+0,charge+jnrF+0,
671 charge+jnrG+0,charge+jnrH+0);
672
673 /**************************
674 * CALCULATE INTERACTIONS *
675 **************************/
676
677 r00 = _mm256_mul_ps(rsq00,rinv00);
678 r00 = _mm256_andnot_ps(dummy_mask,r00);
679
680 /* Compute parameters for interactions between i and j atoms */
681 qq00 = _mm256_mul_ps(iq0,jq0);
682
683 /* EWALD ELECTROSTATICS */
684
685 /* Analytical PME correction */
686 zeta2 = _mm256_mul_ps(beta2,rsq00);
687 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
688 pmecorrF = avx256_pmecorrF_f(zeta2);
689 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
690 felec = _mm256_mul_ps(qq00,felec);
691
692 fscal = felec;
693
694 fscal = _mm256_andnot_ps(dummy_mask,fscal);
695
696 /* Calculate temporary vectorial force */
697 tx = _mm256_mul_ps(fscal,dx00);
698 ty = _mm256_mul_ps(fscal,dy00);
699 tz = _mm256_mul_ps(fscal,dz00);
700
701 /* Update vectorial force */
702 fix0 = _mm256_add_ps(fix0,tx);
703 fiy0 = _mm256_add_ps(fiy0,ty);
704 fiz0 = _mm256_add_ps(fiz0,tz);
705
706 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
707 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
708 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
709 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
710 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
711 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
712 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
713 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
714 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
715
716 /* Inner loop uses 57 flops */
717 }
718
719 /* End of innermost loop */
720
721 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
722 f+i_coord_offset,fshift+i_shift_offset);
723
724 /* Increment number of inner iterations */
725 inneriter += j_index_end - j_index_start;
726
727 /* Outer loop uses 7 flops */
728 }
729
730 /* Increment number of outer iterations */
731 outeriter += nri;
732
733 /* Update outer/inner flops */
734
735 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*57);
736 }
The ultimate molecular dynamics simulation package