2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_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_ElecEwSh_VdwNone_GeomW3W3_VF_avx_256_single
54 * Electrostatics interaction: Ewald
55 * VdW interaction: None
56 * Geometry: Water3-Water3
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_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 real * vdwioffsetptr1;
91 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
92 real * vdwioffsetptr2;
93 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
94 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
95 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
96 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
97 __m256 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
98 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
99 __m256 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
100 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
101 __m256 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
102 __m256 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
103 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
104 __m256 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
105 __m256 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
106 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
107 __m256 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
108 __m256 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
109 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
112 __m128i ewitab_lo,ewitab_hi;
113 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
114 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
116 __m256 dummy_mask,cutoff_mask;
117 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
118 __m256 one = _mm256_set1_ps(1.0);
119 __m256 two = _mm256_set1_ps(2.0);
125 jindex = nlist->jindex;
127 shiftidx = nlist->shift;
129 shiftvec = fr->shift_vec[0];
130 fshift = fr->fshift[0];
131 facel = _mm256_set1_ps(fr->epsfac);
132 charge = mdatoms->chargeA;
134 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
135 beta = _mm256_set1_ps(fr->ic->ewaldcoeff);
136 beta2 = _mm256_mul_ps(beta,beta);
137 beta3 = _mm256_mul_ps(beta,beta2);
139 ewtab = fr->ic->tabq_coul_FDV0;
140 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
141 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
143 /* Setup water-specific parameters */
144 inr = nlist->iinr[0];
145 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
146 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
147 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
149 jq0 = _mm256_set1_ps(charge[inr+0]);
150 jq1 = _mm256_set1_ps(charge[inr+1]);
151 jq2 = _mm256_set1_ps(charge[inr+2]);
152 qq00 = _mm256_mul_ps(iq0,jq0);
153 qq01 = _mm256_mul_ps(iq0,jq1);
154 qq02 = _mm256_mul_ps(iq0,jq2);
155 qq10 = _mm256_mul_ps(iq1,jq0);
156 qq11 = _mm256_mul_ps(iq1,jq1);
157 qq12 = _mm256_mul_ps(iq1,jq2);
158 qq20 = _mm256_mul_ps(iq2,jq0);
159 qq21 = _mm256_mul_ps(iq2,jq1);
160 qq22 = _mm256_mul_ps(iq2,jq2);
162 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
163 rcutoff_scalar = fr->rcoulomb;
164 rcutoff = _mm256_set1_ps(rcutoff_scalar);
165 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
167 /* Avoid stupid compiler warnings */
168 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
181 for(iidx=0;iidx<4*DIM;iidx++)
186 /* Start outer loop over neighborlists */
187 for(iidx=0; iidx<nri; iidx++)
189 /* Load shift vector for this list */
190 i_shift_offset = DIM*shiftidx[iidx];
192 /* Load limits for loop over neighbors */
193 j_index_start = jindex[iidx];
194 j_index_end = jindex[iidx+1];
196 /* Get outer coordinate index */
198 i_coord_offset = DIM*inr;
200 /* Load i particle coords and add shift vector */
201 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
202 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
204 fix0 = _mm256_setzero_ps();
205 fiy0 = _mm256_setzero_ps();
206 fiz0 = _mm256_setzero_ps();
207 fix1 = _mm256_setzero_ps();
208 fiy1 = _mm256_setzero_ps();
209 fiz1 = _mm256_setzero_ps();
210 fix2 = _mm256_setzero_ps();
211 fiy2 = _mm256_setzero_ps();
212 fiz2 = _mm256_setzero_ps();
214 /* Reset potential sums */
215 velecsum = _mm256_setzero_ps();
217 /* Start inner kernel loop */
218 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
221 /* Get j neighbor index, and coordinate index */
230 j_coord_offsetA = DIM*jnrA;
231 j_coord_offsetB = DIM*jnrB;
232 j_coord_offsetC = DIM*jnrC;
233 j_coord_offsetD = DIM*jnrD;
234 j_coord_offsetE = DIM*jnrE;
235 j_coord_offsetF = DIM*jnrF;
236 j_coord_offsetG = DIM*jnrG;
237 j_coord_offsetH = DIM*jnrH;
239 /* load j atom coordinates */
240 gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
241 x+j_coord_offsetC,x+j_coord_offsetD,
242 x+j_coord_offsetE,x+j_coord_offsetF,
243 x+j_coord_offsetG,x+j_coord_offsetH,
244 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
246 /* Calculate displacement vector */
247 dx00 = _mm256_sub_ps(ix0,jx0);
248 dy00 = _mm256_sub_ps(iy0,jy0);
249 dz00 = _mm256_sub_ps(iz0,jz0);
250 dx01 = _mm256_sub_ps(ix0,jx1);
251 dy01 = _mm256_sub_ps(iy0,jy1);
252 dz01 = _mm256_sub_ps(iz0,jz1);
253 dx02 = _mm256_sub_ps(ix0,jx2);
254 dy02 = _mm256_sub_ps(iy0,jy2);
255 dz02 = _mm256_sub_ps(iz0,jz2);
256 dx10 = _mm256_sub_ps(ix1,jx0);
257 dy10 = _mm256_sub_ps(iy1,jy0);
258 dz10 = _mm256_sub_ps(iz1,jz0);
259 dx11 = _mm256_sub_ps(ix1,jx1);
260 dy11 = _mm256_sub_ps(iy1,jy1);
261 dz11 = _mm256_sub_ps(iz1,jz1);
262 dx12 = _mm256_sub_ps(ix1,jx2);
263 dy12 = _mm256_sub_ps(iy1,jy2);
264 dz12 = _mm256_sub_ps(iz1,jz2);
265 dx20 = _mm256_sub_ps(ix2,jx0);
266 dy20 = _mm256_sub_ps(iy2,jy0);
267 dz20 = _mm256_sub_ps(iz2,jz0);
268 dx21 = _mm256_sub_ps(ix2,jx1);
269 dy21 = _mm256_sub_ps(iy2,jy1);
270 dz21 = _mm256_sub_ps(iz2,jz1);
271 dx22 = _mm256_sub_ps(ix2,jx2);
272 dy22 = _mm256_sub_ps(iy2,jy2);
273 dz22 = _mm256_sub_ps(iz2,jz2);
275 /* Calculate squared distance and things based on it */
276 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
277 rsq01 = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
278 rsq02 = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
279 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
280 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
281 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
282 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
283 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
284 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
286 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
287 rinv01 = gmx_mm256_invsqrt_ps(rsq01);
288 rinv02 = gmx_mm256_invsqrt_ps(rsq02);
289 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
290 rinv11 = gmx_mm256_invsqrt_ps(rsq11);
291 rinv12 = gmx_mm256_invsqrt_ps(rsq12);
292 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
293 rinv21 = gmx_mm256_invsqrt_ps(rsq21);
294 rinv22 = gmx_mm256_invsqrt_ps(rsq22);
296 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
297 rinvsq01 = _mm256_mul_ps(rinv01,rinv01);
298 rinvsq02 = _mm256_mul_ps(rinv02,rinv02);
299 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
300 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
301 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
302 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
303 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
304 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
306 fjx0 = _mm256_setzero_ps();
307 fjy0 = _mm256_setzero_ps();
308 fjz0 = _mm256_setzero_ps();
309 fjx1 = _mm256_setzero_ps();
310 fjy1 = _mm256_setzero_ps();
311 fjz1 = _mm256_setzero_ps();
312 fjx2 = _mm256_setzero_ps();
313 fjy2 = _mm256_setzero_ps();
314 fjz2 = _mm256_setzero_ps();
316 /**************************
317 * CALCULATE INTERACTIONS *
318 **************************/
320 if (gmx_mm256_any_lt(rsq00,rcutoff2))
323 r00 = _mm256_mul_ps(rsq00,rinv00);
325 /* EWALD ELECTROSTATICS */
327 /* Analytical PME correction */
328 zeta2 = _mm256_mul_ps(beta2,rsq00);
329 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
330 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
331 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
332 felec = _mm256_mul_ps(qq00,felec);
333 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
334 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
335 velec = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
336 velec = _mm256_mul_ps(qq00,velec);
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 velecsum = _mm256_add_ps(velecsum,velec);
346 fscal = _mm256_and_ps(fscal,cutoff_mask);
348 /* Calculate temporary vectorial force */
349 tx = _mm256_mul_ps(fscal,dx00);
350 ty = _mm256_mul_ps(fscal,dy00);
351 tz = _mm256_mul_ps(fscal,dz00);
353 /* Update vectorial force */
354 fix0 = _mm256_add_ps(fix0,tx);
355 fiy0 = _mm256_add_ps(fiy0,ty);
356 fiz0 = _mm256_add_ps(fiz0,tz);
358 fjx0 = _mm256_add_ps(fjx0,tx);
359 fjy0 = _mm256_add_ps(fjy0,ty);
360 fjz0 = _mm256_add_ps(fjz0,tz);
364 /**************************
365 * CALCULATE INTERACTIONS *
366 **************************/
368 if (gmx_mm256_any_lt(rsq01,rcutoff2))
371 r01 = _mm256_mul_ps(rsq01,rinv01);
373 /* EWALD ELECTROSTATICS */
375 /* Analytical PME correction */
376 zeta2 = _mm256_mul_ps(beta2,rsq01);
377 rinv3 = _mm256_mul_ps(rinvsq01,rinv01);
378 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
379 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
380 felec = _mm256_mul_ps(qq01,felec);
381 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
382 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
383 velec = _mm256_sub_ps(_mm256_sub_ps(rinv01,sh_ewald),pmecorrV);
384 velec = _mm256_mul_ps(qq01,velec);
386 cutoff_mask = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
388 /* Update potential sum for this i atom from the interaction with this j atom. */
389 velec = _mm256_and_ps(velec,cutoff_mask);
390 velecsum = _mm256_add_ps(velecsum,velec);
394 fscal = _mm256_and_ps(fscal,cutoff_mask);
396 /* Calculate temporary vectorial force */
397 tx = _mm256_mul_ps(fscal,dx01);
398 ty = _mm256_mul_ps(fscal,dy01);
399 tz = _mm256_mul_ps(fscal,dz01);
401 /* Update vectorial force */
402 fix0 = _mm256_add_ps(fix0,tx);
403 fiy0 = _mm256_add_ps(fiy0,ty);
404 fiz0 = _mm256_add_ps(fiz0,tz);
406 fjx1 = _mm256_add_ps(fjx1,tx);
407 fjy1 = _mm256_add_ps(fjy1,ty);
408 fjz1 = _mm256_add_ps(fjz1,tz);
412 /**************************
413 * CALCULATE INTERACTIONS *
414 **************************/
416 if (gmx_mm256_any_lt(rsq02,rcutoff2))
419 r02 = _mm256_mul_ps(rsq02,rinv02);
421 /* EWALD ELECTROSTATICS */
423 /* Analytical PME correction */
424 zeta2 = _mm256_mul_ps(beta2,rsq02);
425 rinv3 = _mm256_mul_ps(rinvsq02,rinv02);
426 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
427 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
428 felec = _mm256_mul_ps(qq02,felec);
429 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
430 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
431 velec = _mm256_sub_ps(_mm256_sub_ps(rinv02,sh_ewald),pmecorrV);
432 velec = _mm256_mul_ps(qq02,velec);
434 cutoff_mask = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
436 /* Update potential sum for this i atom from the interaction with this j atom. */
437 velec = _mm256_and_ps(velec,cutoff_mask);
438 velecsum = _mm256_add_ps(velecsum,velec);
442 fscal = _mm256_and_ps(fscal,cutoff_mask);
444 /* Calculate temporary vectorial force */
445 tx = _mm256_mul_ps(fscal,dx02);
446 ty = _mm256_mul_ps(fscal,dy02);
447 tz = _mm256_mul_ps(fscal,dz02);
449 /* Update vectorial force */
450 fix0 = _mm256_add_ps(fix0,tx);
451 fiy0 = _mm256_add_ps(fiy0,ty);
452 fiz0 = _mm256_add_ps(fiz0,tz);
454 fjx2 = _mm256_add_ps(fjx2,tx);
455 fjy2 = _mm256_add_ps(fjy2,ty);
456 fjz2 = _mm256_add_ps(fjz2,tz);
460 /**************************
461 * CALCULATE INTERACTIONS *
462 **************************/
464 if (gmx_mm256_any_lt(rsq10,rcutoff2))
467 r10 = _mm256_mul_ps(rsq10,rinv10);
469 /* EWALD ELECTROSTATICS */
471 /* Analytical PME correction */
472 zeta2 = _mm256_mul_ps(beta2,rsq10);
473 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
474 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
475 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
476 felec = _mm256_mul_ps(qq10,felec);
477 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
478 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
479 velec = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
480 velec = _mm256_mul_ps(qq10,velec);
482 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
484 /* Update potential sum for this i atom from the interaction with this j atom. */
485 velec = _mm256_and_ps(velec,cutoff_mask);
486 velecsum = _mm256_add_ps(velecsum,velec);
490 fscal = _mm256_and_ps(fscal,cutoff_mask);
492 /* Calculate temporary vectorial force */
493 tx = _mm256_mul_ps(fscal,dx10);
494 ty = _mm256_mul_ps(fscal,dy10);
495 tz = _mm256_mul_ps(fscal,dz10);
497 /* Update vectorial force */
498 fix1 = _mm256_add_ps(fix1,tx);
499 fiy1 = _mm256_add_ps(fiy1,ty);
500 fiz1 = _mm256_add_ps(fiz1,tz);
502 fjx0 = _mm256_add_ps(fjx0,tx);
503 fjy0 = _mm256_add_ps(fjy0,ty);
504 fjz0 = _mm256_add_ps(fjz0,tz);
508 /**************************
509 * CALCULATE INTERACTIONS *
510 **************************/
512 if (gmx_mm256_any_lt(rsq11,rcutoff2))
515 r11 = _mm256_mul_ps(rsq11,rinv11);
517 /* EWALD ELECTROSTATICS */
519 /* Analytical PME correction */
520 zeta2 = _mm256_mul_ps(beta2,rsq11);
521 rinv3 = _mm256_mul_ps(rinvsq11,rinv11);
522 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
523 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
524 felec = _mm256_mul_ps(qq11,felec);
525 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
526 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
527 velec = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
528 velec = _mm256_mul_ps(qq11,velec);
530 cutoff_mask = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
532 /* Update potential sum for this i atom from the interaction with this j atom. */
533 velec = _mm256_and_ps(velec,cutoff_mask);
534 velecsum = _mm256_add_ps(velecsum,velec);
538 fscal = _mm256_and_ps(fscal,cutoff_mask);
540 /* Calculate temporary vectorial force */
541 tx = _mm256_mul_ps(fscal,dx11);
542 ty = _mm256_mul_ps(fscal,dy11);
543 tz = _mm256_mul_ps(fscal,dz11);
545 /* Update vectorial force */
546 fix1 = _mm256_add_ps(fix1,tx);
547 fiy1 = _mm256_add_ps(fiy1,ty);
548 fiz1 = _mm256_add_ps(fiz1,tz);
550 fjx1 = _mm256_add_ps(fjx1,tx);
551 fjy1 = _mm256_add_ps(fjy1,ty);
552 fjz1 = _mm256_add_ps(fjz1,tz);
556 /**************************
557 * CALCULATE INTERACTIONS *
558 **************************/
560 if (gmx_mm256_any_lt(rsq12,rcutoff2))
563 r12 = _mm256_mul_ps(rsq12,rinv12);
565 /* EWALD ELECTROSTATICS */
567 /* Analytical PME correction */
568 zeta2 = _mm256_mul_ps(beta2,rsq12);
569 rinv3 = _mm256_mul_ps(rinvsq12,rinv12);
570 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
571 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
572 felec = _mm256_mul_ps(qq12,felec);
573 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
574 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
575 velec = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
576 velec = _mm256_mul_ps(qq12,velec);
578 cutoff_mask = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
580 /* Update potential sum for this i atom from the interaction with this j atom. */
581 velec = _mm256_and_ps(velec,cutoff_mask);
582 velecsum = _mm256_add_ps(velecsum,velec);
586 fscal = _mm256_and_ps(fscal,cutoff_mask);
588 /* Calculate temporary vectorial force */
589 tx = _mm256_mul_ps(fscal,dx12);
590 ty = _mm256_mul_ps(fscal,dy12);
591 tz = _mm256_mul_ps(fscal,dz12);
593 /* Update vectorial force */
594 fix1 = _mm256_add_ps(fix1,tx);
595 fiy1 = _mm256_add_ps(fiy1,ty);
596 fiz1 = _mm256_add_ps(fiz1,tz);
598 fjx2 = _mm256_add_ps(fjx2,tx);
599 fjy2 = _mm256_add_ps(fjy2,ty);
600 fjz2 = _mm256_add_ps(fjz2,tz);
604 /**************************
605 * CALCULATE INTERACTIONS *
606 **************************/
608 if (gmx_mm256_any_lt(rsq20,rcutoff2))
611 r20 = _mm256_mul_ps(rsq20,rinv20);
613 /* EWALD ELECTROSTATICS */
615 /* Analytical PME correction */
616 zeta2 = _mm256_mul_ps(beta2,rsq20);
617 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
618 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
619 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
620 felec = _mm256_mul_ps(qq20,felec);
621 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
622 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
623 velec = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
624 velec = _mm256_mul_ps(qq20,velec);
626 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
628 /* Update potential sum for this i atom from the interaction with this j atom. */
629 velec = _mm256_and_ps(velec,cutoff_mask);
630 velecsum = _mm256_add_ps(velecsum,velec);
634 fscal = _mm256_and_ps(fscal,cutoff_mask);
636 /* Calculate temporary vectorial force */
637 tx = _mm256_mul_ps(fscal,dx20);
638 ty = _mm256_mul_ps(fscal,dy20);
639 tz = _mm256_mul_ps(fscal,dz20);
641 /* Update vectorial force */
642 fix2 = _mm256_add_ps(fix2,tx);
643 fiy2 = _mm256_add_ps(fiy2,ty);
644 fiz2 = _mm256_add_ps(fiz2,tz);
646 fjx0 = _mm256_add_ps(fjx0,tx);
647 fjy0 = _mm256_add_ps(fjy0,ty);
648 fjz0 = _mm256_add_ps(fjz0,tz);
652 /**************************
653 * CALCULATE INTERACTIONS *
654 **************************/
656 if (gmx_mm256_any_lt(rsq21,rcutoff2))
659 r21 = _mm256_mul_ps(rsq21,rinv21);
661 /* EWALD ELECTROSTATICS */
663 /* Analytical PME correction */
664 zeta2 = _mm256_mul_ps(beta2,rsq21);
665 rinv3 = _mm256_mul_ps(rinvsq21,rinv21);
666 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
667 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
668 felec = _mm256_mul_ps(qq21,felec);
669 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
670 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
671 velec = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
672 velec = _mm256_mul_ps(qq21,velec);
674 cutoff_mask = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
676 /* Update potential sum for this i atom from the interaction with this j atom. */
677 velec = _mm256_and_ps(velec,cutoff_mask);
678 velecsum = _mm256_add_ps(velecsum,velec);
682 fscal = _mm256_and_ps(fscal,cutoff_mask);
684 /* Calculate temporary vectorial force */
685 tx = _mm256_mul_ps(fscal,dx21);
686 ty = _mm256_mul_ps(fscal,dy21);
687 tz = _mm256_mul_ps(fscal,dz21);
689 /* Update vectorial force */
690 fix2 = _mm256_add_ps(fix2,tx);
691 fiy2 = _mm256_add_ps(fiy2,ty);
692 fiz2 = _mm256_add_ps(fiz2,tz);
694 fjx1 = _mm256_add_ps(fjx1,tx);
695 fjy1 = _mm256_add_ps(fjy1,ty);
696 fjz1 = _mm256_add_ps(fjz1,tz);
700 /**************************
701 * CALCULATE INTERACTIONS *
702 **************************/
704 if (gmx_mm256_any_lt(rsq22,rcutoff2))
707 r22 = _mm256_mul_ps(rsq22,rinv22);
709 /* EWALD ELECTROSTATICS */
711 /* Analytical PME correction */
712 zeta2 = _mm256_mul_ps(beta2,rsq22);
713 rinv3 = _mm256_mul_ps(rinvsq22,rinv22);
714 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
715 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
716 felec = _mm256_mul_ps(qq22,felec);
717 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
718 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
719 velec = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
720 velec = _mm256_mul_ps(qq22,velec);
722 cutoff_mask = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
724 /* Update potential sum for this i atom from the interaction with this j atom. */
725 velec = _mm256_and_ps(velec,cutoff_mask);
726 velecsum = _mm256_add_ps(velecsum,velec);
730 fscal = _mm256_and_ps(fscal,cutoff_mask);
732 /* Calculate temporary vectorial force */
733 tx = _mm256_mul_ps(fscal,dx22);
734 ty = _mm256_mul_ps(fscal,dy22);
735 tz = _mm256_mul_ps(fscal,dz22);
737 /* Update vectorial force */
738 fix2 = _mm256_add_ps(fix2,tx);
739 fiy2 = _mm256_add_ps(fiy2,ty);
740 fiz2 = _mm256_add_ps(fiz2,tz);
742 fjx2 = _mm256_add_ps(fjx2,tx);
743 fjy2 = _mm256_add_ps(fjy2,ty);
744 fjz2 = _mm256_add_ps(fjz2,tz);
748 fjptrA = f+j_coord_offsetA;
749 fjptrB = f+j_coord_offsetB;
750 fjptrC = f+j_coord_offsetC;
751 fjptrD = f+j_coord_offsetD;
752 fjptrE = f+j_coord_offsetE;
753 fjptrF = f+j_coord_offsetF;
754 fjptrG = f+j_coord_offsetG;
755 fjptrH = f+j_coord_offsetH;
757 gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
758 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
760 /* Inner loop uses 981 flops */
766 /* Get j neighbor index, and coordinate index */
767 jnrlistA = jjnr[jidx];
768 jnrlistB = jjnr[jidx+1];
769 jnrlistC = jjnr[jidx+2];
770 jnrlistD = jjnr[jidx+3];
771 jnrlistE = jjnr[jidx+4];
772 jnrlistF = jjnr[jidx+5];
773 jnrlistG = jjnr[jidx+6];
774 jnrlistH = jjnr[jidx+7];
775 /* Sign of each element will be negative for non-real atoms.
776 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
777 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
779 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
780 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
782 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
783 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
784 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
785 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
786 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
787 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
788 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
789 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
790 j_coord_offsetA = DIM*jnrA;
791 j_coord_offsetB = DIM*jnrB;
792 j_coord_offsetC = DIM*jnrC;
793 j_coord_offsetD = DIM*jnrD;
794 j_coord_offsetE = DIM*jnrE;
795 j_coord_offsetF = DIM*jnrF;
796 j_coord_offsetG = DIM*jnrG;
797 j_coord_offsetH = DIM*jnrH;
799 /* load j atom coordinates */
800 gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
801 x+j_coord_offsetC,x+j_coord_offsetD,
802 x+j_coord_offsetE,x+j_coord_offsetF,
803 x+j_coord_offsetG,x+j_coord_offsetH,
804 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
806 /* Calculate displacement vector */
807 dx00 = _mm256_sub_ps(ix0,jx0);
808 dy00 = _mm256_sub_ps(iy0,jy0);
809 dz00 = _mm256_sub_ps(iz0,jz0);
810 dx01 = _mm256_sub_ps(ix0,jx1);
811 dy01 = _mm256_sub_ps(iy0,jy1);
812 dz01 = _mm256_sub_ps(iz0,jz1);
813 dx02 = _mm256_sub_ps(ix0,jx2);
814 dy02 = _mm256_sub_ps(iy0,jy2);
815 dz02 = _mm256_sub_ps(iz0,jz2);
816 dx10 = _mm256_sub_ps(ix1,jx0);
817 dy10 = _mm256_sub_ps(iy1,jy0);
818 dz10 = _mm256_sub_ps(iz1,jz0);
819 dx11 = _mm256_sub_ps(ix1,jx1);
820 dy11 = _mm256_sub_ps(iy1,jy1);
821 dz11 = _mm256_sub_ps(iz1,jz1);
822 dx12 = _mm256_sub_ps(ix1,jx2);
823 dy12 = _mm256_sub_ps(iy1,jy2);
824 dz12 = _mm256_sub_ps(iz1,jz2);
825 dx20 = _mm256_sub_ps(ix2,jx0);
826 dy20 = _mm256_sub_ps(iy2,jy0);
827 dz20 = _mm256_sub_ps(iz2,jz0);
828 dx21 = _mm256_sub_ps(ix2,jx1);
829 dy21 = _mm256_sub_ps(iy2,jy1);
830 dz21 = _mm256_sub_ps(iz2,jz1);
831 dx22 = _mm256_sub_ps(ix2,jx2);
832 dy22 = _mm256_sub_ps(iy2,jy2);
833 dz22 = _mm256_sub_ps(iz2,jz2);
835 /* Calculate squared distance and things based on it */
836 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
837 rsq01 = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
838 rsq02 = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
839 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
840 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
841 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
842 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
843 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
844 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
846 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
847 rinv01 = gmx_mm256_invsqrt_ps(rsq01);
848 rinv02 = gmx_mm256_invsqrt_ps(rsq02);
849 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
850 rinv11 = gmx_mm256_invsqrt_ps(rsq11);
851 rinv12 = gmx_mm256_invsqrt_ps(rsq12);
852 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
853 rinv21 = gmx_mm256_invsqrt_ps(rsq21);
854 rinv22 = gmx_mm256_invsqrt_ps(rsq22);
856 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
857 rinvsq01 = _mm256_mul_ps(rinv01,rinv01);
858 rinvsq02 = _mm256_mul_ps(rinv02,rinv02);
859 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
860 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
861 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
862 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
863 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
864 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
866 fjx0 = _mm256_setzero_ps();
867 fjy0 = _mm256_setzero_ps();
868 fjz0 = _mm256_setzero_ps();
869 fjx1 = _mm256_setzero_ps();
870 fjy1 = _mm256_setzero_ps();
871 fjz1 = _mm256_setzero_ps();
872 fjx2 = _mm256_setzero_ps();
873 fjy2 = _mm256_setzero_ps();
874 fjz2 = _mm256_setzero_ps();
876 /**************************
877 * CALCULATE INTERACTIONS *
878 **************************/
880 if (gmx_mm256_any_lt(rsq00,rcutoff2))
883 r00 = _mm256_mul_ps(rsq00,rinv00);
884 r00 = _mm256_andnot_ps(dummy_mask,r00);
886 /* EWALD ELECTROSTATICS */
888 /* Analytical PME correction */
889 zeta2 = _mm256_mul_ps(beta2,rsq00);
890 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
891 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
892 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
893 felec = _mm256_mul_ps(qq00,felec);
894 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
895 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
896 velec = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
897 velec = _mm256_mul_ps(qq00,velec);
899 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
901 /* Update potential sum for this i atom from the interaction with this j atom. */
902 velec = _mm256_and_ps(velec,cutoff_mask);
903 velec = _mm256_andnot_ps(dummy_mask,velec);
904 velecsum = _mm256_add_ps(velecsum,velec);
908 fscal = _mm256_and_ps(fscal,cutoff_mask);
910 fscal = _mm256_andnot_ps(dummy_mask,fscal);
912 /* Calculate temporary vectorial force */
913 tx = _mm256_mul_ps(fscal,dx00);
914 ty = _mm256_mul_ps(fscal,dy00);
915 tz = _mm256_mul_ps(fscal,dz00);
917 /* Update vectorial force */
918 fix0 = _mm256_add_ps(fix0,tx);
919 fiy0 = _mm256_add_ps(fiy0,ty);
920 fiz0 = _mm256_add_ps(fiz0,tz);
922 fjx0 = _mm256_add_ps(fjx0,tx);
923 fjy0 = _mm256_add_ps(fjy0,ty);
924 fjz0 = _mm256_add_ps(fjz0,tz);
928 /**************************
929 * CALCULATE INTERACTIONS *
930 **************************/
932 if (gmx_mm256_any_lt(rsq01,rcutoff2))
935 r01 = _mm256_mul_ps(rsq01,rinv01);
936 r01 = _mm256_andnot_ps(dummy_mask,r01);
938 /* EWALD ELECTROSTATICS */
940 /* Analytical PME correction */
941 zeta2 = _mm256_mul_ps(beta2,rsq01);
942 rinv3 = _mm256_mul_ps(rinvsq01,rinv01);
943 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
944 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
945 felec = _mm256_mul_ps(qq01,felec);
946 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
947 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
948 velec = _mm256_sub_ps(_mm256_sub_ps(rinv01,sh_ewald),pmecorrV);
949 velec = _mm256_mul_ps(qq01,velec);
951 cutoff_mask = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
953 /* Update potential sum for this i atom from the interaction with this j atom. */
954 velec = _mm256_and_ps(velec,cutoff_mask);
955 velec = _mm256_andnot_ps(dummy_mask,velec);
956 velecsum = _mm256_add_ps(velecsum,velec);
960 fscal = _mm256_and_ps(fscal,cutoff_mask);
962 fscal = _mm256_andnot_ps(dummy_mask,fscal);
964 /* Calculate temporary vectorial force */
965 tx = _mm256_mul_ps(fscal,dx01);
966 ty = _mm256_mul_ps(fscal,dy01);
967 tz = _mm256_mul_ps(fscal,dz01);
969 /* Update vectorial force */
970 fix0 = _mm256_add_ps(fix0,tx);
971 fiy0 = _mm256_add_ps(fiy0,ty);
972 fiz0 = _mm256_add_ps(fiz0,tz);
974 fjx1 = _mm256_add_ps(fjx1,tx);
975 fjy1 = _mm256_add_ps(fjy1,ty);
976 fjz1 = _mm256_add_ps(fjz1,tz);
980 /**************************
981 * CALCULATE INTERACTIONS *
982 **************************/
984 if (gmx_mm256_any_lt(rsq02,rcutoff2))
987 r02 = _mm256_mul_ps(rsq02,rinv02);
988 r02 = _mm256_andnot_ps(dummy_mask,r02);
990 /* EWALD ELECTROSTATICS */
992 /* Analytical PME correction */
993 zeta2 = _mm256_mul_ps(beta2,rsq02);
994 rinv3 = _mm256_mul_ps(rinvsq02,rinv02);
995 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
996 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
997 felec = _mm256_mul_ps(qq02,felec);
998 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
999 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1000 velec = _mm256_sub_ps(_mm256_sub_ps(rinv02,sh_ewald),pmecorrV);
1001 velec = _mm256_mul_ps(qq02,velec);
1003 cutoff_mask = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1005 /* Update potential sum for this i atom from the interaction with this j atom. */
1006 velec = _mm256_and_ps(velec,cutoff_mask);
1007 velec = _mm256_andnot_ps(dummy_mask,velec);
1008 velecsum = _mm256_add_ps(velecsum,velec);
1012 fscal = _mm256_and_ps(fscal,cutoff_mask);
1014 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1016 /* Calculate temporary vectorial force */
1017 tx = _mm256_mul_ps(fscal,dx02);
1018 ty = _mm256_mul_ps(fscal,dy02);
1019 tz = _mm256_mul_ps(fscal,dz02);
1021 /* Update vectorial force */
1022 fix0 = _mm256_add_ps(fix0,tx);
1023 fiy0 = _mm256_add_ps(fiy0,ty);
1024 fiz0 = _mm256_add_ps(fiz0,tz);
1026 fjx2 = _mm256_add_ps(fjx2,tx);
1027 fjy2 = _mm256_add_ps(fjy2,ty);
1028 fjz2 = _mm256_add_ps(fjz2,tz);
1032 /**************************
1033 * CALCULATE INTERACTIONS *
1034 **************************/
1036 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1039 r10 = _mm256_mul_ps(rsq10,rinv10);
1040 r10 = _mm256_andnot_ps(dummy_mask,r10);
1042 /* EWALD ELECTROSTATICS */
1044 /* Analytical PME correction */
1045 zeta2 = _mm256_mul_ps(beta2,rsq10);
1046 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
1047 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1048 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1049 felec = _mm256_mul_ps(qq10,felec);
1050 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
1051 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1052 velec = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
1053 velec = _mm256_mul_ps(qq10,velec);
1055 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1057 /* Update potential sum for this i atom from the interaction with this j atom. */
1058 velec = _mm256_and_ps(velec,cutoff_mask);
1059 velec = _mm256_andnot_ps(dummy_mask,velec);
1060 velecsum = _mm256_add_ps(velecsum,velec);
1064 fscal = _mm256_and_ps(fscal,cutoff_mask);
1066 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1068 /* Calculate temporary vectorial force */
1069 tx = _mm256_mul_ps(fscal,dx10);
1070 ty = _mm256_mul_ps(fscal,dy10);
1071 tz = _mm256_mul_ps(fscal,dz10);
1073 /* Update vectorial force */
1074 fix1 = _mm256_add_ps(fix1,tx);
1075 fiy1 = _mm256_add_ps(fiy1,ty);
1076 fiz1 = _mm256_add_ps(fiz1,tz);
1078 fjx0 = _mm256_add_ps(fjx0,tx);
1079 fjy0 = _mm256_add_ps(fjy0,ty);
1080 fjz0 = _mm256_add_ps(fjz0,tz);
1084 /**************************
1085 * CALCULATE INTERACTIONS *
1086 **************************/
1088 if (gmx_mm256_any_lt(rsq11,rcutoff2))
1091 r11 = _mm256_mul_ps(rsq11,rinv11);
1092 r11 = _mm256_andnot_ps(dummy_mask,r11);
1094 /* EWALD ELECTROSTATICS */
1096 /* Analytical PME correction */
1097 zeta2 = _mm256_mul_ps(beta2,rsq11);
1098 rinv3 = _mm256_mul_ps(rinvsq11,rinv11);
1099 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1100 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1101 felec = _mm256_mul_ps(qq11,felec);
1102 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
1103 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1104 velec = _mm256_sub_ps(_mm256_sub_ps(rinv11,sh_ewald),pmecorrV);
1105 velec = _mm256_mul_ps(qq11,velec);
1107 cutoff_mask = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1109 /* Update potential sum for this i atom from the interaction with this j atom. */
1110 velec = _mm256_and_ps(velec,cutoff_mask);
1111 velec = _mm256_andnot_ps(dummy_mask,velec);
1112 velecsum = _mm256_add_ps(velecsum,velec);
1116 fscal = _mm256_and_ps(fscal,cutoff_mask);
1118 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1120 /* Calculate temporary vectorial force */
1121 tx = _mm256_mul_ps(fscal,dx11);
1122 ty = _mm256_mul_ps(fscal,dy11);
1123 tz = _mm256_mul_ps(fscal,dz11);
1125 /* Update vectorial force */
1126 fix1 = _mm256_add_ps(fix1,tx);
1127 fiy1 = _mm256_add_ps(fiy1,ty);
1128 fiz1 = _mm256_add_ps(fiz1,tz);
1130 fjx1 = _mm256_add_ps(fjx1,tx);
1131 fjy1 = _mm256_add_ps(fjy1,ty);
1132 fjz1 = _mm256_add_ps(fjz1,tz);
1136 /**************************
1137 * CALCULATE INTERACTIONS *
1138 **************************/
1140 if (gmx_mm256_any_lt(rsq12,rcutoff2))
1143 r12 = _mm256_mul_ps(rsq12,rinv12);
1144 r12 = _mm256_andnot_ps(dummy_mask,r12);
1146 /* EWALD ELECTROSTATICS */
1148 /* Analytical PME correction */
1149 zeta2 = _mm256_mul_ps(beta2,rsq12);
1150 rinv3 = _mm256_mul_ps(rinvsq12,rinv12);
1151 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1152 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1153 felec = _mm256_mul_ps(qq12,felec);
1154 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
1155 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1156 velec = _mm256_sub_ps(_mm256_sub_ps(rinv12,sh_ewald),pmecorrV);
1157 velec = _mm256_mul_ps(qq12,velec);
1159 cutoff_mask = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1161 /* Update potential sum for this i atom from the interaction with this j atom. */
1162 velec = _mm256_and_ps(velec,cutoff_mask);
1163 velec = _mm256_andnot_ps(dummy_mask,velec);
1164 velecsum = _mm256_add_ps(velecsum,velec);
1168 fscal = _mm256_and_ps(fscal,cutoff_mask);
1170 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1172 /* Calculate temporary vectorial force */
1173 tx = _mm256_mul_ps(fscal,dx12);
1174 ty = _mm256_mul_ps(fscal,dy12);
1175 tz = _mm256_mul_ps(fscal,dz12);
1177 /* Update vectorial force */
1178 fix1 = _mm256_add_ps(fix1,tx);
1179 fiy1 = _mm256_add_ps(fiy1,ty);
1180 fiz1 = _mm256_add_ps(fiz1,tz);
1182 fjx2 = _mm256_add_ps(fjx2,tx);
1183 fjy2 = _mm256_add_ps(fjy2,ty);
1184 fjz2 = _mm256_add_ps(fjz2,tz);
1188 /**************************
1189 * CALCULATE INTERACTIONS *
1190 **************************/
1192 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1195 r20 = _mm256_mul_ps(rsq20,rinv20);
1196 r20 = _mm256_andnot_ps(dummy_mask,r20);
1198 /* EWALD ELECTROSTATICS */
1200 /* Analytical PME correction */
1201 zeta2 = _mm256_mul_ps(beta2,rsq20);
1202 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
1203 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1204 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1205 felec = _mm256_mul_ps(qq20,felec);
1206 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
1207 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1208 velec = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
1209 velec = _mm256_mul_ps(qq20,velec);
1211 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1213 /* Update potential sum for this i atom from the interaction with this j atom. */
1214 velec = _mm256_and_ps(velec,cutoff_mask);
1215 velec = _mm256_andnot_ps(dummy_mask,velec);
1216 velecsum = _mm256_add_ps(velecsum,velec);
1220 fscal = _mm256_and_ps(fscal,cutoff_mask);
1222 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1224 /* Calculate temporary vectorial force */
1225 tx = _mm256_mul_ps(fscal,dx20);
1226 ty = _mm256_mul_ps(fscal,dy20);
1227 tz = _mm256_mul_ps(fscal,dz20);
1229 /* Update vectorial force */
1230 fix2 = _mm256_add_ps(fix2,tx);
1231 fiy2 = _mm256_add_ps(fiy2,ty);
1232 fiz2 = _mm256_add_ps(fiz2,tz);
1234 fjx0 = _mm256_add_ps(fjx0,tx);
1235 fjy0 = _mm256_add_ps(fjy0,ty);
1236 fjz0 = _mm256_add_ps(fjz0,tz);
1240 /**************************
1241 * CALCULATE INTERACTIONS *
1242 **************************/
1244 if (gmx_mm256_any_lt(rsq21,rcutoff2))
1247 r21 = _mm256_mul_ps(rsq21,rinv21);
1248 r21 = _mm256_andnot_ps(dummy_mask,r21);
1250 /* EWALD ELECTROSTATICS */
1252 /* Analytical PME correction */
1253 zeta2 = _mm256_mul_ps(beta2,rsq21);
1254 rinv3 = _mm256_mul_ps(rinvsq21,rinv21);
1255 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1256 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1257 felec = _mm256_mul_ps(qq21,felec);
1258 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
1259 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1260 velec = _mm256_sub_ps(_mm256_sub_ps(rinv21,sh_ewald),pmecorrV);
1261 velec = _mm256_mul_ps(qq21,velec);
1263 cutoff_mask = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1265 /* Update potential sum for this i atom from the interaction with this j atom. */
1266 velec = _mm256_and_ps(velec,cutoff_mask);
1267 velec = _mm256_andnot_ps(dummy_mask,velec);
1268 velecsum = _mm256_add_ps(velecsum,velec);
1272 fscal = _mm256_and_ps(fscal,cutoff_mask);
1274 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1276 /* Calculate temporary vectorial force */
1277 tx = _mm256_mul_ps(fscal,dx21);
1278 ty = _mm256_mul_ps(fscal,dy21);
1279 tz = _mm256_mul_ps(fscal,dz21);
1281 /* Update vectorial force */
1282 fix2 = _mm256_add_ps(fix2,tx);
1283 fiy2 = _mm256_add_ps(fiy2,ty);
1284 fiz2 = _mm256_add_ps(fiz2,tz);
1286 fjx1 = _mm256_add_ps(fjx1,tx);
1287 fjy1 = _mm256_add_ps(fjy1,ty);
1288 fjz1 = _mm256_add_ps(fjz1,tz);
1292 /**************************
1293 * CALCULATE INTERACTIONS *
1294 **************************/
1296 if (gmx_mm256_any_lt(rsq22,rcutoff2))
1299 r22 = _mm256_mul_ps(rsq22,rinv22);
1300 r22 = _mm256_andnot_ps(dummy_mask,r22);
1302 /* EWALD ELECTROSTATICS */
1304 /* Analytical PME correction */
1305 zeta2 = _mm256_mul_ps(beta2,rsq22);
1306 rinv3 = _mm256_mul_ps(rinvsq22,rinv22);
1307 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1308 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1309 felec = _mm256_mul_ps(qq22,felec);
1310 pmecorrV = gmx_mm256_pmecorrV_ps(zeta2);
1311 pmecorrV = _mm256_mul_ps(pmecorrV,beta);
1312 velec = _mm256_sub_ps(_mm256_sub_ps(rinv22,sh_ewald),pmecorrV);
1313 velec = _mm256_mul_ps(qq22,velec);
1315 cutoff_mask = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1317 /* Update potential sum for this i atom from the interaction with this j atom. */
1318 velec = _mm256_and_ps(velec,cutoff_mask);
1319 velec = _mm256_andnot_ps(dummy_mask,velec);
1320 velecsum = _mm256_add_ps(velecsum,velec);
1324 fscal = _mm256_and_ps(fscal,cutoff_mask);
1326 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1328 /* Calculate temporary vectorial force */
1329 tx = _mm256_mul_ps(fscal,dx22);
1330 ty = _mm256_mul_ps(fscal,dy22);
1331 tz = _mm256_mul_ps(fscal,dz22);
1333 /* Update vectorial force */
1334 fix2 = _mm256_add_ps(fix2,tx);
1335 fiy2 = _mm256_add_ps(fiy2,ty);
1336 fiz2 = _mm256_add_ps(fiz2,tz);
1338 fjx2 = _mm256_add_ps(fjx2,tx);
1339 fjy2 = _mm256_add_ps(fjy2,ty);
1340 fjz2 = _mm256_add_ps(fjz2,tz);
1344 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1345 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1346 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1347 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1348 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1349 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1350 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1351 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1353 gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1354 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1356 /* Inner loop uses 990 flops */
1359 /* End of innermost loop */
1361 gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1362 f+i_coord_offset,fshift+i_shift_offset);
1365 /* Update potential energies */
1366 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1368 /* Increment number of inner iterations */
1369 inneriter += j_index_end - j_index_start;
1371 /* Outer loop uses 19 flops */
1374 /* Increment number of outer iterations */
1377 /* Update outer/inner flops */
1379 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*990);
1382 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_256_single
1383 * Electrostatics interaction: Ewald
1384 * VdW interaction: None
1385 * Geometry: Water3-Water3
1386 * Calculate force/pot: Force
1389 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_256_single
1390 (t_nblist * gmx_restrict nlist,
1391 rvec * gmx_restrict xx,
1392 rvec * gmx_restrict ff,
1393 t_forcerec * gmx_restrict fr,
1394 t_mdatoms * gmx_restrict mdatoms,
1395 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1396 t_nrnb * gmx_restrict nrnb)
1398 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1399 * just 0 for non-waters.
1400 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1401 * jnr indices corresponding to data put in the four positions in the SIMD register.
1403 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1404 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1405 int jnrA,jnrB,jnrC,jnrD;
1406 int jnrE,jnrF,jnrG,jnrH;
1407 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1408 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1409 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1410 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1411 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1412 real rcutoff_scalar;
1413 real *shiftvec,*fshift,*x,*f;
1414 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1415 real scratch[4*DIM];
1416 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1417 real * vdwioffsetptr0;
1418 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1419 real * vdwioffsetptr1;
1420 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1421 real * vdwioffsetptr2;
1422 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1423 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1424 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1425 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1426 __m256 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1427 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1428 __m256 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1429 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1430 __m256 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1431 __m256 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1432 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1433 __m256 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1434 __m256 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1435 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1436 __m256 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1437 __m256 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1438 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
1441 __m128i ewitab_lo,ewitab_hi;
1442 __m256 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1443 __m256 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1445 __m256 dummy_mask,cutoff_mask;
1446 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1447 __m256 one = _mm256_set1_ps(1.0);
1448 __m256 two = _mm256_set1_ps(2.0);
1454 jindex = nlist->jindex;
1456 shiftidx = nlist->shift;
1458 shiftvec = fr->shift_vec[0];
1459 fshift = fr->fshift[0];
1460 facel = _mm256_set1_ps(fr->epsfac);
1461 charge = mdatoms->chargeA;
1463 sh_ewald = _mm256_set1_ps(fr->ic->sh_ewald);
1464 beta = _mm256_set1_ps(fr->ic->ewaldcoeff);
1465 beta2 = _mm256_mul_ps(beta,beta);
1466 beta3 = _mm256_mul_ps(beta,beta2);
1468 ewtab = fr->ic->tabq_coul_F;
1469 ewtabscale = _mm256_set1_ps(fr->ic->tabq_scale);
1470 ewtabhalfspace = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1472 /* Setup water-specific parameters */
1473 inr = nlist->iinr[0];
1474 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1475 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1476 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1478 jq0 = _mm256_set1_ps(charge[inr+0]);
1479 jq1 = _mm256_set1_ps(charge[inr+1]);
1480 jq2 = _mm256_set1_ps(charge[inr+2]);
1481 qq00 = _mm256_mul_ps(iq0,jq0);
1482 qq01 = _mm256_mul_ps(iq0,jq1);
1483 qq02 = _mm256_mul_ps(iq0,jq2);
1484 qq10 = _mm256_mul_ps(iq1,jq0);
1485 qq11 = _mm256_mul_ps(iq1,jq1);
1486 qq12 = _mm256_mul_ps(iq1,jq2);
1487 qq20 = _mm256_mul_ps(iq2,jq0);
1488 qq21 = _mm256_mul_ps(iq2,jq1);
1489 qq22 = _mm256_mul_ps(iq2,jq2);
1491 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1492 rcutoff_scalar = fr->rcoulomb;
1493 rcutoff = _mm256_set1_ps(rcutoff_scalar);
1494 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
1496 /* Avoid stupid compiler warnings */
1497 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1498 j_coord_offsetA = 0;
1499 j_coord_offsetB = 0;
1500 j_coord_offsetC = 0;
1501 j_coord_offsetD = 0;
1502 j_coord_offsetE = 0;
1503 j_coord_offsetF = 0;
1504 j_coord_offsetG = 0;
1505 j_coord_offsetH = 0;
1510 for(iidx=0;iidx<4*DIM;iidx++)
1512 scratch[iidx] = 0.0;
1515 /* Start outer loop over neighborlists */
1516 for(iidx=0; iidx<nri; iidx++)
1518 /* Load shift vector for this list */
1519 i_shift_offset = DIM*shiftidx[iidx];
1521 /* Load limits for loop over neighbors */
1522 j_index_start = jindex[iidx];
1523 j_index_end = jindex[iidx+1];
1525 /* Get outer coordinate index */
1527 i_coord_offset = DIM*inr;
1529 /* Load i particle coords and add shift vector */
1530 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1531 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1533 fix0 = _mm256_setzero_ps();
1534 fiy0 = _mm256_setzero_ps();
1535 fiz0 = _mm256_setzero_ps();
1536 fix1 = _mm256_setzero_ps();
1537 fiy1 = _mm256_setzero_ps();
1538 fiz1 = _mm256_setzero_ps();
1539 fix2 = _mm256_setzero_ps();
1540 fiy2 = _mm256_setzero_ps();
1541 fiz2 = _mm256_setzero_ps();
1543 /* Start inner kernel loop */
1544 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1547 /* Get j neighbor index, and coordinate index */
1549 jnrB = jjnr[jidx+1];
1550 jnrC = jjnr[jidx+2];
1551 jnrD = jjnr[jidx+3];
1552 jnrE = jjnr[jidx+4];
1553 jnrF = jjnr[jidx+5];
1554 jnrG = jjnr[jidx+6];
1555 jnrH = jjnr[jidx+7];
1556 j_coord_offsetA = DIM*jnrA;
1557 j_coord_offsetB = DIM*jnrB;
1558 j_coord_offsetC = DIM*jnrC;
1559 j_coord_offsetD = DIM*jnrD;
1560 j_coord_offsetE = DIM*jnrE;
1561 j_coord_offsetF = DIM*jnrF;
1562 j_coord_offsetG = DIM*jnrG;
1563 j_coord_offsetH = DIM*jnrH;
1565 /* load j atom coordinates */
1566 gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1567 x+j_coord_offsetC,x+j_coord_offsetD,
1568 x+j_coord_offsetE,x+j_coord_offsetF,
1569 x+j_coord_offsetG,x+j_coord_offsetH,
1570 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1572 /* Calculate displacement vector */
1573 dx00 = _mm256_sub_ps(ix0,jx0);
1574 dy00 = _mm256_sub_ps(iy0,jy0);
1575 dz00 = _mm256_sub_ps(iz0,jz0);
1576 dx01 = _mm256_sub_ps(ix0,jx1);
1577 dy01 = _mm256_sub_ps(iy0,jy1);
1578 dz01 = _mm256_sub_ps(iz0,jz1);
1579 dx02 = _mm256_sub_ps(ix0,jx2);
1580 dy02 = _mm256_sub_ps(iy0,jy2);
1581 dz02 = _mm256_sub_ps(iz0,jz2);
1582 dx10 = _mm256_sub_ps(ix1,jx0);
1583 dy10 = _mm256_sub_ps(iy1,jy0);
1584 dz10 = _mm256_sub_ps(iz1,jz0);
1585 dx11 = _mm256_sub_ps(ix1,jx1);
1586 dy11 = _mm256_sub_ps(iy1,jy1);
1587 dz11 = _mm256_sub_ps(iz1,jz1);
1588 dx12 = _mm256_sub_ps(ix1,jx2);
1589 dy12 = _mm256_sub_ps(iy1,jy2);
1590 dz12 = _mm256_sub_ps(iz1,jz2);
1591 dx20 = _mm256_sub_ps(ix2,jx0);
1592 dy20 = _mm256_sub_ps(iy2,jy0);
1593 dz20 = _mm256_sub_ps(iz2,jz0);
1594 dx21 = _mm256_sub_ps(ix2,jx1);
1595 dy21 = _mm256_sub_ps(iy2,jy1);
1596 dz21 = _mm256_sub_ps(iz2,jz1);
1597 dx22 = _mm256_sub_ps(ix2,jx2);
1598 dy22 = _mm256_sub_ps(iy2,jy2);
1599 dz22 = _mm256_sub_ps(iz2,jz2);
1601 /* Calculate squared distance and things based on it */
1602 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1603 rsq01 = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1604 rsq02 = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1605 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1606 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1607 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1608 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1609 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1610 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1612 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
1613 rinv01 = gmx_mm256_invsqrt_ps(rsq01);
1614 rinv02 = gmx_mm256_invsqrt_ps(rsq02);
1615 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
1616 rinv11 = gmx_mm256_invsqrt_ps(rsq11);
1617 rinv12 = gmx_mm256_invsqrt_ps(rsq12);
1618 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
1619 rinv21 = gmx_mm256_invsqrt_ps(rsq21);
1620 rinv22 = gmx_mm256_invsqrt_ps(rsq22);
1622 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
1623 rinvsq01 = _mm256_mul_ps(rinv01,rinv01);
1624 rinvsq02 = _mm256_mul_ps(rinv02,rinv02);
1625 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1626 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
1627 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
1628 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1629 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
1630 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
1632 fjx0 = _mm256_setzero_ps();
1633 fjy0 = _mm256_setzero_ps();
1634 fjz0 = _mm256_setzero_ps();
1635 fjx1 = _mm256_setzero_ps();
1636 fjy1 = _mm256_setzero_ps();
1637 fjz1 = _mm256_setzero_ps();
1638 fjx2 = _mm256_setzero_ps();
1639 fjy2 = _mm256_setzero_ps();
1640 fjz2 = _mm256_setzero_ps();
1642 /**************************
1643 * CALCULATE INTERACTIONS *
1644 **************************/
1646 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1649 r00 = _mm256_mul_ps(rsq00,rinv00);
1651 /* EWALD ELECTROSTATICS */
1653 /* Analytical PME correction */
1654 zeta2 = _mm256_mul_ps(beta2,rsq00);
1655 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
1656 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1657 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1658 felec = _mm256_mul_ps(qq00,felec);
1660 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1664 fscal = _mm256_and_ps(fscal,cutoff_mask);
1666 /* Calculate temporary vectorial force */
1667 tx = _mm256_mul_ps(fscal,dx00);
1668 ty = _mm256_mul_ps(fscal,dy00);
1669 tz = _mm256_mul_ps(fscal,dz00);
1671 /* Update vectorial force */
1672 fix0 = _mm256_add_ps(fix0,tx);
1673 fiy0 = _mm256_add_ps(fiy0,ty);
1674 fiz0 = _mm256_add_ps(fiz0,tz);
1676 fjx0 = _mm256_add_ps(fjx0,tx);
1677 fjy0 = _mm256_add_ps(fjy0,ty);
1678 fjz0 = _mm256_add_ps(fjz0,tz);
1682 /**************************
1683 * CALCULATE INTERACTIONS *
1684 **************************/
1686 if (gmx_mm256_any_lt(rsq01,rcutoff2))
1689 r01 = _mm256_mul_ps(rsq01,rinv01);
1691 /* EWALD ELECTROSTATICS */
1693 /* Analytical PME correction */
1694 zeta2 = _mm256_mul_ps(beta2,rsq01);
1695 rinv3 = _mm256_mul_ps(rinvsq01,rinv01);
1696 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1697 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1698 felec = _mm256_mul_ps(qq01,felec);
1700 cutoff_mask = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1704 fscal = _mm256_and_ps(fscal,cutoff_mask);
1706 /* Calculate temporary vectorial force */
1707 tx = _mm256_mul_ps(fscal,dx01);
1708 ty = _mm256_mul_ps(fscal,dy01);
1709 tz = _mm256_mul_ps(fscal,dz01);
1711 /* Update vectorial force */
1712 fix0 = _mm256_add_ps(fix0,tx);
1713 fiy0 = _mm256_add_ps(fiy0,ty);
1714 fiz0 = _mm256_add_ps(fiz0,tz);
1716 fjx1 = _mm256_add_ps(fjx1,tx);
1717 fjy1 = _mm256_add_ps(fjy1,ty);
1718 fjz1 = _mm256_add_ps(fjz1,tz);
1722 /**************************
1723 * CALCULATE INTERACTIONS *
1724 **************************/
1726 if (gmx_mm256_any_lt(rsq02,rcutoff2))
1729 r02 = _mm256_mul_ps(rsq02,rinv02);
1731 /* EWALD ELECTROSTATICS */
1733 /* Analytical PME correction */
1734 zeta2 = _mm256_mul_ps(beta2,rsq02);
1735 rinv3 = _mm256_mul_ps(rinvsq02,rinv02);
1736 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1737 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1738 felec = _mm256_mul_ps(qq02,felec);
1740 cutoff_mask = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1744 fscal = _mm256_and_ps(fscal,cutoff_mask);
1746 /* Calculate temporary vectorial force */
1747 tx = _mm256_mul_ps(fscal,dx02);
1748 ty = _mm256_mul_ps(fscal,dy02);
1749 tz = _mm256_mul_ps(fscal,dz02);
1751 /* Update vectorial force */
1752 fix0 = _mm256_add_ps(fix0,tx);
1753 fiy0 = _mm256_add_ps(fiy0,ty);
1754 fiz0 = _mm256_add_ps(fiz0,tz);
1756 fjx2 = _mm256_add_ps(fjx2,tx);
1757 fjy2 = _mm256_add_ps(fjy2,ty);
1758 fjz2 = _mm256_add_ps(fjz2,tz);
1762 /**************************
1763 * CALCULATE INTERACTIONS *
1764 **************************/
1766 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1769 r10 = _mm256_mul_ps(rsq10,rinv10);
1771 /* EWALD ELECTROSTATICS */
1773 /* Analytical PME correction */
1774 zeta2 = _mm256_mul_ps(beta2,rsq10);
1775 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
1776 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1777 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1778 felec = _mm256_mul_ps(qq10,felec);
1780 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1784 fscal = _mm256_and_ps(fscal,cutoff_mask);
1786 /* Calculate temporary vectorial force */
1787 tx = _mm256_mul_ps(fscal,dx10);
1788 ty = _mm256_mul_ps(fscal,dy10);
1789 tz = _mm256_mul_ps(fscal,dz10);
1791 /* Update vectorial force */
1792 fix1 = _mm256_add_ps(fix1,tx);
1793 fiy1 = _mm256_add_ps(fiy1,ty);
1794 fiz1 = _mm256_add_ps(fiz1,tz);
1796 fjx0 = _mm256_add_ps(fjx0,tx);
1797 fjy0 = _mm256_add_ps(fjy0,ty);
1798 fjz0 = _mm256_add_ps(fjz0,tz);
1802 /**************************
1803 * CALCULATE INTERACTIONS *
1804 **************************/
1806 if (gmx_mm256_any_lt(rsq11,rcutoff2))
1809 r11 = _mm256_mul_ps(rsq11,rinv11);
1811 /* EWALD ELECTROSTATICS */
1813 /* Analytical PME correction */
1814 zeta2 = _mm256_mul_ps(beta2,rsq11);
1815 rinv3 = _mm256_mul_ps(rinvsq11,rinv11);
1816 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1817 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1818 felec = _mm256_mul_ps(qq11,felec);
1820 cutoff_mask = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1824 fscal = _mm256_and_ps(fscal,cutoff_mask);
1826 /* Calculate temporary vectorial force */
1827 tx = _mm256_mul_ps(fscal,dx11);
1828 ty = _mm256_mul_ps(fscal,dy11);
1829 tz = _mm256_mul_ps(fscal,dz11);
1831 /* Update vectorial force */
1832 fix1 = _mm256_add_ps(fix1,tx);
1833 fiy1 = _mm256_add_ps(fiy1,ty);
1834 fiz1 = _mm256_add_ps(fiz1,tz);
1836 fjx1 = _mm256_add_ps(fjx1,tx);
1837 fjy1 = _mm256_add_ps(fjy1,ty);
1838 fjz1 = _mm256_add_ps(fjz1,tz);
1842 /**************************
1843 * CALCULATE INTERACTIONS *
1844 **************************/
1846 if (gmx_mm256_any_lt(rsq12,rcutoff2))
1849 r12 = _mm256_mul_ps(rsq12,rinv12);
1851 /* EWALD ELECTROSTATICS */
1853 /* Analytical PME correction */
1854 zeta2 = _mm256_mul_ps(beta2,rsq12);
1855 rinv3 = _mm256_mul_ps(rinvsq12,rinv12);
1856 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1857 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1858 felec = _mm256_mul_ps(qq12,felec);
1860 cutoff_mask = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1864 fscal = _mm256_and_ps(fscal,cutoff_mask);
1866 /* Calculate temporary vectorial force */
1867 tx = _mm256_mul_ps(fscal,dx12);
1868 ty = _mm256_mul_ps(fscal,dy12);
1869 tz = _mm256_mul_ps(fscal,dz12);
1871 /* Update vectorial force */
1872 fix1 = _mm256_add_ps(fix1,tx);
1873 fiy1 = _mm256_add_ps(fiy1,ty);
1874 fiz1 = _mm256_add_ps(fiz1,tz);
1876 fjx2 = _mm256_add_ps(fjx2,tx);
1877 fjy2 = _mm256_add_ps(fjy2,ty);
1878 fjz2 = _mm256_add_ps(fjz2,tz);
1882 /**************************
1883 * CALCULATE INTERACTIONS *
1884 **************************/
1886 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1889 r20 = _mm256_mul_ps(rsq20,rinv20);
1891 /* EWALD ELECTROSTATICS */
1893 /* Analytical PME correction */
1894 zeta2 = _mm256_mul_ps(beta2,rsq20);
1895 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
1896 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1897 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1898 felec = _mm256_mul_ps(qq20,felec);
1900 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1904 fscal = _mm256_and_ps(fscal,cutoff_mask);
1906 /* Calculate temporary vectorial force */
1907 tx = _mm256_mul_ps(fscal,dx20);
1908 ty = _mm256_mul_ps(fscal,dy20);
1909 tz = _mm256_mul_ps(fscal,dz20);
1911 /* Update vectorial force */
1912 fix2 = _mm256_add_ps(fix2,tx);
1913 fiy2 = _mm256_add_ps(fiy2,ty);
1914 fiz2 = _mm256_add_ps(fiz2,tz);
1916 fjx0 = _mm256_add_ps(fjx0,tx);
1917 fjy0 = _mm256_add_ps(fjy0,ty);
1918 fjz0 = _mm256_add_ps(fjz0,tz);
1922 /**************************
1923 * CALCULATE INTERACTIONS *
1924 **************************/
1926 if (gmx_mm256_any_lt(rsq21,rcutoff2))
1929 r21 = _mm256_mul_ps(rsq21,rinv21);
1931 /* EWALD ELECTROSTATICS */
1933 /* Analytical PME correction */
1934 zeta2 = _mm256_mul_ps(beta2,rsq21);
1935 rinv3 = _mm256_mul_ps(rinvsq21,rinv21);
1936 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1937 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1938 felec = _mm256_mul_ps(qq21,felec);
1940 cutoff_mask = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1944 fscal = _mm256_and_ps(fscal,cutoff_mask);
1946 /* Calculate temporary vectorial force */
1947 tx = _mm256_mul_ps(fscal,dx21);
1948 ty = _mm256_mul_ps(fscal,dy21);
1949 tz = _mm256_mul_ps(fscal,dz21);
1951 /* Update vectorial force */
1952 fix2 = _mm256_add_ps(fix2,tx);
1953 fiy2 = _mm256_add_ps(fiy2,ty);
1954 fiz2 = _mm256_add_ps(fiz2,tz);
1956 fjx1 = _mm256_add_ps(fjx1,tx);
1957 fjy1 = _mm256_add_ps(fjy1,ty);
1958 fjz1 = _mm256_add_ps(fjz1,tz);
1962 /**************************
1963 * CALCULATE INTERACTIONS *
1964 **************************/
1966 if (gmx_mm256_any_lt(rsq22,rcutoff2))
1969 r22 = _mm256_mul_ps(rsq22,rinv22);
1971 /* EWALD ELECTROSTATICS */
1973 /* Analytical PME correction */
1974 zeta2 = _mm256_mul_ps(beta2,rsq22);
1975 rinv3 = _mm256_mul_ps(rinvsq22,rinv22);
1976 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
1977 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1978 felec = _mm256_mul_ps(qq22,felec);
1980 cutoff_mask = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1984 fscal = _mm256_and_ps(fscal,cutoff_mask);
1986 /* Calculate temporary vectorial force */
1987 tx = _mm256_mul_ps(fscal,dx22);
1988 ty = _mm256_mul_ps(fscal,dy22);
1989 tz = _mm256_mul_ps(fscal,dz22);
1991 /* Update vectorial force */
1992 fix2 = _mm256_add_ps(fix2,tx);
1993 fiy2 = _mm256_add_ps(fiy2,ty);
1994 fiz2 = _mm256_add_ps(fiz2,tz);
1996 fjx2 = _mm256_add_ps(fjx2,tx);
1997 fjy2 = _mm256_add_ps(fjy2,ty);
1998 fjz2 = _mm256_add_ps(fjz2,tz);
2002 fjptrA = f+j_coord_offsetA;
2003 fjptrB = f+j_coord_offsetB;
2004 fjptrC = f+j_coord_offsetC;
2005 fjptrD = f+j_coord_offsetD;
2006 fjptrE = f+j_coord_offsetE;
2007 fjptrF = f+j_coord_offsetF;
2008 fjptrG = f+j_coord_offsetG;
2009 fjptrH = f+j_coord_offsetH;
2011 gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2012 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2014 /* Inner loop uses 531 flops */
2017 if(jidx<j_index_end)
2020 /* Get j neighbor index, and coordinate index */
2021 jnrlistA = jjnr[jidx];
2022 jnrlistB = jjnr[jidx+1];
2023 jnrlistC = jjnr[jidx+2];
2024 jnrlistD = jjnr[jidx+3];
2025 jnrlistE = jjnr[jidx+4];
2026 jnrlistF = jjnr[jidx+5];
2027 jnrlistG = jjnr[jidx+6];
2028 jnrlistH = jjnr[jidx+7];
2029 /* Sign of each element will be negative for non-real atoms.
2030 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2031 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2033 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2034 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2036 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
2037 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
2038 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
2039 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
2040 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
2041 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
2042 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
2043 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
2044 j_coord_offsetA = DIM*jnrA;
2045 j_coord_offsetB = DIM*jnrB;
2046 j_coord_offsetC = DIM*jnrC;
2047 j_coord_offsetD = DIM*jnrD;
2048 j_coord_offsetE = DIM*jnrE;
2049 j_coord_offsetF = DIM*jnrF;
2050 j_coord_offsetG = DIM*jnrG;
2051 j_coord_offsetH = DIM*jnrH;
2053 /* load j atom coordinates */
2054 gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2055 x+j_coord_offsetC,x+j_coord_offsetD,
2056 x+j_coord_offsetE,x+j_coord_offsetF,
2057 x+j_coord_offsetG,x+j_coord_offsetH,
2058 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2060 /* Calculate displacement vector */
2061 dx00 = _mm256_sub_ps(ix0,jx0);
2062 dy00 = _mm256_sub_ps(iy0,jy0);
2063 dz00 = _mm256_sub_ps(iz0,jz0);
2064 dx01 = _mm256_sub_ps(ix0,jx1);
2065 dy01 = _mm256_sub_ps(iy0,jy1);
2066 dz01 = _mm256_sub_ps(iz0,jz1);
2067 dx02 = _mm256_sub_ps(ix0,jx2);
2068 dy02 = _mm256_sub_ps(iy0,jy2);
2069 dz02 = _mm256_sub_ps(iz0,jz2);
2070 dx10 = _mm256_sub_ps(ix1,jx0);
2071 dy10 = _mm256_sub_ps(iy1,jy0);
2072 dz10 = _mm256_sub_ps(iz1,jz0);
2073 dx11 = _mm256_sub_ps(ix1,jx1);
2074 dy11 = _mm256_sub_ps(iy1,jy1);
2075 dz11 = _mm256_sub_ps(iz1,jz1);
2076 dx12 = _mm256_sub_ps(ix1,jx2);
2077 dy12 = _mm256_sub_ps(iy1,jy2);
2078 dz12 = _mm256_sub_ps(iz1,jz2);
2079 dx20 = _mm256_sub_ps(ix2,jx0);
2080 dy20 = _mm256_sub_ps(iy2,jy0);
2081 dz20 = _mm256_sub_ps(iz2,jz0);
2082 dx21 = _mm256_sub_ps(ix2,jx1);
2083 dy21 = _mm256_sub_ps(iy2,jy1);
2084 dz21 = _mm256_sub_ps(iz2,jz1);
2085 dx22 = _mm256_sub_ps(ix2,jx2);
2086 dy22 = _mm256_sub_ps(iy2,jy2);
2087 dz22 = _mm256_sub_ps(iz2,jz2);
2089 /* Calculate squared distance and things based on it */
2090 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2091 rsq01 = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
2092 rsq02 = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
2093 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
2094 rsq11 = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2095 rsq12 = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2096 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
2097 rsq21 = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2098 rsq22 = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2100 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
2101 rinv01 = gmx_mm256_invsqrt_ps(rsq01);
2102 rinv02 = gmx_mm256_invsqrt_ps(rsq02);
2103 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
2104 rinv11 = gmx_mm256_invsqrt_ps(rsq11);
2105 rinv12 = gmx_mm256_invsqrt_ps(rsq12);
2106 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
2107 rinv21 = gmx_mm256_invsqrt_ps(rsq21);
2108 rinv22 = gmx_mm256_invsqrt_ps(rsq22);
2110 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
2111 rinvsq01 = _mm256_mul_ps(rinv01,rinv01);
2112 rinvsq02 = _mm256_mul_ps(rinv02,rinv02);
2113 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
2114 rinvsq11 = _mm256_mul_ps(rinv11,rinv11);
2115 rinvsq12 = _mm256_mul_ps(rinv12,rinv12);
2116 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
2117 rinvsq21 = _mm256_mul_ps(rinv21,rinv21);
2118 rinvsq22 = _mm256_mul_ps(rinv22,rinv22);
2120 fjx0 = _mm256_setzero_ps();
2121 fjy0 = _mm256_setzero_ps();
2122 fjz0 = _mm256_setzero_ps();
2123 fjx1 = _mm256_setzero_ps();
2124 fjy1 = _mm256_setzero_ps();
2125 fjz1 = _mm256_setzero_ps();
2126 fjx2 = _mm256_setzero_ps();
2127 fjy2 = _mm256_setzero_ps();
2128 fjz2 = _mm256_setzero_ps();
2130 /**************************
2131 * CALCULATE INTERACTIONS *
2132 **************************/
2134 if (gmx_mm256_any_lt(rsq00,rcutoff2))
2137 r00 = _mm256_mul_ps(rsq00,rinv00);
2138 r00 = _mm256_andnot_ps(dummy_mask,r00);
2140 /* EWALD ELECTROSTATICS */
2142 /* Analytical PME correction */
2143 zeta2 = _mm256_mul_ps(beta2,rsq00);
2144 rinv3 = _mm256_mul_ps(rinvsq00,rinv00);
2145 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
2146 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2147 felec = _mm256_mul_ps(qq00,felec);
2149 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
2153 fscal = _mm256_and_ps(fscal,cutoff_mask);
2155 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2157 /* Calculate temporary vectorial force */
2158 tx = _mm256_mul_ps(fscal,dx00);
2159 ty = _mm256_mul_ps(fscal,dy00);
2160 tz = _mm256_mul_ps(fscal,dz00);
2162 /* Update vectorial force */
2163 fix0 = _mm256_add_ps(fix0,tx);
2164 fiy0 = _mm256_add_ps(fiy0,ty);
2165 fiz0 = _mm256_add_ps(fiz0,tz);
2167 fjx0 = _mm256_add_ps(fjx0,tx);
2168 fjy0 = _mm256_add_ps(fjy0,ty);
2169 fjz0 = _mm256_add_ps(fjz0,tz);
2173 /**************************
2174 * CALCULATE INTERACTIONS *
2175 **************************/
2177 if (gmx_mm256_any_lt(rsq01,rcutoff2))
2180 r01 = _mm256_mul_ps(rsq01,rinv01);
2181 r01 = _mm256_andnot_ps(dummy_mask,r01);
2183 /* EWALD ELECTROSTATICS */
2185 /* Analytical PME correction */
2186 zeta2 = _mm256_mul_ps(beta2,rsq01);
2187 rinv3 = _mm256_mul_ps(rinvsq01,rinv01);
2188 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
2189 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2190 felec = _mm256_mul_ps(qq01,felec);
2192 cutoff_mask = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
2196 fscal = _mm256_and_ps(fscal,cutoff_mask);
2198 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2200 /* Calculate temporary vectorial force */
2201 tx = _mm256_mul_ps(fscal,dx01);
2202 ty = _mm256_mul_ps(fscal,dy01);
2203 tz = _mm256_mul_ps(fscal,dz01);
2205 /* Update vectorial force */
2206 fix0 = _mm256_add_ps(fix0,tx);
2207 fiy0 = _mm256_add_ps(fiy0,ty);
2208 fiz0 = _mm256_add_ps(fiz0,tz);
2210 fjx1 = _mm256_add_ps(fjx1,tx);
2211 fjy1 = _mm256_add_ps(fjy1,ty);
2212 fjz1 = _mm256_add_ps(fjz1,tz);
2216 /**************************
2217 * CALCULATE INTERACTIONS *
2218 **************************/
2220 if (gmx_mm256_any_lt(rsq02,rcutoff2))
2223 r02 = _mm256_mul_ps(rsq02,rinv02);
2224 r02 = _mm256_andnot_ps(dummy_mask,r02);
2226 /* EWALD ELECTROSTATICS */
2228 /* Analytical PME correction */
2229 zeta2 = _mm256_mul_ps(beta2,rsq02);
2230 rinv3 = _mm256_mul_ps(rinvsq02,rinv02);
2231 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
2232 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2233 felec = _mm256_mul_ps(qq02,felec);
2235 cutoff_mask = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
2239 fscal = _mm256_and_ps(fscal,cutoff_mask);
2241 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2243 /* Calculate temporary vectorial force */
2244 tx = _mm256_mul_ps(fscal,dx02);
2245 ty = _mm256_mul_ps(fscal,dy02);
2246 tz = _mm256_mul_ps(fscal,dz02);
2248 /* Update vectorial force */
2249 fix0 = _mm256_add_ps(fix0,tx);
2250 fiy0 = _mm256_add_ps(fiy0,ty);
2251 fiz0 = _mm256_add_ps(fiz0,tz);
2253 fjx2 = _mm256_add_ps(fjx2,tx);
2254 fjy2 = _mm256_add_ps(fjy2,ty);
2255 fjz2 = _mm256_add_ps(fjz2,tz);
2259 /**************************
2260 * CALCULATE INTERACTIONS *
2261 **************************/
2263 if (gmx_mm256_any_lt(rsq10,rcutoff2))
2266 r10 = _mm256_mul_ps(rsq10,rinv10);
2267 r10 = _mm256_andnot_ps(dummy_mask,r10);
2269 /* EWALD ELECTROSTATICS */
2271 /* Analytical PME correction */
2272 zeta2 = _mm256_mul_ps(beta2,rsq10);
2273 rinv3 = _mm256_mul_ps(rinvsq10,rinv10);
2274 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
2275 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2276 felec = _mm256_mul_ps(qq10,felec);
2278 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
2282 fscal = _mm256_and_ps(fscal,cutoff_mask);
2284 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2286 /* Calculate temporary vectorial force */
2287 tx = _mm256_mul_ps(fscal,dx10);
2288 ty = _mm256_mul_ps(fscal,dy10);
2289 tz = _mm256_mul_ps(fscal,dz10);
2291 /* Update vectorial force */
2292 fix1 = _mm256_add_ps(fix1,tx);
2293 fiy1 = _mm256_add_ps(fiy1,ty);
2294 fiz1 = _mm256_add_ps(fiz1,tz);
2296 fjx0 = _mm256_add_ps(fjx0,tx);
2297 fjy0 = _mm256_add_ps(fjy0,ty);
2298 fjz0 = _mm256_add_ps(fjz0,tz);
2302 /**************************
2303 * CALCULATE INTERACTIONS *
2304 **************************/
2306 if (gmx_mm256_any_lt(rsq11,rcutoff2))
2309 r11 = _mm256_mul_ps(rsq11,rinv11);
2310 r11 = _mm256_andnot_ps(dummy_mask,r11);
2312 /* EWALD ELECTROSTATICS */
2314 /* Analytical PME correction */
2315 zeta2 = _mm256_mul_ps(beta2,rsq11);
2316 rinv3 = _mm256_mul_ps(rinvsq11,rinv11);
2317 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
2318 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2319 felec = _mm256_mul_ps(qq11,felec);
2321 cutoff_mask = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
2325 fscal = _mm256_and_ps(fscal,cutoff_mask);
2327 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2329 /* Calculate temporary vectorial force */
2330 tx = _mm256_mul_ps(fscal,dx11);
2331 ty = _mm256_mul_ps(fscal,dy11);
2332 tz = _mm256_mul_ps(fscal,dz11);
2334 /* Update vectorial force */
2335 fix1 = _mm256_add_ps(fix1,tx);
2336 fiy1 = _mm256_add_ps(fiy1,ty);
2337 fiz1 = _mm256_add_ps(fiz1,tz);
2339 fjx1 = _mm256_add_ps(fjx1,tx);
2340 fjy1 = _mm256_add_ps(fjy1,ty);
2341 fjz1 = _mm256_add_ps(fjz1,tz);
2345 /**************************
2346 * CALCULATE INTERACTIONS *
2347 **************************/
2349 if (gmx_mm256_any_lt(rsq12,rcutoff2))
2352 r12 = _mm256_mul_ps(rsq12,rinv12);
2353 r12 = _mm256_andnot_ps(dummy_mask,r12);
2355 /* EWALD ELECTROSTATICS */
2357 /* Analytical PME correction */
2358 zeta2 = _mm256_mul_ps(beta2,rsq12);
2359 rinv3 = _mm256_mul_ps(rinvsq12,rinv12);
2360 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
2361 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2362 felec = _mm256_mul_ps(qq12,felec);
2364 cutoff_mask = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2368 fscal = _mm256_and_ps(fscal,cutoff_mask);
2370 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2372 /* Calculate temporary vectorial force */
2373 tx = _mm256_mul_ps(fscal,dx12);
2374 ty = _mm256_mul_ps(fscal,dy12);
2375 tz = _mm256_mul_ps(fscal,dz12);
2377 /* Update vectorial force */
2378 fix1 = _mm256_add_ps(fix1,tx);
2379 fiy1 = _mm256_add_ps(fiy1,ty);
2380 fiz1 = _mm256_add_ps(fiz1,tz);
2382 fjx2 = _mm256_add_ps(fjx2,tx);
2383 fjy2 = _mm256_add_ps(fjy2,ty);
2384 fjz2 = _mm256_add_ps(fjz2,tz);
2388 /**************************
2389 * CALCULATE INTERACTIONS *
2390 **************************/
2392 if (gmx_mm256_any_lt(rsq20,rcutoff2))
2395 r20 = _mm256_mul_ps(rsq20,rinv20);
2396 r20 = _mm256_andnot_ps(dummy_mask,r20);
2398 /* EWALD ELECTROSTATICS */
2400 /* Analytical PME correction */
2401 zeta2 = _mm256_mul_ps(beta2,rsq20);
2402 rinv3 = _mm256_mul_ps(rinvsq20,rinv20);
2403 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
2404 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2405 felec = _mm256_mul_ps(qq20,felec);
2407 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
2411 fscal = _mm256_and_ps(fscal,cutoff_mask);
2413 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2415 /* Calculate temporary vectorial force */
2416 tx = _mm256_mul_ps(fscal,dx20);
2417 ty = _mm256_mul_ps(fscal,dy20);
2418 tz = _mm256_mul_ps(fscal,dz20);
2420 /* Update vectorial force */
2421 fix2 = _mm256_add_ps(fix2,tx);
2422 fiy2 = _mm256_add_ps(fiy2,ty);
2423 fiz2 = _mm256_add_ps(fiz2,tz);
2425 fjx0 = _mm256_add_ps(fjx0,tx);
2426 fjy0 = _mm256_add_ps(fjy0,ty);
2427 fjz0 = _mm256_add_ps(fjz0,tz);
2431 /**************************
2432 * CALCULATE INTERACTIONS *
2433 **************************/
2435 if (gmx_mm256_any_lt(rsq21,rcutoff2))
2438 r21 = _mm256_mul_ps(rsq21,rinv21);
2439 r21 = _mm256_andnot_ps(dummy_mask,r21);
2441 /* EWALD ELECTROSTATICS */
2443 /* Analytical PME correction */
2444 zeta2 = _mm256_mul_ps(beta2,rsq21);
2445 rinv3 = _mm256_mul_ps(rinvsq21,rinv21);
2446 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
2447 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2448 felec = _mm256_mul_ps(qq21,felec);
2450 cutoff_mask = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2454 fscal = _mm256_and_ps(fscal,cutoff_mask);
2456 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2458 /* Calculate temporary vectorial force */
2459 tx = _mm256_mul_ps(fscal,dx21);
2460 ty = _mm256_mul_ps(fscal,dy21);
2461 tz = _mm256_mul_ps(fscal,dz21);
2463 /* Update vectorial force */
2464 fix2 = _mm256_add_ps(fix2,tx);
2465 fiy2 = _mm256_add_ps(fiy2,ty);
2466 fiz2 = _mm256_add_ps(fiz2,tz);
2468 fjx1 = _mm256_add_ps(fjx1,tx);
2469 fjy1 = _mm256_add_ps(fjy1,ty);
2470 fjz1 = _mm256_add_ps(fjz1,tz);
2474 /**************************
2475 * CALCULATE INTERACTIONS *
2476 **************************/
2478 if (gmx_mm256_any_lt(rsq22,rcutoff2))
2481 r22 = _mm256_mul_ps(rsq22,rinv22);
2482 r22 = _mm256_andnot_ps(dummy_mask,r22);
2484 /* EWALD ELECTROSTATICS */
2486 /* Analytical PME correction */
2487 zeta2 = _mm256_mul_ps(beta2,rsq22);
2488 rinv3 = _mm256_mul_ps(rinvsq22,rinv22);
2489 pmecorrF = gmx_mm256_pmecorrF_ps(zeta2);
2490 felec = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2491 felec = _mm256_mul_ps(qq22,felec);
2493 cutoff_mask = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2497 fscal = _mm256_and_ps(fscal,cutoff_mask);
2499 fscal = _mm256_andnot_ps(dummy_mask,fscal);
2501 /* Calculate temporary vectorial force */
2502 tx = _mm256_mul_ps(fscal,dx22);
2503 ty = _mm256_mul_ps(fscal,dy22);
2504 tz = _mm256_mul_ps(fscal,dz22);
2506 /* Update vectorial force */
2507 fix2 = _mm256_add_ps(fix2,tx);
2508 fiy2 = _mm256_add_ps(fiy2,ty);
2509 fiz2 = _mm256_add_ps(fiz2,tz);
2511 fjx2 = _mm256_add_ps(fjx2,tx);
2512 fjy2 = _mm256_add_ps(fjy2,ty);
2513 fjz2 = _mm256_add_ps(fjz2,tz);
2517 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2518 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2519 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2520 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2521 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2522 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2523 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2524 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2526 gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2527 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2529 /* Inner loop uses 540 flops */
2532 /* End of innermost loop */
2534 gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2535 f+i_coord_offset,fshift+i_shift_offset);
2537 /* Increment number of inner iterations */
2538 inneriter += j_index_end - j_index_start;
2540 /* Outer loop uses 18 flops */
2543 /* Increment number of outer iterations */
2546 /* Update outer/inner flops */
2548 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*540);