2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, by the GROMACS development team, led by
5 * David van der Spoel, Berk Hess, Erik Lindahl, and including many
6 * others, as listed in the AUTHORS file in the top-level source
7 * directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_256_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gmx_math_x86_avx_256_single.h"
50 #include "kernelutil_x86_avx_256_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_256_single
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: None
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_VF_avx_256_single
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,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrE,jnrF,jnrG,jnrH;
78 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
79 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
80 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
81 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
82 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
84 real *shiftvec,*fshift,*x,*f;
85 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
87 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
88 real * vdwioffsetptr0;
89 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
91 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
95 __m256 dummy_mask,cutoff_mask;
96 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
97 __m256 one = _mm256_set1_ps(1.0);
98 __m256 two = _mm256_set1_ps(2.0);
104 jindex = nlist->jindex;
106 shiftidx = nlist->shift;
108 shiftvec = fr->shift_vec[0];
109 fshift = fr->fshift[0];
110 facel = _mm256_set1_ps(fr->epsfac);
111 charge = mdatoms->chargeA;
112 krf = _mm256_set1_ps(fr->ic->k_rf);
113 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
114 crf = _mm256_set1_ps(fr->ic->c_rf);
116 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
117 rcutoff_scalar = fr->rcoulomb;
118 rcutoff = _mm256_set1_ps(rcutoff_scalar);
119 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
121 /* Avoid stupid compiler warnings */
122 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
135 for(iidx=0;iidx<4*DIM;iidx++)
140 /* Start outer loop over neighborlists */
141 for(iidx=0; iidx<nri; iidx++)
143 /* Load shift vector for this list */
144 i_shift_offset = DIM*shiftidx[iidx];
146 /* Load limits for loop over neighbors */
147 j_index_start = jindex[iidx];
148 j_index_end = jindex[iidx+1];
150 /* Get outer coordinate index */
152 i_coord_offset = DIM*inr;
154 /* Load i particle coords and add shift vector */
155 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
157 fix0 = _mm256_setzero_ps();
158 fiy0 = _mm256_setzero_ps();
159 fiz0 = _mm256_setzero_ps();
161 /* Load parameters for i particles */
162 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
164 /* Reset potential sums */
165 velecsum = _mm256_setzero_ps();
167 /* Start inner kernel loop */
168 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
171 /* Get j neighbor index, and coordinate index */
180 j_coord_offsetA = DIM*jnrA;
181 j_coord_offsetB = DIM*jnrB;
182 j_coord_offsetC = DIM*jnrC;
183 j_coord_offsetD = DIM*jnrD;
184 j_coord_offsetE = DIM*jnrE;
185 j_coord_offsetF = DIM*jnrF;
186 j_coord_offsetG = DIM*jnrG;
187 j_coord_offsetH = DIM*jnrH;
189 /* load j atom coordinates */
190 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
191 x+j_coord_offsetC,x+j_coord_offsetD,
192 x+j_coord_offsetE,x+j_coord_offsetF,
193 x+j_coord_offsetG,x+j_coord_offsetH,
196 /* Calculate displacement vector */
197 dx00 = _mm256_sub_ps(ix0,jx0);
198 dy00 = _mm256_sub_ps(iy0,jy0);
199 dz00 = _mm256_sub_ps(iz0,jz0);
201 /* Calculate squared distance and things based on it */
202 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
204 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
206 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
208 /* Load parameters for j particles */
209 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
210 charge+jnrC+0,charge+jnrD+0,
211 charge+jnrE+0,charge+jnrF+0,
212 charge+jnrG+0,charge+jnrH+0);
214 /**************************
215 * CALCULATE INTERACTIONS *
216 **************************/
218 if (gmx_mm256_any_lt(rsq00,rcutoff2))
221 /* Compute parameters for interactions between i and j atoms */
222 qq00 = _mm256_mul_ps(iq0,jq0);
224 /* REACTION-FIELD ELECTROSTATICS */
225 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
226 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
228 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
230 /* Update potential sum for this i atom from the interaction with this j atom. */
231 velec = _mm256_and_ps(velec,cutoff_mask);
232 velecsum = _mm256_add_ps(velecsum,velec);
236 fscal = _mm256_and_ps(fscal,cutoff_mask);
238 /* Calculate temporary vectorial force */
239 tx = _mm256_mul_ps(fscal,dx00);
240 ty = _mm256_mul_ps(fscal,dy00);
241 tz = _mm256_mul_ps(fscal,dz00);
243 /* Update vectorial force */
244 fix0 = _mm256_add_ps(fix0,tx);
245 fiy0 = _mm256_add_ps(fiy0,ty);
246 fiz0 = _mm256_add_ps(fiz0,tz);
248 fjptrA = f+j_coord_offsetA;
249 fjptrB = f+j_coord_offsetB;
250 fjptrC = f+j_coord_offsetC;
251 fjptrD = f+j_coord_offsetD;
252 fjptrE = f+j_coord_offsetE;
253 fjptrF = f+j_coord_offsetF;
254 fjptrG = f+j_coord_offsetG;
255 fjptrH = f+j_coord_offsetH;
256 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
260 /* Inner loop uses 36 flops */
266 /* Get j neighbor index, and coordinate index */
267 jnrlistA = jjnr[jidx];
268 jnrlistB = jjnr[jidx+1];
269 jnrlistC = jjnr[jidx+2];
270 jnrlistD = jjnr[jidx+3];
271 jnrlistE = jjnr[jidx+4];
272 jnrlistF = jjnr[jidx+5];
273 jnrlistG = jjnr[jidx+6];
274 jnrlistH = jjnr[jidx+7];
275 /* Sign of each element will be negative for non-real atoms.
276 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
277 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
279 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
280 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
282 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
283 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
284 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
285 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
286 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
287 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
288 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
289 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
290 j_coord_offsetA = DIM*jnrA;
291 j_coord_offsetB = DIM*jnrB;
292 j_coord_offsetC = DIM*jnrC;
293 j_coord_offsetD = DIM*jnrD;
294 j_coord_offsetE = DIM*jnrE;
295 j_coord_offsetF = DIM*jnrF;
296 j_coord_offsetG = DIM*jnrG;
297 j_coord_offsetH = DIM*jnrH;
299 /* load j atom coordinates */
300 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
301 x+j_coord_offsetC,x+j_coord_offsetD,
302 x+j_coord_offsetE,x+j_coord_offsetF,
303 x+j_coord_offsetG,x+j_coord_offsetH,
306 /* Calculate displacement vector */
307 dx00 = _mm256_sub_ps(ix0,jx0);
308 dy00 = _mm256_sub_ps(iy0,jy0);
309 dz00 = _mm256_sub_ps(iz0,jz0);
311 /* Calculate squared distance and things based on it */
312 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
314 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
316 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
318 /* Load parameters for j particles */
319 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
320 charge+jnrC+0,charge+jnrD+0,
321 charge+jnrE+0,charge+jnrF+0,
322 charge+jnrG+0,charge+jnrH+0);
324 /**************************
325 * CALCULATE INTERACTIONS *
326 **************************/
328 if (gmx_mm256_any_lt(rsq00,rcutoff2))
331 /* Compute parameters for interactions between i and j atoms */
332 qq00 = _mm256_mul_ps(iq0,jq0);
334 /* REACTION-FIELD ELECTROSTATICS */
335 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
336 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
338 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
340 /* Update potential sum for this i atom from the interaction with this j atom. */
341 velec = _mm256_and_ps(velec,cutoff_mask);
342 velec = _mm256_andnot_ps(dummy_mask,velec);
343 velecsum = _mm256_add_ps(velecsum,velec);
347 fscal = _mm256_and_ps(fscal,cutoff_mask);
349 fscal = _mm256_andnot_ps(dummy_mask,fscal);
351 /* Calculate temporary vectorial force */
352 tx = _mm256_mul_ps(fscal,dx00);
353 ty = _mm256_mul_ps(fscal,dy00);
354 tz = _mm256_mul_ps(fscal,dz00);
356 /* Update vectorial force */
357 fix0 = _mm256_add_ps(fix0,tx);
358 fiy0 = _mm256_add_ps(fiy0,ty);
359 fiz0 = _mm256_add_ps(fiz0,tz);
361 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
362 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
363 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
364 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
365 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
366 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
367 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
368 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
369 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
373 /* Inner loop uses 36 flops */
376 /* End of innermost loop */
378 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
379 f+i_coord_offset,fshift+i_shift_offset);
382 /* Update potential energies */
383 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
385 /* Increment number of inner iterations */
386 inneriter += j_index_end - j_index_start;
388 /* Outer loop uses 8 flops */
391 /* Increment number of outer iterations */
394 /* Update outer/inner flops */
396 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
399 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_single
400 * Electrostatics interaction: ReactionField
401 * VdW interaction: None
402 * Geometry: Particle-Particle
403 * Calculate force/pot: Force
406 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_256_single
407 (t_nblist * gmx_restrict nlist,
408 rvec * gmx_restrict xx,
409 rvec * gmx_restrict ff,
410 t_forcerec * gmx_restrict fr,
411 t_mdatoms * gmx_restrict mdatoms,
412 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
413 t_nrnb * gmx_restrict nrnb)
415 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
416 * just 0 for non-waters.
417 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
418 * jnr indices corresponding to data put in the four positions in the SIMD register.
420 int i_shift_offset,i_coord_offset,outeriter,inneriter;
421 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
422 int jnrA,jnrB,jnrC,jnrD;
423 int jnrE,jnrF,jnrG,jnrH;
424 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
425 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
426 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
427 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
428 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
430 real *shiftvec,*fshift,*x,*f;
431 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
433 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
434 real * vdwioffsetptr0;
435 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
436 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
437 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
438 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
439 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
441 __m256 dummy_mask,cutoff_mask;
442 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
443 __m256 one = _mm256_set1_ps(1.0);
444 __m256 two = _mm256_set1_ps(2.0);
450 jindex = nlist->jindex;
452 shiftidx = nlist->shift;
454 shiftvec = fr->shift_vec[0];
455 fshift = fr->fshift[0];
456 facel = _mm256_set1_ps(fr->epsfac);
457 charge = mdatoms->chargeA;
458 krf = _mm256_set1_ps(fr->ic->k_rf);
459 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
460 crf = _mm256_set1_ps(fr->ic->c_rf);
462 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
463 rcutoff_scalar = fr->rcoulomb;
464 rcutoff = _mm256_set1_ps(rcutoff_scalar);
465 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
467 /* Avoid stupid compiler warnings */
468 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
481 for(iidx=0;iidx<4*DIM;iidx++)
486 /* Start outer loop over neighborlists */
487 for(iidx=0; iidx<nri; iidx++)
489 /* Load shift vector for this list */
490 i_shift_offset = DIM*shiftidx[iidx];
492 /* Load limits for loop over neighbors */
493 j_index_start = jindex[iidx];
494 j_index_end = jindex[iidx+1];
496 /* Get outer coordinate index */
498 i_coord_offset = DIM*inr;
500 /* Load i particle coords and add shift vector */
501 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
503 fix0 = _mm256_setzero_ps();
504 fiy0 = _mm256_setzero_ps();
505 fiz0 = _mm256_setzero_ps();
507 /* Load parameters for i particles */
508 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
510 /* Start inner kernel loop */
511 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
514 /* Get j neighbor index, and coordinate index */
523 j_coord_offsetA = DIM*jnrA;
524 j_coord_offsetB = DIM*jnrB;
525 j_coord_offsetC = DIM*jnrC;
526 j_coord_offsetD = DIM*jnrD;
527 j_coord_offsetE = DIM*jnrE;
528 j_coord_offsetF = DIM*jnrF;
529 j_coord_offsetG = DIM*jnrG;
530 j_coord_offsetH = DIM*jnrH;
532 /* load j atom coordinates */
533 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
534 x+j_coord_offsetC,x+j_coord_offsetD,
535 x+j_coord_offsetE,x+j_coord_offsetF,
536 x+j_coord_offsetG,x+j_coord_offsetH,
539 /* Calculate displacement vector */
540 dx00 = _mm256_sub_ps(ix0,jx0);
541 dy00 = _mm256_sub_ps(iy0,jy0);
542 dz00 = _mm256_sub_ps(iz0,jz0);
544 /* Calculate squared distance and things based on it */
545 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
547 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
549 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
551 /* Load parameters for j particles */
552 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
553 charge+jnrC+0,charge+jnrD+0,
554 charge+jnrE+0,charge+jnrF+0,
555 charge+jnrG+0,charge+jnrH+0);
557 /**************************
558 * CALCULATE INTERACTIONS *
559 **************************/
561 if (gmx_mm256_any_lt(rsq00,rcutoff2))
564 /* Compute parameters for interactions between i and j atoms */
565 qq00 = _mm256_mul_ps(iq0,jq0);
567 /* REACTION-FIELD ELECTROSTATICS */
568 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
570 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
574 fscal = _mm256_and_ps(fscal,cutoff_mask);
576 /* Calculate temporary vectorial force */
577 tx = _mm256_mul_ps(fscal,dx00);
578 ty = _mm256_mul_ps(fscal,dy00);
579 tz = _mm256_mul_ps(fscal,dz00);
581 /* Update vectorial force */
582 fix0 = _mm256_add_ps(fix0,tx);
583 fiy0 = _mm256_add_ps(fiy0,ty);
584 fiz0 = _mm256_add_ps(fiz0,tz);
586 fjptrA = f+j_coord_offsetA;
587 fjptrB = f+j_coord_offsetB;
588 fjptrC = f+j_coord_offsetC;
589 fjptrD = f+j_coord_offsetD;
590 fjptrE = f+j_coord_offsetE;
591 fjptrF = f+j_coord_offsetF;
592 fjptrG = f+j_coord_offsetG;
593 fjptrH = f+j_coord_offsetH;
594 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
598 /* Inner loop uses 30 flops */
604 /* Get j neighbor index, and coordinate index */
605 jnrlistA = jjnr[jidx];
606 jnrlistB = jjnr[jidx+1];
607 jnrlistC = jjnr[jidx+2];
608 jnrlistD = jjnr[jidx+3];
609 jnrlistE = jjnr[jidx+4];
610 jnrlistF = jjnr[jidx+5];
611 jnrlistG = jjnr[jidx+6];
612 jnrlistH = jjnr[jidx+7];
613 /* Sign of each element will be negative for non-real atoms.
614 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
615 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
617 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
618 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
620 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
621 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
622 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
623 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
624 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
625 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
626 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
627 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
628 j_coord_offsetA = DIM*jnrA;
629 j_coord_offsetB = DIM*jnrB;
630 j_coord_offsetC = DIM*jnrC;
631 j_coord_offsetD = DIM*jnrD;
632 j_coord_offsetE = DIM*jnrE;
633 j_coord_offsetF = DIM*jnrF;
634 j_coord_offsetG = DIM*jnrG;
635 j_coord_offsetH = DIM*jnrH;
637 /* load j atom coordinates */
638 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
639 x+j_coord_offsetC,x+j_coord_offsetD,
640 x+j_coord_offsetE,x+j_coord_offsetF,
641 x+j_coord_offsetG,x+j_coord_offsetH,
644 /* Calculate displacement vector */
645 dx00 = _mm256_sub_ps(ix0,jx0);
646 dy00 = _mm256_sub_ps(iy0,jy0);
647 dz00 = _mm256_sub_ps(iz0,jz0);
649 /* Calculate squared distance and things based on it */
650 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
652 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
654 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
656 /* Load parameters for j particles */
657 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
658 charge+jnrC+0,charge+jnrD+0,
659 charge+jnrE+0,charge+jnrF+0,
660 charge+jnrG+0,charge+jnrH+0);
662 /**************************
663 * CALCULATE INTERACTIONS *
664 **************************/
666 if (gmx_mm256_any_lt(rsq00,rcutoff2))
669 /* Compute parameters for interactions between i and j atoms */
670 qq00 = _mm256_mul_ps(iq0,jq0);
672 /* REACTION-FIELD ELECTROSTATICS */
673 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
675 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
679 fscal = _mm256_and_ps(fscal,cutoff_mask);
681 fscal = _mm256_andnot_ps(dummy_mask,fscal);
683 /* Calculate temporary vectorial force */
684 tx = _mm256_mul_ps(fscal,dx00);
685 ty = _mm256_mul_ps(fscal,dy00);
686 tz = _mm256_mul_ps(fscal,dz00);
688 /* Update vectorial force */
689 fix0 = _mm256_add_ps(fix0,tx);
690 fiy0 = _mm256_add_ps(fiy0,ty);
691 fiz0 = _mm256_add_ps(fiz0,tz);
693 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
694 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
695 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
696 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
697 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
698 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
699 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
700 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
701 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
705 /* Inner loop uses 30 flops */
708 /* End of innermost loop */
710 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
711 f+i_coord_offset,fshift+i_shift_offset);
713 /* Increment number of inner iterations */
714 inneriter += j_index_end - j_index_start;
716 /* Outer loop uses 7 flops */
719 /* Increment number of outer iterations */
722 /* Update outer/inner flops */
724 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);