2 * Note: this file was generated by the Gromacs avx_256_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_avx_256_double.h"
34 #include "kernelutil_x86_avx_256_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_avx_256_double
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Water3-Water3
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_avx_256_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
63 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
64 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
66 real *shiftvec,*fshift,*x,*f;
67 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
69 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 real * vdwioffsetptr0;
71 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
72 real * vdwioffsetptr1;
73 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 real * vdwioffsetptr2;
75 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
76 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
77 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
78 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
79 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
80 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
81 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
82 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
83 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
84 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
85 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
86 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
87 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
88 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
89 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
90 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
91 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
94 __m256d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
95 __m256d beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
97 __m256d dummy_mask,cutoff_mask;
98 __m128 tmpmask0,tmpmask1;
99 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
100 __m256d one = _mm256_set1_pd(1.0);
101 __m256d two = _mm256_set1_pd(2.0);
107 jindex = nlist->jindex;
109 shiftidx = nlist->shift;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm256_set1_pd(fr->epsfac);
114 charge = mdatoms->chargeA;
116 sh_ewald = _mm256_set1_pd(fr->ic->sh_ewald);
117 beta = _mm256_set1_pd(fr->ic->ewaldcoeff);
118 beta2 = _mm256_mul_pd(beta,beta);
119 beta3 = _mm256_mul_pd(beta,beta2);
121 ewtab = fr->ic->tabq_coul_FDV0;
122 ewtabscale = _mm256_set1_pd(fr->ic->tabq_scale);
123 ewtabhalfspace = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
125 /* Setup water-specific parameters */
126 inr = nlist->iinr[0];
127 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
128 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
129 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
131 jq0 = _mm256_set1_pd(charge[inr+0]);
132 jq1 = _mm256_set1_pd(charge[inr+1]);
133 jq2 = _mm256_set1_pd(charge[inr+2]);
134 qq00 = _mm256_mul_pd(iq0,jq0);
135 qq01 = _mm256_mul_pd(iq0,jq1);
136 qq02 = _mm256_mul_pd(iq0,jq2);
137 qq10 = _mm256_mul_pd(iq1,jq0);
138 qq11 = _mm256_mul_pd(iq1,jq1);
139 qq12 = _mm256_mul_pd(iq1,jq2);
140 qq20 = _mm256_mul_pd(iq2,jq0);
141 qq21 = _mm256_mul_pd(iq2,jq1);
142 qq22 = _mm256_mul_pd(iq2,jq2);
144 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
145 rcutoff_scalar = fr->rcoulomb;
146 rcutoff = _mm256_set1_pd(rcutoff_scalar);
147 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
149 /* Avoid stupid compiler warnings */
150 jnrA = jnrB = jnrC = jnrD = 0;
159 for(iidx=0;iidx<4*DIM;iidx++)
164 /* Start outer loop over neighborlists */
165 for(iidx=0; iidx<nri; iidx++)
167 /* Load shift vector for this list */
168 i_shift_offset = DIM*shiftidx[iidx];
170 /* Load limits for loop over neighbors */
171 j_index_start = jindex[iidx];
172 j_index_end = jindex[iidx+1];
174 /* Get outer coordinate index */
176 i_coord_offset = DIM*inr;
178 /* Load i particle coords and add shift vector */
179 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
180 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
182 fix0 = _mm256_setzero_pd();
183 fiy0 = _mm256_setzero_pd();
184 fiz0 = _mm256_setzero_pd();
185 fix1 = _mm256_setzero_pd();
186 fiy1 = _mm256_setzero_pd();
187 fiz1 = _mm256_setzero_pd();
188 fix2 = _mm256_setzero_pd();
189 fiy2 = _mm256_setzero_pd();
190 fiz2 = _mm256_setzero_pd();
192 /* Reset potential sums */
193 velecsum = _mm256_setzero_pd();
195 /* Start inner kernel loop */
196 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
199 /* Get j neighbor index, and coordinate index */
204 j_coord_offsetA = DIM*jnrA;
205 j_coord_offsetB = DIM*jnrB;
206 j_coord_offsetC = DIM*jnrC;
207 j_coord_offsetD = DIM*jnrD;
209 /* load j atom coordinates */
210 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
211 x+j_coord_offsetC,x+j_coord_offsetD,
212 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
214 /* Calculate displacement vector */
215 dx00 = _mm256_sub_pd(ix0,jx0);
216 dy00 = _mm256_sub_pd(iy0,jy0);
217 dz00 = _mm256_sub_pd(iz0,jz0);
218 dx01 = _mm256_sub_pd(ix0,jx1);
219 dy01 = _mm256_sub_pd(iy0,jy1);
220 dz01 = _mm256_sub_pd(iz0,jz1);
221 dx02 = _mm256_sub_pd(ix0,jx2);
222 dy02 = _mm256_sub_pd(iy0,jy2);
223 dz02 = _mm256_sub_pd(iz0,jz2);
224 dx10 = _mm256_sub_pd(ix1,jx0);
225 dy10 = _mm256_sub_pd(iy1,jy0);
226 dz10 = _mm256_sub_pd(iz1,jz0);
227 dx11 = _mm256_sub_pd(ix1,jx1);
228 dy11 = _mm256_sub_pd(iy1,jy1);
229 dz11 = _mm256_sub_pd(iz1,jz1);
230 dx12 = _mm256_sub_pd(ix1,jx2);
231 dy12 = _mm256_sub_pd(iy1,jy2);
232 dz12 = _mm256_sub_pd(iz1,jz2);
233 dx20 = _mm256_sub_pd(ix2,jx0);
234 dy20 = _mm256_sub_pd(iy2,jy0);
235 dz20 = _mm256_sub_pd(iz2,jz0);
236 dx21 = _mm256_sub_pd(ix2,jx1);
237 dy21 = _mm256_sub_pd(iy2,jy1);
238 dz21 = _mm256_sub_pd(iz2,jz1);
239 dx22 = _mm256_sub_pd(ix2,jx2);
240 dy22 = _mm256_sub_pd(iy2,jy2);
241 dz22 = _mm256_sub_pd(iz2,jz2);
243 /* Calculate squared distance and things based on it */
244 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
245 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
246 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
247 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
248 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
249 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
250 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
251 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
252 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
254 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
255 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
256 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
257 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
258 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
259 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
260 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
261 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
262 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
264 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
265 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
266 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
267 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
268 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
269 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
270 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
271 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
272 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
274 fjx0 = _mm256_setzero_pd();
275 fjy0 = _mm256_setzero_pd();
276 fjz0 = _mm256_setzero_pd();
277 fjx1 = _mm256_setzero_pd();
278 fjy1 = _mm256_setzero_pd();
279 fjz1 = _mm256_setzero_pd();
280 fjx2 = _mm256_setzero_pd();
281 fjy2 = _mm256_setzero_pd();
282 fjz2 = _mm256_setzero_pd();
284 /**************************
285 * CALCULATE INTERACTIONS *
286 **************************/
288 if (gmx_mm256_any_lt(rsq00,rcutoff2))
291 r00 = _mm256_mul_pd(rsq00,rinv00);
293 /* EWALD ELECTROSTATICS */
295 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
296 ewrt = _mm256_mul_pd(r00,ewtabscale);
297 ewitab = _mm256_cvttpd_epi32(ewrt);
298 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
299 ewitab = _mm_slli_epi32(ewitab,2);
300 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
301 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
302 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
303 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
304 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
305 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
306 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
307 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
308 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
310 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
312 /* Update potential sum for this i atom from the interaction with this j atom. */
313 velec = _mm256_and_pd(velec,cutoff_mask);
314 velecsum = _mm256_add_pd(velecsum,velec);
318 fscal = _mm256_and_pd(fscal,cutoff_mask);
320 /* Calculate temporary vectorial force */
321 tx = _mm256_mul_pd(fscal,dx00);
322 ty = _mm256_mul_pd(fscal,dy00);
323 tz = _mm256_mul_pd(fscal,dz00);
325 /* Update vectorial force */
326 fix0 = _mm256_add_pd(fix0,tx);
327 fiy0 = _mm256_add_pd(fiy0,ty);
328 fiz0 = _mm256_add_pd(fiz0,tz);
330 fjx0 = _mm256_add_pd(fjx0,tx);
331 fjy0 = _mm256_add_pd(fjy0,ty);
332 fjz0 = _mm256_add_pd(fjz0,tz);
336 /**************************
337 * CALCULATE INTERACTIONS *
338 **************************/
340 if (gmx_mm256_any_lt(rsq01,rcutoff2))
343 r01 = _mm256_mul_pd(rsq01,rinv01);
345 /* EWALD ELECTROSTATICS */
347 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
348 ewrt = _mm256_mul_pd(r01,ewtabscale);
349 ewitab = _mm256_cvttpd_epi32(ewrt);
350 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
351 ewitab = _mm_slli_epi32(ewitab,2);
352 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
353 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
354 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
355 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
356 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
357 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
358 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
359 velec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_sub_pd(rinv01,sh_ewald),velec));
360 felec = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
362 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
364 /* Update potential sum for this i atom from the interaction with this j atom. */
365 velec = _mm256_and_pd(velec,cutoff_mask);
366 velecsum = _mm256_add_pd(velecsum,velec);
370 fscal = _mm256_and_pd(fscal,cutoff_mask);
372 /* Calculate temporary vectorial force */
373 tx = _mm256_mul_pd(fscal,dx01);
374 ty = _mm256_mul_pd(fscal,dy01);
375 tz = _mm256_mul_pd(fscal,dz01);
377 /* Update vectorial force */
378 fix0 = _mm256_add_pd(fix0,tx);
379 fiy0 = _mm256_add_pd(fiy0,ty);
380 fiz0 = _mm256_add_pd(fiz0,tz);
382 fjx1 = _mm256_add_pd(fjx1,tx);
383 fjy1 = _mm256_add_pd(fjy1,ty);
384 fjz1 = _mm256_add_pd(fjz1,tz);
388 /**************************
389 * CALCULATE INTERACTIONS *
390 **************************/
392 if (gmx_mm256_any_lt(rsq02,rcutoff2))
395 r02 = _mm256_mul_pd(rsq02,rinv02);
397 /* EWALD ELECTROSTATICS */
399 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
400 ewrt = _mm256_mul_pd(r02,ewtabscale);
401 ewitab = _mm256_cvttpd_epi32(ewrt);
402 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
403 ewitab = _mm_slli_epi32(ewitab,2);
404 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
405 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
406 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
407 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
408 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
409 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
410 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
411 velec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_sub_pd(rinv02,sh_ewald),velec));
412 felec = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
414 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
416 /* Update potential sum for this i atom from the interaction with this j atom. */
417 velec = _mm256_and_pd(velec,cutoff_mask);
418 velecsum = _mm256_add_pd(velecsum,velec);
422 fscal = _mm256_and_pd(fscal,cutoff_mask);
424 /* Calculate temporary vectorial force */
425 tx = _mm256_mul_pd(fscal,dx02);
426 ty = _mm256_mul_pd(fscal,dy02);
427 tz = _mm256_mul_pd(fscal,dz02);
429 /* Update vectorial force */
430 fix0 = _mm256_add_pd(fix0,tx);
431 fiy0 = _mm256_add_pd(fiy0,ty);
432 fiz0 = _mm256_add_pd(fiz0,tz);
434 fjx2 = _mm256_add_pd(fjx2,tx);
435 fjy2 = _mm256_add_pd(fjy2,ty);
436 fjz2 = _mm256_add_pd(fjz2,tz);
440 /**************************
441 * CALCULATE INTERACTIONS *
442 **************************/
444 if (gmx_mm256_any_lt(rsq10,rcutoff2))
447 r10 = _mm256_mul_pd(rsq10,rinv10);
449 /* EWALD ELECTROSTATICS */
451 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
452 ewrt = _mm256_mul_pd(r10,ewtabscale);
453 ewitab = _mm256_cvttpd_epi32(ewrt);
454 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
455 ewitab = _mm_slli_epi32(ewitab,2);
456 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
457 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
458 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
459 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
460 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
461 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
462 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
463 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
464 felec = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
466 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
468 /* Update potential sum for this i atom from the interaction with this j atom. */
469 velec = _mm256_and_pd(velec,cutoff_mask);
470 velecsum = _mm256_add_pd(velecsum,velec);
474 fscal = _mm256_and_pd(fscal,cutoff_mask);
476 /* Calculate temporary vectorial force */
477 tx = _mm256_mul_pd(fscal,dx10);
478 ty = _mm256_mul_pd(fscal,dy10);
479 tz = _mm256_mul_pd(fscal,dz10);
481 /* Update vectorial force */
482 fix1 = _mm256_add_pd(fix1,tx);
483 fiy1 = _mm256_add_pd(fiy1,ty);
484 fiz1 = _mm256_add_pd(fiz1,tz);
486 fjx0 = _mm256_add_pd(fjx0,tx);
487 fjy0 = _mm256_add_pd(fjy0,ty);
488 fjz0 = _mm256_add_pd(fjz0,tz);
492 /**************************
493 * CALCULATE INTERACTIONS *
494 **************************/
496 if (gmx_mm256_any_lt(rsq11,rcutoff2))
499 r11 = _mm256_mul_pd(rsq11,rinv11);
501 /* EWALD ELECTROSTATICS */
503 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
504 ewrt = _mm256_mul_pd(r11,ewtabscale);
505 ewitab = _mm256_cvttpd_epi32(ewrt);
506 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
507 ewitab = _mm_slli_epi32(ewitab,2);
508 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
509 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
510 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
511 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
512 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
513 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
514 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
515 velec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_sub_pd(rinv11,sh_ewald),velec));
516 felec = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
518 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
520 /* Update potential sum for this i atom from the interaction with this j atom. */
521 velec = _mm256_and_pd(velec,cutoff_mask);
522 velecsum = _mm256_add_pd(velecsum,velec);
526 fscal = _mm256_and_pd(fscal,cutoff_mask);
528 /* Calculate temporary vectorial force */
529 tx = _mm256_mul_pd(fscal,dx11);
530 ty = _mm256_mul_pd(fscal,dy11);
531 tz = _mm256_mul_pd(fscal,dz11);
533 /* Update vectorial force */
534 fix1 = _mm256_add_pd(fix1,tx);
535 fiy1 = _mm256_add_pd(fiy1,ty);
536 fiz1 = _mm256_add_pd(fiz1,tz);
538 fjx1 = _mm256_add_pd(fjx1,tx);
539 fjy1 = _mm256_add_pd(fjy1,ty);
540 fjz1 = _mm256_add_pd(fjz1,tz);
544 /**************************
545 * CALCULATE INTERACTIONS *
546 **************************/
548 if (gmx_mm256_any_lt(rsq12,rcutoff2))
551 r12 = _mm256_mul_pd(rsq12,rinv12);
553 /* EWALD ELECTROSTATICS */
555 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
556 ewrt = _mm256_mul_pd(r12,ewtabscale);
557 ewitab = _mm256_cvttpd_epi32(ewrt);
558 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
559 ewitab = _mm_slli_epi32(ewitab,2);
560 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
561 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
562 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
563 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
564 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
565 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
566 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
567 velec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_sub_pd(rinv12,sh_ewald),velec));
568 felec = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
570 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
572 /* Update potential sum for this i atom from the interaction with this j atom. */
573 velec = _mm256_and_pd(velec,cutoff_mask);
574 velecsum = _mm256_add_pd(velecsum,velec);
578 fscal = _mm256_and_pd(fscal,cutoff_mask);
580 /* Calculate temporary vectorial force */
581 tx = _mm256_mul_pd(fscal,dx12);
582 ty = _mm256_mul_pd(fscal,dy12);
583 tz = _mm256_mul_pd(fscal,dz12);
585 /* Update vectorial force */
586 fix1 = _mm256_add_pd(fix1,tx);
587 fiy1 = _mm256_add_pd(fiy1,ty);
588 fiz1 = _mm256_add_pd(fiz1,tz);
590 fjx2 = _mm256_add_pd(fjx2,tx);
591 fjy2 = _mm256_add_pd(fjy2,ty);
592 fjz2 = _mm256_add_pd(fjz2,tz);
596 /**************************
597 * CALCULATE INTERACTIONS *
598 **************************/
600 if (gmx_mm256_any_lt(rsq20,rcutoff2))
603 r20 = _mm256_mul_pd(rsq20,rinv20);
605 /* EWALD ELECTROSTATICS */
607 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
608 ewrt = _mm256_mul_pd(r20,ewtabscale);
609 ewitab = _mm256_cvttpd_epi32(ewrt);
610 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
611 ewitab = _mm_slli_epi32(ewitab,2);
612 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
613 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
614 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
615 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
616 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
617 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
618 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
619 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
620 felec = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
622 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
624 /* Update potential sum for this i atom from the interaction with this j atom. */
625 velec = _mm256_and_pd(velec,cutoff_mask);
626 velecsum = _mm256_add_pd(velecsum,velec);
630 fscal = _mm256_and_pd(fscal,cutoff_mask);
632 /* Calculate temporary vectorial force */
633 tx = _mm256_mul_pd(fscal,dx20);
634 ty = _mm256_mul_pd(fscal,dy20);
635 tz = _mm256_mul_pd(fscal,dz20);
637 /* Update vectorial force */
638 fix2 = _mm256_add_pd(fix2,tx);
639 fiy2 = _mm256_add_pd(fiy2,ty);
640 fiz2 = _mm256_add_pd(fiz2,tz);
642 fjx0 = _mm256_add_pd(fjx0,tx);
643 fjy0 = _mm256_add_pd(fjy0,ty);
644 fjz0 = _mm256_add_pd(fjz0,tz);
648 /**************************
649 * CALCULATE INTERACTIONS *
650 **************************/
652 if (gmx_mm256_any_lt(rsq21,rcutoff2))
655 r21 = _mm256_mul_pd(rsq21,rinv21);
657 /* EWALD ELECTROSTATICS */
659 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
660 ewrt = _mm256_mul_pd(r21,ewtabscale);
661 ewitab = _mm256_cvttpd_epi32(ewrt);
662 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
663 ewitab = _mm_slli_epi32(ewitab,2);
664 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
665 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
666 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
667 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
668 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
669 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
670 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
671 velec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_sub_pd(rinv21,sh_ewald),velec));
672 felec = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
674 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
676 /* Update potential sum for this i atom from the interaction with this j atom. */
677 velec = _mm256_and_pd(velec,cutoff_mask);
678 velecsum = _mm256_add_pd(velecsum,velec);
682 fscal = _mm256_and_pd(fscal,cutoff_mask);
684 /* Calculate temporary vectorial force */
685 tx = _mm256_mul_pd(fscal,dx21);
686 ty = _mm256_mul_pd(fscal,dy21);
687 tz = _mm256_mul_pd(fscal,dz21);
689 /* Update vectorial force */
690 fix2 = _mm256_add_pd(fix2,tx);
691 fiy2 = _mm256_add_pd(fiy2,ty);
692 fiz2 = _mm256_add_pd(fiz2,tz);
694 fjx1 = _mm256_add_pd(fjx1,tx);
695 fjy1 = _mm256_add_pd(fjy1,ty);
696 fjz1 = _mm256_add_pd(fjz1,tz);
700 /**************************
701 * CALCULATE INTERACTIONS *
702 **************************/
704 if (gmx_mm256_any_lt(rsq22,rcutoff2))
707 r22 = _mm256_mul_pd(rsq22,rinv22);
709 /* EWALD ELECTROSTATICS */
711 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
712 ewrt = _mm256_mul_pd(r22,ewtabscale);
713 ewitab = _mm256_cvttpd_epi32(ewrt);
714 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
715 ewitab = _mm_slli_epi32(ewitab,2);
716 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
717 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
718 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
719 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
720 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
721 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
722 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
723 velec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_sub_pd(rinv22,sh_ewald),velec));
724 felec = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
726 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
728 /* Update potential sum for this i atom from the interaction with this j atom. */
729 velec = _mm256_and_pd(velec,cutoff_mask);
730 velecsum = _mm256_add_pd(velecsum,velec);
734 fscal = _mm256_and_pd(fscal,cutoff_mask);
736 /* Calculate temporary vectorial force */
737 tx = _mm256_mul_pd(fscal,dx22);
738 ty = _mm256_mul_pd(fscal,dy22);
739 tz = _mm256_mul_pd(fscal,dz22);
741 /* Update vectorial force */
742 fix2 = _mm256_add_pd(fix2,tx);
743 fiy2 = _mm256_add_pd(fiy2,ty);
744 fiz2 = _mm256_add_pd(fiz2,tz);
746 fjx2 = _mm256_add_pd(fjx2,tx);
747 fjy2 = _mm256_add_pd(fjy2,ty);
748 fjz2 = _mm256_add_pd(fjz2,tz);
752 fjptrA = f+j_coord_offsetA;
753 fjptrB = f+j_coord_offsetB;
754 fjptrC = f+j_coord_offsetC;
755 fjptrD = f+j_coord_offsetD;
757 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
758 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
760 /* Inner loop uses 414 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 /* Sign of each element will be negative for non-real atoms.
772 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
773 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
775 tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
777 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
778 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
779 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
781 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
782 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
783 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
784 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
785 j_coord_offsetA = DIM*jnrA;
786 j_coord_offsetB = DIM*jnrB;
787 j_coord_offsetC = DIM*jnrC;
788 j_coord_offsetD = DIM*jnrD;
790 /* load j atom coordinates */
791 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
792 x+j_coord_offsetC,x+j_coord_offsetD,
793 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
795 /* Calculate displacement vector */
796 dx00 = _mm256_sub_pd(ix0,jx0);
797 dy00 = _mm256_sub_pd(iy0,jy0);
798 dz00 = _mm256_sub_pd(iz0,jz0);
799 dx01 = _mm256_sub_pd(ix0,jx1);
800 dy01 = _mm256_sub_pd(iy0,jy1);
801 dz01 = _mm256_sub_pd(iz0,jz1);
802 dx02 = _mm256_sub_pd(ix0,jx2);
803 dy02 = _mm256_sub_pd(iy0,jy2);
804 dz02 = _mm256_sub_pd(iz0,jz2);
805 dx10 = _mm256_sub_pd(ix1,jx0);
806 dy10 = _mm256_sub_pd(iy1,jy0);
807 dz10 = _mm256_sub_pd(iz1,jz0);
808 dx11 = _mm256_sub_pd(ix1,jx1);
809 dy11 = _mm256_sub_pd(iy1,jy1);
810 dz11 = _mm256_sub_pd(iz1,jz1);
811 dx12 = _mm256_sub_pd(ix1,jx2);
812 dy12 = _mm256_sub_pd(iy1,jy2);
813 dz12 = _mm256_sub_pd(iz1,jz2);
814 dx20 = _mm256_sub_pd(ix2,jx0);
815 dy20 = _mm256_sub_pd(iy2,jy0);
816 dz20 = _mm256_sub_pd(iz2,jz0);
817 dx21 = _mm256_sub_pd(ix2,jx1);
818 dy21 = _mm256_sub_pd(iy2,jy1);
819 dz21 = _mm256_sub_pd(iz2,jz1);
820 dx22 = _mm256_sub_pd(ix2,jx2);
821 dy22 = _mm256_sub_pd(iy2,jy2);
822 dz22 = _mm256_sub_pd(iz2,jz2);
824 /* Calculate squared distance and things based on it */
825 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
826 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
827 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
828 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
829 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
830 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
831 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
832 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
833 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
835 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
836 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
837 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
838 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
839 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
840 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
841 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
842 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
843 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
845 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
846 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
847 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
848 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
849 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
850 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
851 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
852 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
853 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
855 fjx0 = _mm256_setzero_pd();
856 fjy0 = _mm256_setzero_pd();
857 fjz0 = _mm256_setzero_pd();
858 fjx1 = _mm256_setzero_pd();
859 fjy1 = _mm256_setzero_pd();
860 fjz1 = _mm256_setzero_pd();
861 fjx2 = _mm256_setzero_pd();
862 fjy2 = _mm256_setzero_pd();
863 fjz2 = _mm256_setzero_pd();
865 /**************************
866 * CALCULATE INTERACTIONS *
867 **************************/
869 if (gmx_mm256_any_lt(rsq00,rcutoff2))
872 r00 = _mm256_mul_pd(rsq00,rinv00);
873 r00 = _mm256_andnot_pd(dummy_mask,r00);
875 /* EWALD ELECTROSTATICS */
877 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
878 ewrt = _mm256_mul_pd(r00,ewtabscale);
879 ewitab = _mm256_cvttpd_epi32(ewrt);
880 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
881 ewitab = _mm_slli_epi32(ewitab,2);
882 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
883 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
884 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
885 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
886 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
887 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
888 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
889 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
890 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
892 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
894 /* Update potential sum for this i atom from the interaction with this j atom. */
895 velec = _mm256_and_pd(velec,cutoff_mask);
896 velec = _mm256_andnot_pd(dummy_mask,velec);
897 velecsum = _mm256_add_pd(velecsum,velec);
901 fscal = _mm256_and_pd(fscal,cutoff_mask);
903 fscal = _mm256_andnot_pd(dummy_mask,fscal);
905 /* Calculate temporary vectorial force */
906 tx = _mm256_mul_pd(fscal,dx00);
907 ty = _mm256_mul_pd(fscal,dy00);
908 tz = _mm256_mul_pd(fscal,dz00);
910 /* Update vectorial force */
911 fix0 = _mm256_add_pd(fix0,tx);
912 fiy0 = _mm256_add_pd(fiy0,ty);
913 fiz0 = _mm256_add_pd(fiz0,tz);
915 fjx0 = _mm256_add_pd(fjx0,tx);
916 fjy0 = _mm256_add_pd(fjy0,ty);
917 fjz0 = _mm256_add_pd(fjz0,tz);
921 /**************************
922 * CALCULATE INTERACTIONS *
923 **************************/
925 if (gmx_mm256_any_lt(rsq01,rcutoff2))
928 r01 = _mm256_mul_pd(rsq01,rinv01);
929 r01 = _mm256_andnot_pd(dummy_mask,r01);
931 /* EWALD ELECTROSTATICS */
933 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
934 ewrt = _mm256_mul_pd(r01,ewtabscale);
935 ewitab = _mm256_cvttpd_epi32(ewrt);
936 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
937 ewitab = _mm_slli_epi32(ewitab,2);
938 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
939 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
940 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
941 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
942 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
943 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
944 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
945 velec = _mm256_mul_pd(qq01,_mm256_sub_pd(_mm256_sub_pd(rinv01,sh_ewald),velec));
946 felec = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
948 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
950 /* Update potential sum for this i atom from the interaction with this j atom. */
951 velec = _mm256_and_pd(velec,cutoff_mask);
952 velec = _mm256_andnot_pd(dummy_mask,velec);
953 velecsum = _mm256_add_pd(velecsum,velec);
957 fscal = _mm256_and_pd(fscal,cutoff_mask);
959 fscal = _mm256_andnot_pd(dummy_mask,fscal);
961 /* Calculate temporary vectorial force */
962 tx = _mm256_mul_pd(fscal,dx01);
963 ty = _mm256_mul_pd(fscal,dy01);
964 tz = _mm256_mul_pd(fscal,dz01);
966 /* Update vectorial force */
967 fix0 = _mm256_add_pd(fix0,tx);
968 fiy0 = _mm256_add_pd(fiy0,ty);
969 fiz0 = _mm256_add_pd(fiz0,tz);
971 fjx1 = _mm256_add_pd(fjx1,tx);
972 fjy1 = _mm256_add_pd(fjy1,ty);
973 fjz1 = _mm256_add_pd(fjz1,tz);
977 /**************************
978 * CALCULATE INTERACTIONS *
979 **************************/
981 if (gmx_mm256_any_lt(rsq02,rcutoff2))
984 r02 = _mm256_mul_pd(rsq02,rinv02);
985 r02 = _mm256_andnot_pd(dummy_mask,r02);
987 /* EWALD ELECTROSTATICS */
989 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
990 ewrt = _mm256_mul_pd(r02,ewtabscale);
991 ewitab = _mm256_cvttpd_epi32(ewrt);
992 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
993 ewitab = _mm_slli_epi32(ewitab,2);
994 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
995 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
996 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
997 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
998 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
999 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1000 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1001 velec = _mm256_mul_pd(qq02,_mm256_sub_pd(_mm256_sub_pd(rinv02,sh_ewald),velec));
1002 felec = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
1004 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1006 /* Update potential sum for this i atom from the interaction with this j atom. */
1007 velec = _mm256_and_pd(velec,cutoff_mask);
1008 velec = _mm256_andnot_pd(dummy_mask,velec);
1009 velecsum = _mm256_add_pd(velecsum,velec);
1013 fscal = _mm256_and_pd(fscal,cutoff_mask);
1015 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1017 /* Calculate temporary vectorial force */
1018 tx = _mm256_mul_pd(fscal,dx02);
1019 ty = _mm256_mul_pd(fscal,dy02);
1020 tz = _mm256_mul_pd(fscal,dz02);
1022 /* Update vectorial force */
1023 fix0 = _mm256_add_pd(fix0,tx);
1024 fiy0 = _mm256_add_pd(fiy0,ty);
1025 fiz0 = _mm256_add_pd(fiz0,tz);
1027 fjx2 = _mm256_add_pd(fjx2,tx);
1028 fjy2 = _mm256_add_pd(fjy2,ty);
1029 fjz2 = _mm256_add_pd(fjz2,tz);
1033 /**************************
1034 * CALCULATE INTERACTIONS *
1035 **************************/
1037 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1040 r10 = _mm256_mul_pd(rsq10,rinv10);
1041 r10 = _mm256_andnot_pd(dummy_mask,r10);
1043 /* EWALD ELECTROSTATICS */
1045 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1046 ewrt = _mm256_mul_pd(r10,ewtabscale);
1047 ewitab = _mm256_cvttpd_epi32(ewrt);
1048 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1049 ewitab = _mm_slli_epi32(ewitab,2);
1050 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1051 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1052 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1053 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1054 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1055 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1056 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1057 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
1058 felec = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1060 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1062 /* Update potential sum for this i atom from the interaction with this j atom. */
1063 velec = _mm256_and_pd(velec,cutoff_mask);
1064 velec = _mm256_andnot_pd(dummy_mask,velec);
1065 velecsum = _mm256_add_pd(velecsum,velec);
1069 fscal = _mm256_and_pd(fscal,cutoff_mask);
1071 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1073 /* Calculate temporary vectorial force */
1074 tx = _mm256_mul_pd(fscal,dx10);
1075 ty = _mm256_mul_pd(fscal,dy10);
1076 tz = _mm256_mul_pd(fscal,dz10);
1078 /* Update vectorial force */
1079 fix1 = _mm256_add_pd(fix1,tx);
1080 fiy1 = _mm256_add_pd(fiy1,ty);
1081 fiz1 = _mm256_add_pd(fiz1,tz);
1083 fjx0 = _mm256_add_pd(fjx0,tx);
1084 fjy0 = _mm256_add_pd(fjy0,ty);
1085 fjz0 = _mm256_add_pd(fjz0,tz);
1089 /**************************
1090 * CALCULATE INTERACTIONS *
1091 **************************/
1093 if (gmx_mm256_any_lt(rsq11,rcutoff2))
1096 r11 = _mm256_mul_pd(rsq11,rinv11);
1097 r11 = _mm256_andnot_pd(dummy_mask,r11);
1099 /* EWALD ELECTROSTATICS */
1101 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1102 ewrt = _mm256_mul_pd(r11,ewtabscale);
1103 ewitab = _mm256_cvttpd_epi32(ewrt);
1104 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1105 ewitab = _mm_slli_epi32(ewitab,2);
1106 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1107 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1108 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1109 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1110 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1111 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1112 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1113 velec = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_sub_pd(rinv11,sh_ewald),velec));
1114 felec = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1116 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1118 /* Update potential sum for this i atom from the interaction with this j atom. */
1119 velec = _mm256_and_pd(velec,cutoff_mask);
1120 velec = _mm256_andnot_pd(dummy_mask,velec);
1121 velecsum = _mm256_add_pd(velecsum,velec);
1125 fscal = _mm256_and_pd(fscal,cutoff_mask);
1127 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1129 /* Calculate temporary vectorial force */
1130 tx = _mm256_mul_pd(fscal,dx11);
1131 ty = _mm256_mul_pd(fscal,dy11);
1132 tz = _mm256_mul_pd(fscal,dz11);
1134 /* Update vectorial force */
1135 fix1 = _mm256_add_pd(fix1,tx);
1136 fiy1 = _mm256_add_pd(fiy1,ty);
1137 fiz1 = _mm256_add_pd(fiz1,tz);
1139 fjx1 = _mm256_add_pd(fjx1,tx);
1140 fjy1 = _mm256_add_pd(fjy1,ty);
1141 fjz1 = _mm256_add_pd(fjz1,tz);
1145 /**************************
1146 * CALCULATE INTERACTIONS *
1147 **************************/
1149 if (gmx_mm256_any_lt(rsq12,rcutoff2))
1152 r12 = _mm256_mul_pd(rsq12,rinv12);
1153 r12 = _mm256_andnot_pd(dummy_mask,r12);
1155 /* EWALD ELECTROSTATICS */
1157 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1158 ewrt = _mm256_mul_pd(r12,ewtabscale);
1159 ewitab = _mm256_cvttpd_epi32(ewrt);
1160 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1161 ewitab = _mm_slli_epi32(ewitab,2);
1162 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1163 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1164 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1165 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1166 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1167 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1168 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1169 velec = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_sub_pd(rinv12,sh_ewald),velec));
1170 felec = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1172 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1174 /* Update potential sum for this i atom from the interaction with this j atom. */
1175 velec = _mm256_and_pd(velec,cutoff_mask);
1176 velec = _mm256_andnot_pd(dummy_mask,velec);
1177 velecsum = _mm256_add_pd(velecsum,velec);
1181 fscal = _mm256_and_pd(fscal,cutoff_mask);
1183 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1185 /* Calculate temporary vectorial force */
1186 tx = _mm256_mul_pd(fscal,dx12);
1187 ty = _mm256_mul_pd(fscal,dy12);
1188 tz = _mm256_mul_pd(fscal,dz12);
1190 /* Update vectorial force */
1191 fix1 = _mm256_add_pd(fix1,tx);
1192 fiy1 = _mm256_add_pd(fiy1,ty);
1193 fiz1 = _mm256_add_pd(fiz1,tz);
1195 fjx2 = _mm256_add_pd(fjx2,tx);
1196 fjy2 = _mm256_add_pd(fjy2,ty);
1197 fjz2 = _mm256_add_pd(fjz2,tz);
1201 /**************************
1202 * CALCULATE INTERACTIONS *
1203 **************************/
1205 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1208 r20 = _mm256_mul_pd(rsq20,rinv20);
1209 r20 = _mm256_andnot_pd(dummy_mask,r20);
1211 /* EWALD ELECTROSTATICS */
1213 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1214 ewrt = _mm256_mul_pd(r20,ewtabscale);
1215 ewitab = _mm256_cvttpd_epi32(ewrt);
1216 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1217 ewitab = _mm_slli_epi32(ewitab,2);
1218 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1219 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1220 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1221 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1222 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1223 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1224 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1225 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
1226 felec = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1228 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1230 /* Update potential sum for this i atom from the interaction with this j atom. */
1231 velec = _mm256_and_pd(velec,cutoff_mask);
1232 velec = _mm256_andnot_pd(dummy_mask,velec);
1233 velecsum = _mm256_add_pd(velecsum,velec);
1237 fscal = _mm256_and_pd(fscal,cutoff_mask);
1239 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1241 /* Calculate temporary vectorial force */
1242 tx = _mm256_mul_pd(fscal,dx20);
1243 ty = _mm256_mul_pd(fscal,dy20);
1244 tz = _mm256_mul_pd(fscal,dz20);
1246 /* Update vectorial force */
1247 fix2 = _mm256_add_pd(fix2,tx);
1248 fiy2 = _mm256_add_pd(fiy2,ty);
1249 fiz2 = _mm256_add_pd(fiz2,tz);
1251 fjx0 = _mm256_add_pd(fjx0,tx);
1252 fjy0 = _mm256_add_pd(fjy0,ty);
1253 fjz0 = _mm256_add_pd(fjz0,tz);
1257 /**************************
1258 * CALCULATE INTERACTIONS *
1259 **************************/
1261 if (gmx_mm256_any_lt(rsq21,rcutoff2))
1264 r21 = _mm256_mul_pd(rsq21,rinv21);
1265 r21 = _mm256_andnot_pd(dummy_mask,r21);
1267 /* EWALD ELECTROSTATICS */
1269 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1270 ewrt = _mm256_mul_pd(r21,ewtabscale);
1271 ewitab = _mm256_cvttpd_epi32(ewrt);
1272 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1273 ewitab = _mm_slli_epi32(ewitab,2);
1274 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1275 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1276 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1277 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1278 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1279 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1280 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1281 velec = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_sub_pd(rinv21,sh_ewald),velec));
1282 felec = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1284 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1286 /* Update potential sum for this i atom from the interaction with this j atom. */
1287 velec = _mm256_and_pd(velec,cutoff_mask);
1288 velec = _mm256_andnot_pd(dummy_mask,velec);
1289 velecsum = _mm256_add_pd(velecsum,velec);
1293 fscal = _mm256_and_pd(fscal,cutoff_mask);
1295 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1297 /* Calculate temporary vectorial force */
1298 tx = _mm256_mul_pd(fscal,dx21);
1299 ty = _mm256_mul_pd(fscal,dy21);
1300 tz = _mm256_mul_pd(fscal,dz21);
1302 /* Update vectorial force */
1303 fix2 = _mm256_add_pd(fix2,tx);
1304 fiy2 = _mm256_add_pd(fiy2,ty);
1305 fiz2 = _mm256_add_pd(fiz2,tz);
1307 fjx1 = _mm256_add_pd(fjx1,tx);
1308 fjy1 = _mm256_add_pd(fjy1,ty);
1309 fjz1 = _mm256_add_pd(fjz1,tz);
1313 /**************************
1314 * CALCULATE INTERACTIONS *
1315 **************************/
1317 if (gmx_mm256_any_lt(rsq22,rcutoff2))
1320 r22 = _mm256_mul_pd(rsq22,rinv22);
1321 r22 = _mm256_andnot_pd(dummy_mask,r22);
1323 /* EWALD ELECTROSTATICS */
1325 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1326 ewrt = _mm256_mul_pd(r22,ewtabscale);
1327 ewitab = _mm256_cvttpd_epi32(ewrt);
1328 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1329 ewitab = _mm_slli_epi32(ewitab,2);
1330 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1331 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1332 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1333 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1334 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1335 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1336 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1337 velec = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_sub_pd(rinv22,sh_ewald),velec));
1338 felec = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1340 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1342 /* Update potential sum for this i atom from the interaction with this j atom. */
1343 velec = _mm256_and_pd(velec,cutoff_mask);
1344 velec = _mm256_andnot_pd(dummy_mask,velec);
1345 velecsum = _mm256_add_pd(velecsum,velec);
1349 fscal = _mm256_and_pd(fscal,cutoff_mask);
1351 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1353 /* Calculate temporary vectorial force */
1354 tx = _mm256_mul_pd(fscal,dx22);
1355 ty = _mm256_mul_pd(fscal,dy22);
1356 tz = _mm256_mul_pd(fscal,dz22);
1358 /* Update vectorial force */
1359 fix2 = _mm256_add_pd(fix2,tx);
1360 fiy2 = _mm256_add_pd(fiy2,ty);
1361 fiz2 = _mm256_add_pd(fiz2,tz);
1363 fjx2 = _mm256_add_pd(fjx2,tx);
1364 fjy2 = _mm256_add_pd(fjy2,ty);
1365 fjz2 = _mm256_add_pd(fjz2,tz);
1369 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1370 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1371 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1372 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1374 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1375 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1377 /* Inner loop uses 423 flops */
1380 /* End of innermost loop */
1382 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1383 f+i_coord_offset,fshift+i_shift_offset);
1386 /* Update potential energies */
1387 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1389 /* Increment number of inner iterations */
1390 inneriter += j_index_end - j_index_start;
1392 /* Outer loop uses 19 flops */
1395 /* Increment number of outer iterations */
1398 /* Update outer/inner flops */
1400 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*423);
1403 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_256_double
1404 * Electrostatics interaction: Ewald
1405 * VdW interaction: None
1406 * Geometry: Water3-Water3
1407 * Calculate force/pot: Force
1410 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_256_double
1411 (t_nblist * gmx_restrict nlist,
1412 rvec * gmx_restrict xx,
1413 rvec * gmx_restrict ff,
1414 t_forcerec * gmx_restrict fr,
1415 t_mdatoms * gmx_restrict mdatoms,
1416 nb_kernel_data_t * gmx_restrict kernel_data,
1417 t_nrnb * gmx_restrict nrnb)
1419 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1420 * just 0 for non-waters.
1421 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1422 * jnr indices corresponding to data put in the four positions in the SIMD register.
1424 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1425 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1426 int jnrA,jnrB,jnrC,jnrD;
1427 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1428 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1429 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1430 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1431 real rcutoff_scalar;
1432 real *shiftvec,*fshift,*x,*f;
1433 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1434 real scratch[4*DIM];
1435 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1436 real * vdwioffsetptr0;
1437 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1438 real * vdwioffsetptr1;
1439 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1440 real * vdwioffsetptr2;
1441 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1442 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1443 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1444 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1445 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1446 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1447 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1448 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1449 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1450 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1451 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1452 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1453 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1454 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1455 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1456 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1457 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
1460 __m256d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1461 __m256d beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1463 __m256d dummy_mask,cutoff_mask;
1464 __m128 tmpmask0,tmpmask1;
1465 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1466 __m256d one = _mm256_set1_pd(1.0);
1467 __m256d two = _mm256_set1_pd(2.0);
1473 jindex = nlist->jindex;
1475 shiftidx = nlist->shift;
1477 shiftvec = fr->shift_vec[0];
1478 fshift = fr->fshift[0];
1479 facel = _mm256_set1_pd(fr->epsfac);
1480 charge = mdatoms->chargeA;
1482 sh_ewald = _mm256_set1_pd(fr->ic->sh_ewald);
1483 beta = _mm256_set1_pd(fr->ic->ewaldcoeff);
1484 beta2 = _mm256_mul_pd(beta,beta);
1485 beta3 = _mm256_mul_pd(beta,beta2);
1487 ewtab = fr->ic->tabq_coul_F;
1488 ewtabscale = _mm256_set1_pd(fr->ic->tabq_scale);
1489 ewtabhalfspace = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1491 /* Setup water-specific parameters */
1492 inr = nlist->iinr[0];
1493 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1494 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1495 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1497 jq0 = _mm256_set1_pd(charge[inr+0]);
1498 jq1 = _mm256_set1_pd(charge[inr+1]);
1499 jq2 = _mm256_set1_pd(charge[inr+2]);
1500 qq00 = _mm256_mul_pd(iq0,jq0);
1501 qq01 = _mm256_mul_pd(iq0,jq1);
1502 qq02 = _mm256_mul_pd(iq0,jq2);
1503 qq10 = _mm256_mul_pd(iq1,jq0);
1504 qq11 = _mm256_mul_pd(iq1,jq1);
1505 qq12 = _mm256_mul_pd(iq1,jq2);
1506 qq20 = _mm256_mul_pd(iq2,jq0);
1507 qq21 = _mm256_mul_pd(iq2,jq1);
1508 qq22 = _mm256_mul_pd(iq2,jq2);
1510 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1511 rcutoff_scalar = fr->rcoulomb;
1512 rcutoff = _mm256_set1_pd(rcutoff_scalar);
1513 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
1515 /* Avoid stupid compiler warnings */
1516 jnrA = jnrB = jnrC = jnrD = 0;
1517 j_coord_offsetA = 0;
1518 j_coord_offsetB = 0;
1519 j_coord_offsetC = 0;
1520 j_coord_offsetD = 0;
1525 for(iidx=0;iidx<4*DIM;iidx++)
1527 scratch[iidx] = 0.0;
1530 /* Start outer loop over neighborlists */
1531 for(iidx=0; iidx<nri; iidx++)
1533 /* Load shift vector for this list */
1534 i_shift_offset = DIM*shiftidx[iidx];
1536 /* Load limits for loop over neighbors */
1537 j_index_start = jindex[iidx];
1538 j_index_end = jindex[iidx+1];
1540 /* Get outer coordinate index */
1542 i_coord_offset = DIM*inr;
1544 /* Load i particle coords and add shift vector */
1545 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1546 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1548 fix0 = _mm256_setzero_pd();
1549 fiy0 = _mm256_setzero_pd();
1550 fiz0 = _mm256_setzero_pd();
1551 fix1 = _mm256_setzero_pd();
1552 fiy1 = _mm256_setzero_pd();
1553 fiz1 = _mm256_setzero_pd();
1554 fix2 = _mm256_setzero_pd();
1555 fiy2 = _mm256_setzero_pd();
1556 fiz2 = _mm256_setzero_pd();
1558 /* Start inner kernel loop */
1559 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1562 /* Get j neighbor index, and coordinate index */
1564 jnrB = jjnr[jidx+1];
1565 jnrC = jjnr[jidx+2];
1566 jnrD = jjnr[jidx+3];
1567 j_coord_offsetA = DIM*jnrA;
1568 j_coord_offsetB = DIM*jnrB;
1569 j_coord_offsetC = DIM*jnrC;
1570 j_coord_offsetD = DIM*jnrD;
1572 /* load j atom coordinates */
1573 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1574 x+j_coord_offsetC,x+j_coord_offsetD,
1575 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1577 /* Calculate displacement vector */
1578 dx00 = _mm256_sub_pd(ix0,jx0);
1579 dy00 = _mm256_sub_pd(iy0,jy0);
1580 dz00 = _mm256_sub_pd(iz0,jz0);
1581 dx01 = _mm256_sub_pd(ix0,jx1);
1582 dy01 = _mm256_sub_pd(iy0,jy1);
1583 dz01 = _mm256_sub_pd(iz0,jz1);
1584 dx02 = _mm256_sub_pd(ix0,jx2);
1585 dy02 = _mm256_sub_pd(iy0,jy2);
1586 dz02 = _mm256_sub_pd(iz0,jz2);
1587 dx10 = _mm256_sub_pd(ix1,jx0);
1588 dy10 = _mm256_sub_pd(iy1,jy0);
1589 dz10 = _mm256_sub_pd(iz1,jz0);
1590 dx11 = _mm256_sub_pd(ix1,jx1);
1591 dy11 = _mm256_sub_pd(iy1,jy1);
1592 dz11 = _mm256_sub_pd(iz1,jz1);
1593 dx12 = _mm256_sub_pd(ix1,jx2);
1594 dy12 = _mm256_sub_pd(iy1,jy2);
1595 dz12 = _mm256_sub_pd(iz1,jz2);
1596 dx20 = _mm256_sub_pd(ix2,jx0);
1597 dy20 = _mm256_sub_pd(iy2,jy0);
1598 dz20 = _mm256_sub_pd(iz2,jz0);
1599 dx21 = _mm256_sub_pd(ix2,jx1);
1600 dy21 = _mm256_sub_pd(iy2,jy1);
1601 dz21 = _mm256_sub_pd(iz2,jz1);
1602 dx22 = _mm256_sub_pd(ix2,jx2);
1603 dy22 = _mm256_sub_pd(iy2,jy2);
1604 dz22 = _mm256_sub_pd(iz2,jz2);
1606 /* Calculate squared distance and things based on it */
1607 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1608 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1609 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1610 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1611 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1612 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1613 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1614 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1615 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1617 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1618 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1619 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1620 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1621 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1622 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1623 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1624 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1625 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1627 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
1628 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
1629 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
1630 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1631 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
1632 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
1633 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1634 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
1635 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
1637 fjx0 = _mm256_setzero_pd();
1638 fjy0 = _mm256_setzero_pd();
1639 fjz0 = _mm256_setzero_pd();
1640 fjx1 = _mm256_setzero_pd();
1641 fjy1 = _mm256_setzero_pd();
1642 fjz1 = _mm256_setzero_pd();
1643 fjx2 = _mm256_setzero_pd();
1644 fjy2 = _mm256_setzero_pd();
1645 fjz2 = _mm256_setzero_pd();
1647 /**************************
1648 * CALCULATE INTERACTIONS *
1649 **************************/
1651 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1654 r00 = _mm256_mul_pd(rsq00,rinv00);
1656 /* EWALD ELECTROSTATICS */
1658 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1659 ewrt = _mm256_mul_pd(r00,ewtabscale);
1660 ewitab = _mm256_cvttpd_epi32(ewrt);
1661 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1662 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1663 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1665 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1666 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1668 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1672 fscal = _mm256_and_pd(fscal,cutoff_mask);
1674 /* Calculate temporary vectorial force */
1675 tx = _mm256_mul_pd(fscal,dx00);
1676 ty = _mm256_mul_pd(fscal,dy00);
1677 tz = _mm256_mul_pd(fscal,dz00);
1679 /* Update vectorial force */
1680 fix0 = _mm256_add_pd(fix0,tx);
1681 fiy0 = _mm256_add_pd(fiy0,ty);
1682 fiz0 = _mm256_add_pd(fiz0,tz);
1684 fjx0 = _mm256_add_pd(fjx0,tx);
1685 fjy0 = _mm256_add_pd(fjy0,ty);
1686 fjz0 = _mm256_add_pd(fjz0,tz);
1690 /**************************
1691 * CALCULATE INTERACTIONS *
1692 **************************/
1694 if (gmx_mm256_any_lt(rsq01,rcutoff2))
1697 r01 = _mm256_mul_pd(rsq01,rinv01);
1699 /* EWALD ELECTROSTATICS */
1701 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1702 ewrt = _mm256_mul_pd(r01,ewtabscale);
1703 ewitab = _mm256_cvttpd_epi32(ewrt);
1704 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1705 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1706 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1708 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1709 felec = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
1711 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
1715 fscal = _mm256_and_pd(fscal,cutoff_mask);
1717 /* Calculate temporary vectorial force */
1718 tx = _mm256_mul_pd(fscal,dx01);
1719 ty = _mm256_mul_pd(fscal,dy01);
1720 tz = _mm256_mul_pd(fscal,dz01);
1722 /* Update vectorial force */
1723 fix0 = _mm256_add_pd(fix0,tx);
1724 fiy0 = _mm256_add_pd(fiy0,ty);
1725 fiz0 = _mm256_add_pd(fiz0,tz);
1727 fjx1 = _mm256_add_pd(fjx1,tx);
1728 fjy1 = _mm256_add_pd(fjy1,ty);
1729 fjz1 = _mm256_add_pd(fjz1,tz);
1733 /**************************
1734 * CALCULATE INTERACTIONS *
1735 **************************/
1737 if (gmx_mm256_any_lt(rsq02,rcutoff2))
1740 r02 = _mm256_mul_pd(rsq02,rinv02);
1742 /* EWALD ELECTROSTATICS */
1744 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1745 ewrt = _mm256_mul_pd(r02,ewtabscale);
1746 ewitab = _mm256_cvttpd_epi32(ewrt);
1747 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1748 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1749 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1751 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1752 felec = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
1754 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
1758 fscal = _mm256_and_pd(fscal,cutoff_mask);
1760 /* Calculate temporary vectorial force */
1761 tx = _mm256_mul_pd(fscal,dx02);
1762 ty = _mm256_mul_pd(fscal,dy02);
1763 tz = _mm256_mul_pd(fscal,dz02);
1765 /* Update vectorial force */
1766 fix0 = _mm256_add_pd(fix0,tx);
1767 fiy0 = _mm256_add_pd(fiy0,ty);
1768 fiz0 = _mm256_add_pd(fiz0,tz);
1770 fjx2 = _mm256_add_pd(fjx2,tx);
1771 fjy2 = _mm256_add_pd(fjy2,ty);
1772 fjz2 = _mm256_add_pd(fjz2,tz);
1776 /**************************
1777 * CALCULATE INTERACTIONS *
1778 **************************/
1780 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1783 r10 = _mm256_mul_pd(rsq10,rinv10);
1785 /* EWALD ELECTROSTATICS */
1787 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1788 ewrt = _mm256_mul_pd(r10,ewtabscale);
1789 ewitab = _mm256_cvttpd_epi32(ewrt);
1790 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1791 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1792 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1794 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1795 felec = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1797 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1801 fscal = _mm256_and_pd(fscal,cutoff_mask);
1803 /* Calculate temporary vectorial force */
1804 tx = _mm256_mul_pd(fscal,dx10);
1805 ty = _mm256_mul_pd(fscal,dy10);
1806 tz = _mm256_mul_pd(fscal,dz10);
1808 /* Update vectorial force */
1809 fix1 = _mm256_add_pd(fix1,tx);
1810 fiy1 = _mm256_add_pd(fiy1,ty);
1811 fiz1 = _mm256_add_pd(fiz1,tz);
1813 fjx0 = _mm256_add_pd(fjx0,tx);
1814 fjy0 = _mm256_add_pd(fjy0,ty);
1815 fjz0 = _mm256_add_pd(fjz0,tz);
1819 /**************************
1820 * CALCULATE INTERACTIONS *
1821 **************************/
1823 if (gmx_mm256_any_lt(rsq11,rcutoff2))
1826 r11 = _mm256_mul_pd(rsq11,rinv11);
1828 /* EWALD ELECTROSTATICS */
1830 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1831 ewrt = _mm256_mul_pd(r11,ewtabscale);
1832 ewitab = _mm256_cvttpd_epi32(ewrt);
1833 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1834 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1835 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1837 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1838 felec = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1840 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1844 fscal = _mm256_and_pd(fscal,cutoff_mask);
1846 /* Calculate temporary vectorial force */
1847 tx = _mm256_mul_pd(fscal,dx11);
1848 ty = _mm256_mul_pd(fscal,dy11);
1849 tz = _mm256_mul_pd(fscal,dz11);
1851 /* Update vectorial force */
1852 fix1 = _mm256_add_pd(fix1,tx);
1853 fiy1 = _mm256_add_pd(fiy1,ty);
1854 fiz1 = _mm256_add_pd(fiz1,tz);
1856 fjx1 = _mm256_add_pd(fjx1,tx);
1857 fjy1 = _mm256_add_pd(fjy1,ty);
1858 fjz1 = _mm256_add_pd(fjz1,tz);
1862 /**************************
1863 * CALCULATE INTERACTIONS *
1864 **************************/
1866 if (gmx_mm256_any_lt(rsq12,rcutoff2))
1869 r12 = _mm256_mul_pd(rsq12,rinv12);
1871 /* EWALD ELECTROSTATICS */
1873 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1874 ewrt = _mm256_mul_pd(r12,ewtabscale);
1875 ewitab = _mm256_cvttpd_epi32(ewrt);
1876 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1877 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1878 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1880 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1881 felec = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1883 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1887 fscal = _mm256_and_pd(fscal,cutoff_mask);
1889 /* Calculate temporary vectorial force */
1890 tx = _mm256_mul_pd(fscal,dx12);
1891 ty = _mm256_mul_pd(fscal,dy12);
1892 tz = _mm256_mul_pd(fscal,dz12);
1894 /* Update vectorial force */
1895 fix1 = _mm256_add_pd(fix1,tx);
1896 fiy1 = _mm256_add_pd(fiy1,ty);
1897 fiz1 = _mm256_add_pd(fiz1,tz);
1899 fjx2 = _mm256_add_pd(fjx2,tx);
1900 fjy2 = _mm256_add_pd(fjy2,ty);
1901 fjz2 = _mm256_add_pd(fjz2,tz);
1905 /**************************
1906 * CALCULATE INTERACTIONS *
1907 **************************/
1909 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1912 r20 = _mm256_mul_pd(rsq20,rinv20);
1914 /* EWALD ELECTROSTATICS */
1916 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1917 ewrt = _mm256_mul_pd(r20,ewtabscale);
1918 ewitab = _mm256_cvttpd_epi32(ewrt);
1919 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1920 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1921 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1923 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1924 felec = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1926 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1930 fscal = _mm256_and_pd(fscal,cutoff_mask);
1932 /* Calculate temporary vectorial force */
1933 tx = _mm256_mul_pd(fscal,dx20);
1934 ty = _mm256_mul_pd(fscal,dy20);
1935 tz = _mm256_mul_pd(fscal,dz20);
1937 /* Update vectorial force */
1938 fix2 = _mm256_add_pd(fix2,tx);
1939 fiy2 = _mm256_add_pd(fiy2,ty);
1940 fiz2 = _mm256_add_pd(fiz2,tz);
1942 fjx0 = _mm256_add_pd(fjx0,tx);
1943 fjy0 = _mm256_add_pd(fjy0,ty);
1944 fjz0 = _mm256_add_pd(fjz0,tz);
1948 /**************************
1949 * CALCULATE INTERACTIONS *
1950 **************************/
1952 if (gmx_mm256_any_lt(rsq21,rcutoff2))
1955 r21 = _mm256_mul_pd(rsq21,rinv21);
1957 /* EWALD ELECTROSTATICS */
1959 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1960 ewrt = _mm256_mul_pd(r21,ewtabscale);
1961 ewitab = _mm256_cvttpd_epi32(ewrt);
1962 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1963 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1964 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1966 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1967 felec = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1969 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1973 fscal = _mm256_and_pd(fscal,cutoff_mask);
1975 /* Calculate temporary vectorial force */
1976 tx = _mm256_mul_pd(fscal,dx21);
1977 ty = _mm256_mul_pd(fscal,dy21);
1978 tz = _mm256_mul_pd(fscal,dz21);
1980 /* Update vectorial force */
1981 fix2 = _mm256_add_pd(fix2,tx);
1982 fiy2 = _mm256_add_pd(fiy2,ty);
1983 fiz2 = _mm256_add_pd(fiz2,tz);
1985 fjx1 = _mm256_add_pd(fjx1,tx);
1986 fjy1 = _mm256_add_pd(fjy1,ty);
1987 fjz1 = _mm256_add_pd(fjz1,tz);
1991 /**************************
1992 * CALCULATE INTERACTIONS *
1993 **************************/
1995 if (gmx_mm256_any_lt(rsq22,rcutoff2))
1998 r22 = _mm256_mul_pd(rsq22,rinv22);
2000 /* EWALD ELECTROSTATICS */
2002 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2003 ewrt = _mm256_mul_pd(r22,ewtabscale);
2004 ewitab = _mm256_cvttpd_epi32(ewrt);
2005 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2006 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2007 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2009 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2010 felec = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2012 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2016 fscal = _mm256_and_pd(fscal,cutoff_mask);
2018 /* Calculate temporary vectorial force */
2019 tx = _mm256_mul_pd(fscal,dx22);
2020 ty = _mm256_mul_pd(fscal,dy22);
2021 tz = _mm256_mul_pd(fscal,dz22);
2023 /* Update vectorial force */
2024 fix2 = _mm256_add_pd(fix2,tx);
2025 fiy2 = _mm256_add_pd(fiy2,ty);
2026 fiz2 = _mm256_add_pd(fiz2,tz);
2028 fjx2 = _mm256_add_pd(fjx2,tx);
2029 fjy2 = _mm256_add_pd(fjy2,ty);
2030 fjz2 = _mm256_add_pd(fjz2,tz);
2034 fjptrA = f+j_coord_offsetA;
2035 fjptrB = f+j_coord_offsetB;
2036 fjptrC = f+j_coord_offsetC;
2037 fjptrD = f+j_coord_offsetD;
2039 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2040 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2042 /* Inner loop uses 351 flops */
2045 if(jidx<j_index_end)
2048 /* Get j neighbor index, and coordinate index */
2049 jnrlistA = jjnr[jidx];
2050 jnrlistB = jjnr[jidx+1];
2051 jnrlistC = jjnr[jidx+2];
2052 jnrlistD = jjnr[jidx+3];
2053 /* Sign of each element will be negative for non-real atoms.
2054 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2055 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
2057 tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2059 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
2060 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
2061 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
2063 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
2064 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
2065 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
2066 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
2067 j_coord_offsetA = DIM*jnrA;
2068 j_coord_offsetB = DIM*jnrB;
2069 j_coord_offsetC = DIM*jnrC;
2070 j_coord_offsetD = DIM*jnrD;
2072 /* load j atom coordinates */
2073 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
2074 x+j_coord_offsetC,x+j_coord_offsetD,
2075 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2077 /* Calculate displacement vector */
2078 dx00 = _mm256_sub_pd(ix0,jx0);
2079 dy00 = _mm256_sub_pd(iy0,jy0);
2080 dz00 = _mm256_sub_pd(iz0,jz0);
2081 dx01 = _mm256_sub_pd(ix0,jx1);
2082 dy01 = _mm256_sub_pd(iy0,jy1);
2083 dz01 = _mm256_sub_pd(iz0,jz1);
2084 dx02 = _mm256_sub_pd(ix0,jx2);
2085 dy02 = _mm256_sub_pd(iy0,jy2);
2086 dz02 = _mm256_sub_pd(iz0,jz2);
2087 dx10 = _mm256_sub_pd(ix1,jx0);
2088 dy10 = _mm256_sub_pd(iy1,jy0);
2089 dz10 = _mm256_sub_pd(iz1,jz0);
2090 dx11 = _mm256_sub_pd(ix1,jx1);
2091 dy11 = _mm256_sub_pd(iy1,jy1);
2092 dz11 = _mm256_sub_pd(iz1,jz1);
2093 dx12 = _mm256_sub_pd(ix1,jx2);
2094 dy12 = _mm256_sub_pd(iy1,jy2);
2095 dz12 = _mm256_sub_pd(iz1,jz2);
2096 dx20 = _mm256_sub_pd(ix2,jx0);
2097 dy20 = _mm256_sub_pd(iy2,jy0);
2098 dz20 = _mm256_sub_pd(iz2,jz0);
2099 dx21 = _mm256_sub_pd(ix2,jx1);
2100 dy21 = _mm256_sub_pd(iy2,jy1);
2101 dz21 = _mm256_sub_pd(iz2,jz1);
2102 dx22 = _mm256_sub_pd(ix2,jx2);
2103 dy22 = _mm256_sub_pd(iy2,jy2);
2104 dz22 = _mm256_sub_pd(iz2,jz2);
2106 /* Calculate squared distance and things based on it */
2107 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
2108 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
2109 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
2110 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
2111 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2112 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2113 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
2114 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2115 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2117 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
2118 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
2119 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
2120 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
2121 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
2122 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
2123 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
2124 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
2125 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
2127 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
2128 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
2129 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
2130 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
2131 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
2132 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
2133 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
2134 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
2135 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
2137 fjx0 = _mm256_setzero_pd();
2138 fjy0 = _mm256_setzero_pd();
2139 fjz0 = _mm256_setzero_pd();
2140 fjx1 = _mm256_setzero_pd();
2141 fjy1 = _mm256_setzero_pd();
2142 fjz1 = _mm256_setzero_pd();
2143 fjx2 = _mm256_setzero_pd();
2144 fjy2 = _mm256_setzero_pd();
2145 fjz2 = _mm256_setzero_pd();
2147 /**************************
2148 * CALCULATE INTERACTIONS *
2149 **************************/
2151 if (gmx_mm256_any_lt(rsq00,rcutoff2))
2154 r00 = _mm256_mul_pd(rsq00,rinv00);
2155 r00 = _mm256_andnot_pd(dummy_mask,r00);
2157 /* EWALD ELECTROSTATICS */
2159 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2160 ewrt = _mm256_mul_pd(r00,ewtabscale);
2161 ewitab = _mm256_cvttpd_epi32(ewrt);
2162 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2163 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2164 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2166 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2167 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
2169 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
2173 fscal = _mm256_and_pd(fscal,cutoff_mask);
2175 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2177 /* Calculate temporary vectorial force */
2178 tx = _mm256_mul_pd(fscal,dx00);
2179 ty = _mm256_mul_pd(fscal,dy00);
2180 tz = _mm256_mul_pd(fscal,dz00);
2182 /* Update vectorial force */
2183 fix0 = _mm256_add_pd(fix0,tx);
2184 fiy0 = _mm256_add_pd(fiy0,ty);
2185 fiz0 = _mm256_add_pd(fiz0,tz);
2187 fjx0 = _mm256_add_pd(fjx0,tx);
2188 fjy0 = _mm256_add_pd(fjy0,ty);
2189 fjz0 = _mm256_add_pd(fjz0,tz);
2193 /**************************
2194 * CALCULATE INTERACTIONS *
2195 **************************/
2197 if (gmx_mm256_any_lt(rsq01,rcutoff2))
2200 r01 = _mm256_mul_pd(rsq01,rinv01);
2201 r01 = _mm256_andnot_pd(dummy_mask,r01);
2203 /* EWALD ELECTROSTATICS */
2205 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2206 ewrt = _mm256_mul_pd(r01,ewtabscale);
2207 ewitab = _mm256_cvttpd_epi32(ewrt);
2208 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2209 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2210 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2212 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2213 felec = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
2215 cutoff_mask = _mm256_cmp_pd(rsq01,rcutoff2,_CMP_LT_OQ);
2219 fscal = _mm256_and_pd(fscal,cutoff_mask);
2221 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2223 /* Calculate temporary vectorial force */
2224 tx = _mm256_mul_pd(fscal,dx01);
2225 ty = _mm256_mul_pd(fscal,dy01);
2226 tz = _mm256_mul_pd(fscal,dz01);
2228 /* Update vectorial force */
2229 fix0 = _mm256_add_pd(fix0,tx);
2230 fiy0 = _mm256_add_pd(fiy0,ty);
2231 fiz0 = _mm256_add_pd(fiz0,tz);
2233 fjx1 = _mm256_add_pd(fjx1,tx);
2234 fjy1 = _mm256_add_pd(fjy1,ty);
2235 fjz1 = _mm256_add_pd(fjz1,tz);
2239 /**************************
2240 * CALCULATE INTERACTIONS *
2241 **************************/
2243 if (gmx_mm256_any_lt(rsq02,rcutoff2))
2246 r02 = _mm256_mul_pd(rsq02,rinv02);
2247 r02 = _mm256_andnot_pd(dummy_mask,r02);
2249 /* EWALD ELECTROSTATICS */
2251 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2252 ewrt = _mm256_mul_pd(r02,ewtabscale);
2253 ewitab = _mm256_cvttpd_epi32(ewrt);
2254 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2255 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2256 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2258 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2259 felec = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
2261 cutoff_mask = _mm256_cmp_pd(rsq02,rcutoff2,_CMP_LT_OQ);
2265 fscal = _mm256_and_pd(fscal,cutoff_mask);
2267 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2269 /* Calculate temporary vectorial force */
2270 tx = _mm256_mul_pd(fscal,dx02);
2271 ty = _mm256_mul_pd(fscal,dy02);
2272 tz = _mm256_mul_pd(fscal,dz02);
2274 /* Update vectorial force */
2275 fix0 = _mm256_add_pd(fix0,tx);
2276 fiy0 = _mm256_add_pd(fiy0,ty);
2277 fiz0 = _mm256_add_pd(fiz0,tz);
2279 fjx2 = _mm256_add_pd(fjx2,tx);
2280 fjy2 = _mm256_add_pd(fjy2,ty);
2281 fjz2 = _mm256_add_pd(fjz2,tz);
2285 /**************************
2286 * CALCULATE INTERACTIONS *
2287 **************************/
2289 if (gmx_mm256_any_lt(rsq10,rcutoff2))
2292 r10 = _mm256_mul_pd(rsq10,rinv10);
2293 r10 = _mm256_andnot_pd(dummy_mask,r10);
2295 /* EWALD ELECTROSTATICS */
2297 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2298 ewrt = _mm256_mul_pd(r10,ewtabscale);
2299 ewitab = _mm256_cvttpd_epi32(ewrt);
2300 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2301 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2302 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2304 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2305 felec = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
2307 cutoff_mask = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
2311 fscal = _mm256_and_pd(fscal,cutoff_mask);
2313 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2315 /* Calculate temporary vectorial force */
2316 tx = _mm256_mul_pd(fscal,dx10);
2317 ty = _mm256_mul_pd(fscal,dy10);
2318 tz = _mm256_mul_pd(fscal,dz10);
2320 /* Update vectorial force */
2321 fix1 = _mm256_add_pd(fix1,tx);
2322 fiy1 = _mm256_add_pd(fiy1,ty);
2323 fiz1 = _mm256_add_pd(fiz1,tz);
2325 fjx0 = _mm256_add_pd(fjx0,tx);
2326 fjy0 = _mm256_add_pd(fjy0,ty);
2327 fjz0 = _mm256_add_pd(fjz0,tz);
2331 /**************************
2332 * CALCULATE INTERACTIONS *
2333 **************************/
2335 if (gmx_mm256_any_lt(rsq11,rcutoff2))
2338 r11 = _mm256_mul_pd(rsq11,rinv11);
2339 r11 = _mm256_andnot_pd(dummy_mask,r11);
2341 /* EWALD ELECTROSTATICS */
2343 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2344 ewrt = _mm256_mul_pd(r11,ewtabscale);
2345 ewitab = _mm256_cvttpd_epi32(ewrt);
2346 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2347 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2348 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2350 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2351 felec = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2353 cutoff_mask = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2357 fscal = _mm256_and_pd(fscal,cutoff_mask);
2359 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2361 /* Calculate temporary vectorial force */
2362 tx = _mm256_mul_pd(fscal,dx11);
2363 ty = _mm256_mul_pd(fscal,dy11);
2364 tz = _mm256_mul_pd(fscal,dz11);
2366 /* Update vectorial force */
2367 fix1 = _mm256_add_pd(fix1,tx);
2368 fiy1 = _mm256_add_pd(fiy1,ty);
2369 fiz1 = _mm256_add_pd(fiz1,tz);
2371 fjx1 = _mm256_add_pd(fjx1,tx);
2372 fjy1 = _mm256_add_pd(fjy1,ty);
2373 fjz1 = _mm256_add_pd(fjz1,tz);
2377 /**************************
2378 * CALCULATE INTERACTIONS *
2379 **************************/
2381 if (gmx_mm256_any_lt(rsq12,rcutoff2))
2384 r12 = _mm256_mul_pd(rsq12,rinv12);
2385 r12 = _mm256_andnot_pd(dummy_mask,r12);
2387 /* EWALD ELECTROSTATICS */
2389 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2390 ewrt = _mm256_mul_pd(r12,ewtabscale);
2391 ewitab = _mm256_cvttpd_epi32(ewrt);
2392 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2393 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2394 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2396 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2397 felec = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2399 cutoff_mask = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2403 fscal = _mm256_and_pd(fscal,cutoff_mask);
2405 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2407 /* Calculate temporary vectorial force */
2408 tx = _mm256_mul_pd(fscal,dx12);
2409 ty = _mm256_mul_pd(fscal,dy12);
2410 tz = _mm256_mul_pd(fscal,dz12);
2412 /* Update vectorial force */
2413 fix1 = _mm256_add_pd(fix1,tx);
2414 fiy1 = _mm256_add_pd(fiy1,ty);
2415 fiz1 = _mm256_add_pd(fiz1,tz);
2417 fjx2 = _mm256_add_pd(fjx2,tx);
2418 fjy2 = _mm256_add_pd(fjy2,ty);
2419 fjz2 = _mm256_add_pd(fjz2,tz);
2423 /**************************
2424 * CALCULATE INTERACTIONS *
2425 **************************/
2427 if (gmx_mm256_any_lt(rsq20,rcutoff2))
2430 r20 = _mm256_mul_pd(rsq20,rinv20);
2431 r20 = _mm256_andnot_pd(dummy_mask,r20);
2433 /* EWALD ELECTROSTATICS */
2435 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2436 ewrt = _mm256_mul_pd(r20,ewtabscale);
2437 ewitab = _mm256_cvttpd_epi32(ewrt);
2438 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2439 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2440 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2442 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2443 felec = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
2445 cutoff_mask = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
2449 fscal = _mm256_and_pd(fscal,cutoff_mask);
2451 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2453 /* Calculate temporary vectorial force */
2454 tx = _mm256_mul_pd(fscal,dx20);
2455 ty = _mm256_mul_pd(fscal,dy20);
2456 tz = _mm256_mul_pd(fscal,dz20);
2458 /* Update vectorial force */
2459 fix2 = _mm256_add_pd(fix2,tx);
2460 fiy2 = _mm256_add_pd(fiy2,ty);
2461 fiz2 = _mm256_add_pd(fiz2,tz);
2463 fjx0 = _mm256_add_pd(fjx0,tx);
2464 fjy0 = _mm256_add_pd(fjy0,ty);
2465 fjz0 = _mm256_add_pd(fjz0,tz);
2469 /**************************
2470 * CALCULATE INTERACTIONS *
2471 **************************/
2473 if (gmx_mm256_any_lt(rsq21,rcutoff2))
2476 r21 = _mm256_mul_pd(rsq21,rinv21);
2477 r21 = _mm256_andnot_pd(dummy_mask,r21);
2479 /* EWALD ELECTROSTATICS */
2481 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2482 ewrt = _mm256_mul_pd(r21,ewtabscale);
2483 ewitab = _mm256_cvttpd_epi32(ewrt);
2484 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2485 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2486 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2488 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2489 felec = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2491 cutoff_mask = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2495 fscal = _mm256_and_pd(fscal,cutoff_mask);
2497 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2499 /* Calculate temporary vectorial force */
2500 tx = _mm256_mul_pd(fscal,dx21);
2501 ty = _mm256_mul_pd(fscal,dy21);
2502 tz = _mm256_mul_pd(fscal,dz21);
2504 /* Update vectorial force */
2505 fix2 = _mm256_add_pd(fix2,tx);
2506 fiy2 = _mm256_add_pd(fiy2,ty);
2507 fiz2 = _mm256_add_pd(fiz2,tz);
2509 fjx1 = _mm256_add_pd(fjx1,tx);
2510 fjy1 = _mm256_add_pd(fjy1,ty);
2511 fjz1 = _mm256_add_pd(fjz1,tz);
2515 /**************************
2516 * CALCULATE INTERACTIONS *
2517 **************************/
2519 if (gmx_mm256_any_lt(rsq22,rcutoff2))
2522 r22 = _mm256_mul_pd(rsq22,rinv22);
2523 r22 = _mm256_andnot_pd(dummy_mask,r22);
2525 /* EWALD ELECTROSTATICS */
2527 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2528 ewrt = _mm256_mul_pd(r22,ewtabscale);
2529 ewitab = _mm256_cvttpd_epi32(ewrt);
2530 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2531 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2532 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2534 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2535 felec = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2537 cutoff_mask = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2541 fscal = _mm256_and_pd(fscal,cutoff_mask);
2543 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2545 /* Calculate temporary vectorial force */
2546 tx = _mm256_mul_pd(fscal,dx22);
2547 ty = _mm256_mul_pd(fscal,dy22);
2548 tz = _mm256_mul_pd(fscal,dz22);
2550 /* Update vectorial force */
2551 fix2 = _mm256_add_pd(fix2,tx);
2552 fiy2 = _mm256_add_pd(fiy2,ty);
2553 fiz2 = _mm256_add_pd(fiz2,tz);
2555 fjx2 = _mm256_add_pd(fjx2,tx);
2556 fjy2 = _mm256_add_pd(fjy2,ty);
2557 fjz2 = _mm256_add_pd(fjz2,tz);
2561 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2562 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2563 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2564 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2566 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2567 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2569 /* Inner loop uses 360 flops */
2572 /* End of innermost loop */
2574 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2575 f+i_coord_offset,fshift+i_shift_offset);
2577 /* Increment number of inner iterations */
2578 inneriter += j_index_end - j_index_start;
2580 /* Outer loop uses 18 flops */
2583 /* Increment number of outer iterations */
2586 /* Update outer/inner flops */
2588 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*360);