2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_256_double.h"
48 #include "kernelutil_x86_avx_256_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3W3_VF_avx_256_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: None
54 * Geometry: Water3-Water3
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEw_VdwNone_GeomW3W3_VF_avx_256_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
83 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 real * vdwioffsetptr0;
85 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 real * vdwioffsetptr1;
87 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 real * vdwioffsetptr2;
89 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
93 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
94 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
95 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
96 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
97 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
98 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
99 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
100 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
101 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
102 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
103 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
104 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
105 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
108 __m256d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
109 __m256d beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
111 __m256d dummy_mask,cutoff_mask;
112 __m128 tmpmask0,tmpmask1;
113 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
114 __m256d one = _mm256_set1_pd(1.0);
115 __m256d two = _mm256_set1_pd(2.0);
121 jindex = nlist->jindex;
123 shiftidx = nlist->shift;
125 shiftvec = fr->shift_vec[0];
126 fshift = fr->fshift[0];
127 facel = _mm256_set1_pd(fr->epsfac);
128 charge = mdatoms->chargeA;
130 sh_ewald = _mm256_set1_pd(fr->ic->sh_ewald);
131 beta = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
132 beta2 = _mm256_mul_pd(beta,beta);
133 beta3 = _mm256_mul_pd(beta,beta2);
135 ewtab = fr->ic->tabq_coul_FDV0;
136 ewtabscale = _mm256_set1_pd(fr->ic->tabq_scale);
137 ewtabhalfspace = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
139 /* Setup water-specific parameters */
140 inr = nlist->iinr[0];
141 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
142 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
143 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
145 jq0 = _mm256_set1_pd(charge[inr+0]);
146 jq1 = _mm256_set1_pd(charge[inr+1]);
147 jq2 = _mm256_set1_pd(charge[inr+2]);
148 qq00 = _mm256_mul_pd(iq0,jq0);
149 qq01 = _mm256_mul_pd(iq0,jq1);
150 qq02 = _mm256_mul_pd(iq0,jq2);
151 qq10 = _mm256_mul_pd(iq1,jq0);
152 qq11 = _mm256_mul_pd(iq1,jq1);
153 qq12 = _mm256_mul_pd(iq1,jq2);
154 qq20 = _mm256_mul_pd(iq2,jq0);
155 qq21 = _mm256_mul_pd(iq2,jq1);
156 qq22 = _mm256_mul_pd(iq2,jq2);
158 /* Avoid stupid compiler warnings */
159 jnrA = jnrB = jnrC = jnrD = 0;
168 for(iidx=0;iidx<4*DIM;iidx++)
173 /* Start outer loop over neighborlists */
174 for(iidx=0; iidx<nri; iidx++)
176 /* Load shift vector for this list */
177 i_shift_offset = DIM*shiftidx[iidx];
179 /* Load limits for loop over neighbors */
180 j_index_start = jindex[iidx];
181 j_index_end = jindex[iidx+1];
183 /* Get outer coordinate index */
185 i_coord_offset = DIM*inr;
187 /* Load i particle coords and add shift vector */
188 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
189 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
191 fix0 = _mm256_setzero_pd();
192 fiy0 = _mm256_setzero_pd();
193 fiz0 = _mm256_setzero_pd();
194 fix1 = _mm256_setzero_pd();
195 fiy1 = _mm256_setzero_pd();
196 fiz1 = _mm256_setzero_pd();
197 fix2 = _mm256_setzero_pd();
198 fiy2 = _mm256_setzero_pd();
199 fiz2 = _mm256_setzero_pd();
201 /* Reset potential sums */
202 velecsum = _mm256_setzero_pd();
204 /* Start inner kernel loop */
205 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
208 /* Get j neighbor index, and coordinate index */
213 j_coord_offsetA = DIM*jnrA;
214 j_coord_offsetB = DIM*jnrB;
215 j_coord_offsetC = DIM*jnrC;
216 j_coord_offsetD = DIM*jnrD;
218 /* load j atom coordinates */
219 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
220 x+j_coord_offsetC,x+j_coord_offsetD,
221 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
223 /* Calculate displacement vector */
224 dx00 = _mm256_sub_pd(ix0,jx0);
225 dy00 = _mm256_sub_pd(iy0,jy0);
226 dz00 = _mm256_sub_pd(iz0,jz0);
227 dx01 = _mm256_sub_pd(ix0,jx1);
228 dy01 = _mm256_sub_pd(iy0,jy1);
229 dz01 = _mm256_sub_pd(iz0,jz1);
230 dx02 = _mm256_sub_pd(ix0,jx2);
231 dy02 = _mm256_sub_pd(iy0,jy2);
232 dz02 = _mm256_sub_pd(iz0,jz2);
233 dx10 = _mm256_sub_pd(ix1,jx0);
234 dy10 = _mm256_sub_pd(iy1,jy0);
235 dz10 = _mm256_sub_pd(iz1,jz0);
236 dx11 = _mm256_sub_pd(ix1,jx1);
237 dy11 = _mm256_sub_pd(iy1,jy1);
238 dz11 = _mm256_sub_pd(iz1,jz1);
239 dx12 = _mm256_sub_pd(ix1,jx2);
240 dy12 = _mm256_sub_pd(iy1,jy2);
241 dz12 = _mm256_sub_pd(iz1,jz2);
242 dx20 = _mm256_sub_pd(ix2,jx0);
243 dy20 = _mm256_sub_pd(iy2,jy0);
244 dz20 = _mm256_sub_pd(iz2,jz0);
245 dx21 = _mm256_sub_pd(ix2,jx1);
246 dy21 = _mm256_sub_pd(iy2,jy1);
247 dz21 = _mm256_sub_pd(iz2,jz1);
248 dx22 = _mm256_sub_pd(ix2,jx2);
249 dy22 = _mm256_sub_pd(iy2,jy2);
250 dz22 = _mm256_sub_pd(iz2,jz2);
252 /* Calculate squared distance and things based on it */
253 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
254 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
255 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
256 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
257 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
258 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
259 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
260 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
261 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
263 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
264 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
265 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
266 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
267 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
268 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
269 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
270 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
271 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
273 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
274 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
275 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
276 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
277 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
278 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
279 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
280 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
281 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
283 fjx0 = _mm256_setzero_pd();
284 fjy0 = _mm256_setzero_pd();
285 fjz0 = _mm256_setzero_pd();
286 fjx1 = _mm256_setzero_pd();
287 fjy1 = _mm256_setzero_pd();
288 fjz1 = _mm256_setzero_pd();
289 fjx2 = _mm256_setzero_pd();
290 fjy2 = _mm256_setzero_pd();
291 fjz2 = _mm256_setzero_pd();
293 /**************************
294 * CALCULATE INTERACTIONS *
295 **************************/
297 r00 = _mm256_mul_pd(rsq00,rinv00);
299 /* EWALD ELECTROSTATICS */
301 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
302 ewrt = _mm256_mul_pd(r00,ewtabscale);
303 ewitab = _mm256_cvttpd_epi32(ewrt);
304 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
305 ewitab = _mm_slli_epi32(ewitab,2);
306 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
307 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
308 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
309 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
310 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
311 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
312 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
313 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
314 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
316 /* Update potential sum for this i atom from the interaction with this j atom. */
317 velecsum = _mm256_add_pd(velecsum,velec);
321 /* Calculate temporary vectorial force */
322 tx = _mm256_mul_pd(fscal,dx00);
323 ty = _mm256_mul_pd(fscal,dy00);
324 tz = _mm256_mul_pd(fscal,dz00);
326 /* Update vectorial force */
327 fix0 = _mm256_add_pd(fix0,tx);
328 fiy0 = _mm256_add_pd(fiy0,ty);
329 fiz0 = _mm256_add_pd(fiz0,tz);
331 fjx0 = _mm256_add_pd(fjx0,tx);
332 fjy0 = _mm256_add_pd(fjy0,ty);
333 fjz0 = _mm256_add_pd(fjz0,tz);
335 /**************************
336 * CALCULATE INTERACTIONS *
337 **************************/
339 r01 = _mm256_mul_pd(rsq01,rinv01);
341 /* EWALD ELECTROSTATICS */
343 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
344 ewrt = _mm256_mul_pd(r01,ewtabscale);
345 ewitab = _mm256_cvttpd_epi32(ewrt);
346 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
347 ewitab = _mm_slli_epi32(ewitab,2);
348 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
349 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
350 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
351 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
352 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
353 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
354 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
355 velec = _mm256_mul_pd(qq01,_mm256_sub_pd(rinv01,velec));
356 felec = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
358 /* Update potential sum for this i atom from the interaction with this j atom. */
359 velecsum = _mm256_add_pd(velecsum,velec);
363 /* Calculate temporary vectorial force */
364 tx = _mm256_mul_pd(fscal,dx01);
365 ty = _mm256_mul_pd(fscal,dy01);
366 tz = _mm256_mul_pd(fscal,dz01);
368 /* Update vectorial force */
369 fix0 = _mm256_add_pd(fix0,tx);
370 fiy0 = _mm256_add_pd(fiy0,ty);
371 fiz0 = _mm256_add_pd(fiz0,tz);
373 fjx1 = _mm256_add_pd(fjx1,tx);
374 fjy1 = _mm256_add_pd(fjy1,ty);
375 fjz1 = _mm256_add_pd(fjz1,tz);
377 /**************************
378 * CALCULATE INTERACTIONS *
379 **************************/
381 r02 = _mm256_mul_pd(rsq02,rinv02);
383 /* EWALD ELECTROSTATICS */
385 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
386 ewrt = _mm256_mul_pd(r02,ewtabscale);
387 ewitab = _mm256_cvttpd_epi32(ewrt);
388 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
389 ewitab = _mm_slli_epi32(ewitab,2);
390 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
391 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
392 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
393 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
394 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
395 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
396 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
397 velec = _mm256_mul_pd(qq02,_mm256_sub_pd(rinv02,velec));
398 felec = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
400 /* Update potential sum for this i atom from the interaction with this j atom. */
401 velecsum = _mm256_add_pd(velecsum,velec);
405 /* Calculate temporary vectorial force */
406 tx = _mm256_mul_pd(fscal,dx02);
407 ty = _mm256_mul_pd(fscal,dy02);
408 tz = _mm256_mul_pd(fscal,dz02);
410 /* Update vectorial force */
411 fix0 = _mm256_add_pd(fix0,tx);
412 fiy0 = _mm256_add_pd(fiy0,ty);
413 fiz0 = _mm256_add_pd(fiz0,tz);
415 fjx2 = _mm256_add_pd(fjx2,tx);
416 fjy2 = _mm256_add_pd(fjy2,ty);
417 fjz2 = _mm256_add_pd(fjz2,tz);
419 /**************************
420 * CALCULATE INTERACTIONS *
421 **************************/
423 r10 = _mm256_mul_pd(rsq10,rinv10);
425 /* EWALD ELECTROSTATICS */
427 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
428 ewrt = _mm256_mul_pd(r10,ewtabscale);
429 ewitab = _mm256_cvttpd_epi32(ewrt);
430 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
431 ewitab = _mm_slli_epi32(ewitab,2);
432 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
433 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
434 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
435 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
436 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
437 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
438 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
439 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
440 felec = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
442 /* Update potential sum for this i atom from the interaction with this j atom. */
443 velecsum = _mm256_add_pd(velecsum,velec);
447 /* Calculate temporary vectorial force */
448 tx = _mm256_mul_pd(fscal,dx10);
449 ty = _mm256_mul_pd(fscal,dy10);
450 tz = _mm256_mul_pd(fscal,dz10);
452 /* Update vectorial force */
453 fix1 = _mm256_add_pd(fix1,tx);
454 fiy1 = _mm256_add_pd(fiy1,ty);
455 fiz1 = _mm256_add_pd(fiz1,tz);
457 fjx0 = _mm256_add_pd(fjx0,tx);
458 fjy0 = _mm256_add_pd(fjy0,ty);
459 fjz0 = _mm256_add_pd(fjz0,tz);
461 /**************************
462 * CALCULATE INTERACTIONS *
463 **************************/
465 r11 = _mm256_mul_pd(rsq11,rinv11);
467 /* EWALD ELECTROSTATICS */
469 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
470 ewrt = _mm256_mul_pd(r11,ewtabscale);
471 ewitab = _mm256_cvttpd_epi32(ewrt);
472 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
473 ewitab = _mm_slli_epi32(ewitab,2);
474 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
475 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
476 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
477 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
478 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
479 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
480 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
481 velec = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
482 felec = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
484 /* Update potential sum for this i atom from the interaction with this j atom. */
485 velecsum = _mm256_add_pd(velecsum,velec);
489 /* Calculate temporary vectorial force */
490 tx = _mm256_mul_pd(fscal,dx11);
491 ty = _mm256_mul_pd(fscal,dy11);
492 tz = _mm256_mul_pd(fscal,dz11);
494 /* Update vectorial force */
495 fix1 = _mm256_add_pd(fix1,tx);
496 fiy1 = _mm256_add_pd(fiy1,ty);
497 fiz1 = _mm256_add_pd(fiz1,tz);
499 fjx1 = _mm256_add_pd(fjx1,tx);
500 fjy1 = _mm256_add_pd(fjy1,ty);
501 fjz1 = _mm256_add_pd(fjz1,tz);
503 /**************************
504 * CALCULATE INTERACTIONS *
505 **************************/
507 r12 = _mm256_mul_pd(rsq12,rinv12);
509 /* EWALD ELECTROSTATICS */
511 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
512 ewrt = _mm256_mul_pd(r12,ewtabscale);
513 ewitab = _mm256_cvttpd_epi32(ewrt);
514 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
515 ewitab = _mm_slli_epi32(ewitab,2);
516 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
517 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
518 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
519 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
520 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
521 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
522 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
523 velec = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
524 felec = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
526 /* Update potential sum for this i atom from the interaction with this j atom. */
527 velecsum = _mm256_add_pd(velecsum,velec);
531 /* Calculate temporary vectorial force */
532 tx = _mm256_mul_pd(fscal,dx12);
533 ty = _mm256_mul_pd(fscal,dy12);
534 tz = _mm256_mul_pd(fscal,dz12);
536 /* Update vectorial force */
537 fix1 = _mm256_add_pd(fix1,tx);
538 fiy1 = _mm256_add_pd(fiy1,ty);
539 fiz1 = _mm256_add_pd(fiz1,tz);
541 fjx2 = _mm256_add_pd(fjx2,tx);
542 fjy2 = _mm256_add_pd(fjy2,ty);
543 fjz2 = _mm256_add_pd(fjz2,tz);
545 /**************************
546 * CALCULATE INTERACTIONS *
547 **************************/
549 r20 = _mm256_mul_pd(rsq20,rinv20);
551 /* EWALD ELECTROSTATICS */
553 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
554 ewrt = _mm256_mul_pd(r20,ewtabscale);
555 ewitab = _mm256_cvttpd_epi32(ewrt);
556 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
557 ewitab = _mm_slli_epi32(ewitab,2);
558 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
559 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
560 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
561 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
562 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
563 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
564 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
565 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
566 felec = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
568 /* Update potential sum for this i atom from the interaction with this j atom. */
569 velecsum = _mm256_add_pd(velecsum,velec);
573 /* Calculate temporary vectorial force */
574 tx = _mm256_mul_pd(fscal,dx20);
575 ty = _mm256_mul_pd(fscal,dy20);
576 tz = _mm256_mul_pd(fscal,dz20);
578 /* Update vectorial force */
579 fix2 = _mm256_add_pd(fix2,tx);
580 fiy2 = _mm256_add_pd(fiy2,ty);
581 fiz2 = _mm256_add_pd(fiz2,tz);
583 fjx0 = _mm256_add_pd(fjx0,tx);
584 fjy0 = _mm256_add_pd(fjy0,ty);
585 fjz0 = _mm256_add_pd(fjz0,tz);
587 /**************************
588 * CALCULATE INTERACTIONS *
589 **************************/
591 r21 = _mm256_mul_pd(rsq21,rinv21);
593 /* EWALD ELECTROSTATICS */
595 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
596 ewrt = _mm256_mul_pd(r21,ewtabscale);
597 ewitab = _mm256_cvttpd_epi32(ewrt);
598 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
599 ewitab = _mm_slli_epi32(ewitab,2);
600 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
601 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
602 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
603 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
604 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
605 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
606 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
607 velec = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
608 felec = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
610 /* Update potential sum for this i atom from the interaction with this j atom. */
611 velecsum = _mm256_add_pd(velecsum,velec);
615 /* Calculate temporary vectorial force */
616 tx = _mm256_mul_pd(fscal,dx21);
617 ty = _mm256_mul_pd(fscal,dy21);
618 tz = _mm256_mul_pd(fscal,dz21);
620 /* Update vectorial force */
621 fix2 = _mm256_add_pd(fix2,tx);
622 fiy2 = _mm256_add_pd(fiy2,ty);
623 fiz2 = _mm256_add_pd(fiz2,tz);
625 fjx1 = _mm256_add_pd(fjx1,tx);
626 fjy1 = _mm256_add_pd(fjy1,ty);
627 fjz1 = _mm256_add_pd(fjz1,tz);
629 /**************************
630 * CALCULATE INTERACTIONS *
631 **************************/
633 r22 = _mm256_mul_pd(rsq22,rinv22);
635 /* EWALD ELECTROSTATICS */
637 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
638 ewrt = _mm256_mul_pd(r22,ewtabscale);
639 ewitab = _mm256_cvttpd_epi32(ewrt);
640 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
641 ewitab = _mm_slli_epi32(ewitab,2);
642 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
643 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
644 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
645 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
646 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
647 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
648 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
649 velec = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
650 felec = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
652 /* Update potential sum for this i atom from the interaction with this j atom. */
653 velecsum = _mm256_add_pd(velecsum,velec);
657 /* Calculate temporary vectorial force */
658 tx = _mm256_mul_pd(fscal,dx22);
659 ty = _mm256_mul_pd(fscal,dy22);
660 tz = _mm256_mul_pd(fscal,dz22);
662 /* Update vectorial force */
663 fix2 = _mm256_add_pd(fix2,tx);
664 fiy2 = _mm256_add_pd(fiy2,ty);
665 fiz2 = _mm256_add_pd(fiz2,tz);
667 fjx2 = _mm256_add_pd(fjx2,tx);
668 fjy2 = _mm256_add_pd(fjy2,ty);
669 fjz2 = _mm256_add_pd(fjz2,tz);
671 fjptrA = f+j_coord_offsetA;
672 fjptrB = f+j_coord_offsetB;
673 fjptrC = f+j_coord_offsetC;
674 fjptrD = f+j_coord_offsetD;
676 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
677 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
679 /* Inner loop uses 369 flops */
685 /* Get j neighbor index, and coordinate index */
686 jnrlistA = jjnr[jidx];
687 jnrlistB = jjnr[jidx+1];
688 jnrlistC = jjnr[jidx+2];
689 jnrlistD = jjnr[jidx+3];
690 /* Sign of each element will be negative for non-real atoms.
691 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
692 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
694 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
696 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
697 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
698 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
700 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
701 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
702 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
703 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
704 j_coord_offsetA = DIM*jnrA;
705 j_coord_offsetB = DIM*jnrB;
706 j_coord_offsetC = DIM*jnrC;
707 j_coord_offsetD = DIM*jnrD;
709 /* load j atom coordinates */
710 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
711 x+j_coord_offsetC,x+j_coord_offsetD,
712 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
714 /* Calculate displacement vector */
715 dx00 = _mm256_sub_pd(ix0,jx0);
716 dy00 = _mm256_sub_pd(iy0,jy0);
717 dz00 = _mm256_sub_pd(iz0,jz0);
718 dx01 = _mm256_sub_pd(ix0,jx1);
719 dy01 = _mm256_sub_pd(iy0,jy1);
720 dz01 = _mm256_sub_pd(iz0,jz1);
721 dx02 = _mm256_sub_pd(ix0,jx2);
722 dy02 = _mm256_sub_pd(iy0,jy2);
723 dz02 = _mm256_sub_pd(iz0,jz2);
724 dx10 = _mm256_sub_pd(ix1,jx0);
725 dy10 = _mm256_sub_pd(iy1,jy0);
726 dz10 = _mm256_sub_pd(iz1,jz0);
727 dx11 = _mm256_sub_pd(ix1,jx1);
728 dy11 = _mm256_sub_pd(iy1,jy1);
729 dz11 = _mm256_sub_pd(iz1,jz1);
730 dx12 = _mm256_sub_pd(ix1,jx2);
731 dy12 = _mm256_sub_pd(iy1,jy2);
732 dz12 = _mm256_sub_pd(iz1,jz2);
733 dx20 = _mm256_sub_pd(ix2,jx0);
734 dy20 = _mm256_sub_pd(iy2,jy0);
735 dz20 = _mm256_sub_pd(iz2,jz0);
736 dx21 = _mm256_sub_pd(ix2,jx1);
737 dy21 = _mm256_sub_pd(iy2,jy1);
738 dz21 = _mm256_sub_pd(iz2,jz1);
739 dx22 = _mm256_sub_pd(ix2,jx2);
740 dy22 = _mm256_sub_pd(iy2,jy2);
741 dz22 = _mm256_sub_pd(iz2,jz2);
743 /* Calculate squared distance and things based on it */
744 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
745 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
746 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
747 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
748 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
749 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
750 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
751 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
752 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
754 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
755 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
756 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
757 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
758 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
759 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
760 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
761 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
762 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
764 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
765 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
766 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
767 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
768 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
769 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
770 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
771 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
772 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
774 fjx0 = _mm256_setzero_pd();
775 fjy0 = _mm256_setzero_pd();
776 fjz0 = _mm256_setzero_pd();
777 fjx1 = _mm256_setzero_pd();
778 fjy1 = _mm256_setzero_pd();
779 fjz1 = _mm256_setzero_pd();
780 fjx2 = _mm256_setzero_pd();
781 fjy2 = _mm256_setzero_pd();
782 fjz2 = _mm256_setzero_pd();
784 /**************************
785 * CALCULATE INTERACTIONS *
786 **************************/
788 r00 = _mm256_mul_pd(rsq00,rinv00);
789 r00 = _mm256_andnot_pd(dummy_mask,r00);
791 /* EWALD ELECTROSTATICS */
793 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
794 ewrt = _mm256_mul_pd(r00,ewtabscale);
795 ewitab = _mm256_cvttpd_epi32(ewrt);
796 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
797 ewitab = _mm_slli_epi32(ewitab,2);
798 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
799 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
800 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
801 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
802 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
803 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
804 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
805 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
806 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
808 /* Update potential sum for this i atom from the interaction with this j atom. */
809 velec = _mm256_andnot_pd(dummy_mask,velec);
810 velecsum = _mm256_add_pd(velecsum,velec);
814 fscal = _mm256_andnot_pd(dummy_mask,fscal);
816 /* Calculate temporary vectorial force */
817 tx = _mm256_mul_pd(fscal,dx00);
818 ty = _mm256_mul_pd(fscal,dy00);
819 tz = _mm256_mul_pd(fscal,dz00);
821 /* Update vectorial force */
822 fix0 = _mm256_add_pd(fix0,tx);
823 fiy0 = _mm256_add_pd(fiy0,ty);
824 fiz0 = _mm256_add_pd(fiz0,tz);
826 fjx0 = _mm256_add_pd(fjx0,tx);
827 fjy0 = _mm256_add_pd(fjy0,ty);
828 fjz0 = _mm256_add_pd(fjz0,tz);
830 /**************************
831 * CALCULATE INTERACTIONS *
832 **************************/
834 r01 = _mm256_mul_pd(rsq01,rinv01);
835 r01 = _mm256_andnot_pd(dummy_mask,r01);
837 /* EWALD ELECTROSTATICS */
839 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
840 ewrt = _mm256_mul_pd(r01,ewtabscale);
841 ewitab = _mm256_cvttpd_epi32(ewrt);
842 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
843 ewitab = _mm_slli_epi32(ewitab,2);
844 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
845 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
846 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
847 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
848 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
849 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
850 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
851 velec = _mm256_mul_pd(qq01,_mm256_sub_pd(rinv01,velec));
852 felec = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
854 /* Update potential sum for this i atom from the interaction with this j atom. */
855 velec = _mm256_andnot_pd(dummy_mask,velec);
856 velecsum = _mm256_add_pd(velecsum,velec);
860 fscal = _mm256_andnot_pd(dummy_mask,fscal);
862 /* Calculate temporary vectorial force */
863 tx = _mm256_mul_pd(fscal,dx01);
864 ty = _mm256_mul_pd(fscal,dy01);
865 tz = _mm256_mul_pd(fscal,dz01);
867 /* Update vectorial force */
868 fix0 = _mm256_add_pd(fix0,tx);
869 fiy0 = _mm256_add_pd(fiy0,ty);
870 fiz0 = _mm256_add_pd(fiz0,tz);
872 fjx1 = _mm256_add_pd(fjx1,tx);
873 fjy1 = _mm256_add_pd(fjy1,ty);
874 fjz1 = _mm256_add_pd(fjz1,tz);
876 /**************************
877 * CALCULATE INTERACTIONS *
878 **************************/
880 r02 = _mm256_mul_pd(rsq02,rinv02);
881 r02 = _mm256_andnot_pd(dummy_mask,r02);
883 /* EWALD ELECTROSTATICS */
885 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
886 ewrt = _mm256_mul_pd(r02,ewtabscale);
887 ewitab = _mm256_cvttpd_epi32(ewrt);
888 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
889 ewitab = _mm_slli_epi32(ewitab,2);
890 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
891 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
892 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
893 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
894 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
895 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
896 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
897 velec = _mm256_mul_pd(qq02,_mm256_sub_pd(rinv02,velec));
898 felec = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
900 /* Update potential sum for this i atom from the interaction with this j atom. */
901 velec = _mm256_andnot_pd(dummy_mask,velec);
902 velecsum = _mm256_add_pd(velecsum,velec);
906 fscal = _mm256_andnot_pd(dummy_mask,fscal);
908 /* Calculate temporary vectorial force */
909 tx = _mm256_mul_pd(fscal,dx02);
910 ty = _mm256_mul_pd(fscal,dy02);
911 tz = _mm256_mul_pd(fscal,dz02);
913 /* Update vectorial force */
914 fix0 = _mm256_add_pd(fix0,tx);
915 fiy0 = _mm256_add_pd(fiy0,ty);
916 fiz0 = _mm256_add_pd(fiz0,tz);
918 fjx2 = _mm256_add_pd(fjx2,tx);
919 fjy2 = _mm256_add_pd(fjy2,ty);
920 fjz2 = _mm256_add_pd(fjz2,tz);
922 /**************************
923 * CALCULATE INTERACTIONS *
924 **************************/
926 r10 = _mm256_mul_pd(rsq10,rinv10);
927 r10 = _mm256_andnot_pd(dummy_mask,r10);
929 /* EWALD ELECTROSTATICS */
931 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
932 ewrt = _mm256_mul_pd(r10,ewtabscale);
933 ewitab = _mm256_cvttpd_epi32(ewrt);
934 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
935 ewitab = _mm_slli_epi32(ewitab,2);
936 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
937 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
938 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
939 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
940 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
941 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
942 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
943 velec = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
944 felec = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
946 /* Update potential sum for this i atom from the interaction with this j atom. */
947 velec = _mm256_andnot_pd(dummy_mask,velec);
948 velecsum = _mm256_add_pd(velecsum,velec);
952 fscal = _mm256_andnot_pd(dummy_mask,fscal);
954 /* Calculate temporary vectorial force */
955 tx = _mm256_mul_pd(fscal,dx10);
956 ty = _mm256_mul_pd(fscal,dy10);
957 tz = _mm256_mul_pd(fscal,dz10);
959 /* Update vectorial force */
960 fix1 = _mm256_add_pd(fix1,tx);
961 fiy1 = _mm256_add_pd(fiy1,ty);
962 fiz1 = _mm256_add_pd(fiz1,tz);
964 fjx0 = _mm256_add_pd(fjx0,tx);
965 fjy0 = _mm256_add_pd(fjy0,ty);
966 fjz0 = _mm256_add_pd(fjz0,tz);
968 /**************************
969 * CALCULATE INTERACTIONS *
970 **************************/
972 r11 = _mm256_mul_pd(rsq11,rinv11);
973 r11 = _mm256_andnot_pd(dummy_mask,r11);
975 /* EWALD ELECTROSTATICS */
977 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
978 ewrt = _mm256_mul_pd(r11,ewtabscale);
979 ewitab = _mm256_cvttpd_epi32(ewrt);
980 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
981 ewitab = _mm_slli_epi32(ewitab,2);
982 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
983 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
984 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
985 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
986 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
987 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
988 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
989 velec = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
990 felec = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
992 /* Update potential sum for this i atom from the interaction with this j atom. */
993 velec = _mm256_andnot_pd(dummy_mask,velec);
994 velecsum = _mm256_add_pd(velecsum,velec);
998 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1000 /* Calculate temporary vectorial force */
1001 tx = _mm256_mul_pd(fscal,dx11);
1002 ty = _mm256_mul_pd(fscal,dy11);
1003 tz = _mm256_mul_pd(fscal,dz11);
1005 /* Update vectorial force */
1006 fix1 = _mm256_add_pd(fix1,tx);
1007 fiy1 = _mm256_add_pd(fiy1,ty);
1008 fiz1 = _mm256_add_pd(fiz1,tz);
1010 fjx1 = _mm256_add_pd(fjx1,tx);
1011 fjy1 = _mm256_add_pd(fjy1,ty);
1012 fjz1 = _mm256_add_pd(fjz1,tz);
1014 /**************************
1015 * CALCULATE INTERACTIONS *
1016 **************************/
1018 r12 = _mm256_mul_pd(rsq12,rinv12);
1019 r12 = _mm256_andnot_pd(dummy_mask,r12);
1021 /* EWALD ELECTROSTATICS */
1023 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1024 ewrt = _mm256_mul_pd(r12,ewtabscale);
1025 ewitab = _mm256_cvttpd_epi32(ewrt);
1026 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1027 ewitab = _mm_slli_epi32(ewitab,2);
1028 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1029 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1030 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1031 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1032 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1033 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1034 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1035 velec = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
1036 felec = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1038 /* Update potential sum for this i atom from the interaction with this j atom. */
1039 velec = _mm256_andnot_pd(dummy_mask,velec);
1040 velecsum = _mm256_add_pd(velecsum,velec);
1044 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1046 /* Calculate temporary vectorial force */
1047 tx = _mm256_mul_pd(fscal,dx12);
1048 ty = _mm256_mul_pd(fscal,dy12);
1049 tz = _mm256_mul_pd(fscal,dz12);
1051 /* Update vectorial force */
1052 fix1 = _mm256_add_pd(fix1,tx);
1053 fiy1 = _mm256_add_pd(fiy1,ty);
1054 fiz1 = _mm256_add_pd(fiz1,tz);
1056 fjx2 = _mm256_add_pd(fjx2,tx);
1057 fjy2 = _mm256_add_pd(fjy2,ty);
1058 fjz2 = _mm256_add_pd(fjz2,tz);
1060 /**************************
1061 * CALCULATE INTERACTIONS *
1062 **************************/
1064 r20 = _mm256_mul_pd(rsq20,rinv20);
1065 r20 = _mm256_andnot_pd(dummy_mask,r20);
1067 /* EWALD ELECTROSTATICS */
1069 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1070 ewrt = _mm256_mul_pd(r20,ewtabscale);
1071 ewitab = _mm256_cvttpd_epi32(ewrt);
1072 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1073 ewitab = _mm_slli_epi32(ewitab,2);
1074 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1075 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1076 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1077 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1078 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1079 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1080 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1081 velec = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
1082 felec = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1084 /* Update potential sum for this i atom from the interaction with this j atom. */
1085 velec = _mm256_andnot_pd(dummy_mask,velec);
1086 velecsum = _mm256_add_pd(velecsum,velec);
1090 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1092 /* Calculate temporary vectorial force */
1093 tx = _mm256_mul_pd(fscal,dx20);
1094 ty = _mm256_mul_pd(fscal,dy20);
1095 tz = _mm256_mul_pd(fscal,dz20);
1097 /* Update vectorial force */
1098 fix2 = _mm256_add_pd(fix2,tx);
1099 fiy2 = _mm256_add_pd(fiy2,ty);
1100 fiz2 = _mm256_add_pd(fiz2,tz);
1102 fjx0 = _mm256_add_pd(fjx0,tx);
1103 fjy0 = _mm256_add_pd(fjy0,ty);
1104 fjz0 = _mm256_add_pd(fjz0,tz);
1106 /**************************
1107 * CALCULATE INTERACTIONS *
1108 **************************/
1110 r21 = _mm256_mul_pd(rsq21,rinv21);
1111 r21 = _mm256_andnot_pd(dummy_mask,r21);
1113 /* EWALD ELECTROSTATICS */
1115 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1116 ewrt = _mm256_mul_pd(r21,ewtabscale);
1117 ewitab = _mm256_cvttpd_epi32(ewrt);
1118 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1119 ewitab = _mm_slli_epi32(ewitab,2);
1120 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1121 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1122 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1123 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1124 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1125 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1126 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1127 velec = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
1128 felec = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1130 /* Update potential sum for this i atom from the interaction with this j atom. */
1131 velec = _mm256_andnot_pd(dummy_mask,velec);
1132 velecsum = _mm256_add_pd(velecsum,velec);
1136 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1138 /* Calculate temporary vectorial force */
1139 tx = _mm256_mul_pd(fscal,dx21);
1140 ty = _mm256_mul_pd(fscal,dy21);
1141 tz = _mm256_mul_pd(fscal,dz21);
1143 /* Update vectorial force */
1144 fix2 = _mm256_add_pd(fix2,tx);
1145 fiy2 = _mm256_add_pd(fiy2,ty);
1146 fiz2 = _mm256_add_pd(fiz2,tz);
1148 fjx1 = _mm256_add_pd(fjx1,tx);
1149 fjy1 = _mm256_add_pd(fjy1,ty);
1150 fjz1 = _mm256_add_pd(fjz1,tz);
1152 /**************************
1153 * CALCULATE INTERACTIONS *
1154 **************************/
1156 r22 = _mm256_mul_pd(rsq22,rinv22);
1157 r22 = _mm256_andnot_pd(dummy_mask,r22);
1159 /* EWALD ELECTROSTATICS */
1161 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1162 ewrt = _mm256_mul_pd(r22,ewtabscale);
1163 ewitab = _mm256_cvttpd_epi32(ewrt);
1164 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1165 ewitab = _mm_slli_epi32(ewitab,2);
1166 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1167 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1168 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1169 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1170 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1171 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1172 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1173 velec = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
1174 felec = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1176 /* Update potential sum for this i atom from the interaction with this j atom. */
1177 velec = _mm256_andnot_pd(dummy_mask,velec);
1178 velecsum = _mm256_add_pd(velecsum,velec);
1182 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1184 /* Calculate temporary vectorial force */
1185 tx = _mm256_mul_pd(fscal,dx22);
1186 ty = _mm256_mul_pd(fscal,dy22);
1187 tz = _mm256_mul_pd(fscal,dz22);
1189 /* Update vectorial force */
1190 fix2 = _mm256_add_pd(fix2,tx);
1191 fiy2 = _mm256_add_pd(fiy2,ty);
1192 fiz2 = _mm256_add_pd(fiz2,tz);
1194 fjx2 = _mm256_add_pd(fjx2,tx);
1195 fjy2 = _mm256_add_pd(fjy2,ty);
1196 fjz2 = _mm256_add_pd(fjz2,tz);
1198 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1199 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1200 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1201 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1203 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1204 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1206 /* Inner loop uses 378 flops */
1209 /* End of innermost loop */
1211 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1212 f+i_coord_offset,fshift+i_shift_offset);
1215 /* Update potential energies */
1216 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1218 /* Increment number of inner iterations */
1219 inneriter += j_index_end - j_index_start;
1221 /* Outer loop uses 19 flops */
1224 /* Increment number of outer iterations */
1227 /* Update outer/inner flops */
1229 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*378);
1232 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_256_double
1233 * Electrostatics interaction: Ewald
1234 * VdW interaction: None
1235 * Geometry: Water3-Water3
1236 * Calculate force/pot: Force
1239 nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_256_double
1240 (t_nblist * gmx_restrict nlist,
1241 rvec * gmx_restrict xx,
1242 rvec * gmx_restrict ff,
1243 t_forcerec * gmx_restrict fr,
1244 t_mdatoms * gmx_restrict mdatoms,
1245 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1246 t_nrnb * gmx_restrict nrnb)
1248 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1249 * just 0 for non-waters.
1250 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1251 * jnr indices corresponding to data put in the four positions in the SIMD register.
1253 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1254 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1255 int jnrA,jnrB,jnrC,jnrD;
1256 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1257 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1258 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1259 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1260 real rcutoff_scalar;
1261 real *shiftvec,*fshift,*x,*f;
1262 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1263 real scratch[4*DIM];
1264 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1265 real * vdwioffsetptr0;
1266 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1267 real * vdwioffsetptr1;
1268 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1269 real * vdwioffsetptr2;
1270 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1271 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1272 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1273 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1274 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1275 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1276 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1277 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1278 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1279 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1280 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1281 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1282 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1283 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1284 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1285 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1286 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
1289 __m256d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1290 __m256d beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1292 __m256d dummy_mask,cutoff_mask;
1293 __m128 tmpmask0,tmpmask1;
1294 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1295 __m256d one = _mm256_set1_pd(1.0);
1296 __m256d two = _mm256_set1_pd(2.0);
1302 jindex = nlist->jindex;
1304 shiftidx = nlist->shift;
1306 shiftvec = fr->shift_vec[0];
1307 fshift = fr->fshift[0];
1308 facel = _mm256_set1_pd(fr->epsfac);
1309 charge = mdatoms->chargeA;
1311 sh_ewald = _mm256_set1_pd(fr->ic->sh_ewald);
1312 beta = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
1313 beta2 = _mm256_mul_pd(beta,beta);
1314 beta3 = _mm256_mul_pd(beta,beta2);
1316 ewtab = fr->ic->tabq_coul_F;
1317 ewtabscale = _mm256_set1_pd(fr->ic->tabq_scale);
1318 ewtabhalfspace = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1320 /* Setup water-specific parameters */
1321 inr = nlist->iinr[0];
1322 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1323 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1324 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1326 jq0 = _mm256_set1_pd(charge[inr+0]);
1327 jq1 = _mm256_set1_pd(charge[inr+1]);
1328 jq2 = _mm256_set1_pd(charge[inr+2]);
1329 qq00 = _mm256_mul_pd(iq0,jq0);
1330 qq01 = _mm256_mul_pd(iq0,jq1);
1331 qq02 = _mm256_mul_pd(iq0,jq2);
1332 qq10 = _mm256_mul_pd(iq1,jq0);
1333 qq11 = _mm256_mul_pd(iq1,jq1);
1334 qq12 = _mm256_mul_pd(iq1,jq2);
1335 qq20 = _mm256_mul_pd(iq2,jq0);
1336 qq21 = _mm256_mul_pd(iq2,jq1);
1337 qq22 = _mm256_mul_pd(iq2,jq2);
1339 /* Avoid stupid compiler warnings */
1340 jnrA = jnrB = jnrC = jnrD = 0;
1341 j_coord_offsetA = 0;
1342 j_coord_offsetB = 0;
1343 j_coord_offsetC = 0;
1344 j_coord_offsetD = 0;
1349 for(iidx=0;iidx<4*DIM;iidx++)
1351 scratch[iidx] = 0.0;
1354 /* Start outer loop over neighborlists */
1355 for(iidx=0; iidx<nri; iidx++)
1357 /* Load shift vector for this list */
1358 i_shift_offset = DIM*shiftidx[iidx];
1360 /* Load limits for loop over neighbors */
1361 j_index_start = jindex[iidx];
1362 j_index_end = jindex[iidx+1];
1364 /* Get outer coordinate index */
1366 i_coord_offset = DIM*inr;
1368 /* Load i particle coords and add shift vector */
1369 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1370 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1372 fix0 = _mm256_setzero_pd();
1373 fiy0 = _mm256_setzero_pd();
1374 fiz0 = _mm256_setzero_pd();
1375 fix1 = _mm256_setzero_pd();
1376 fiy1 = _mm256_setzero_pd();
1377 fiz1 = _mm256_setzero_pd();
1378 fix2 = _mm256_setzero_pd();
1379 fiy2 = _mm256_setzero_pd();
1380 fiz2 = _mm256_setzero_pd();
1382 /* Start inner kernel loop */
1383 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1386 /* Get j neighbor index, and coordinate index */
1388 jnrB = jjnr[jidx+1];
1389 jnrC = jjnr[jidx+2];
1390 jnrD = jjnr[jidx+3];
1391 j_coord_offsetA = DIM*jnrA;
1392 j_coord_offsetB = DIM*jnrB;
1393 j_coord_offsetC = DIM*jnrC;
1394 j_coord_offsetD = DIM*jnrD;
1396 /* load j atom coordinates */
1397 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1398 x+j_coord_offsetC,x+j_coord_offsetD,
1399 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1401 /* Calculate displacement vector */
1402 dx00 = _mm256_sub_pd(ix0,jx0);
1403 dy00 = _mm256_sub_pd(iy0,jy0);
1404 dz00 = _mm256_sub_pd(iz0,jz0);
1405 dx01 = _mm256_sub_pd(ix0,jx1);
1406 dy01 = _mm256_sub_pd(iy0,jy1);
1407 dz01 = _mm256_sub_pd(iz0,jz1);
1408 dx02 = _mm256_sub_pd(ix0,jx2);
1409 dy02 = _mm256_sub_pd(iy0,jy2);
1410 dz02 = _mm256_sub_pd(iz0,jz2);
1411 dx10 = _mm256_sub_pd(ix1,jx0);
1412 dy10 = _mm256_sub_pd(iy1,jy0);
1413 dz10 = _mm256_sub_pd(iz1,jz0);
1414 dx11 = _mm256_sub_pd(ix1,jx1);
1415 dy11 = _mm256_sub_pd(iy1,jy1);
1416 dz11 = _mm256_sub_pd(iz1,jz1);
1417 dx12 = _mm256_sub_pd(ix1,jx2);
1418 dy12 = _mm256_sub_pd(iy1,jy2);
1419 dz12 = _mm256_sub_pd(iz1,jz2);
1420 dx20 = _mm256_sub_pd(ix2,jx0);
1421 dy20 = _mm256_sub_pd(iy2,jy0);
1422 dz20 = _mm256_sub_pd(iz2,jz0);
1423 dx21 = _mm256_sub_pd(ix2,jx1);
1424 dy21 = _mm256_sub_pd(iy2,jy1);
1425 dz21 = _mm256_sub_pd(iz2,jz1);
1426 dx22 = _mm256_sub_pd(ix2,jx2);
1427 dy22 = _mm256_sub_pd(iy2,jy2);
1428 dz22 = _mm256_sub_pd(iz2,jz2);
1430 /* Calculate squared distance and things based on it */
1431 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1432 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1433 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1434 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1435 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1436 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1437 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1438 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1439 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1441 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1442 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1443 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1444 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1445 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1446 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1447 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1448 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1449 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1451 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
1452 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
1453 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
1454 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1455 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
1456 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
1457 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1458 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
1459 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
1461 fjx0 = _mm256_setzero_pd();
1462 fjy0 = _mm256_setzero_pd();
1463 fjz0 = _mm256_setzero_pd();
1464 fjx1 = _mm256_setzero_pd();
1465 fjy1 = _mm256_setzero_pd();
1466 fjz1 = _mm256_setzero_pd();
1467 fjx2 = _mm256_setzero_pd();
1468 fjy2 = _mm256_setzero_pd();
1469 fjz2 = _mm256_setzero_pd();
1471 /**************************
1472 * CALCULATE INTERACTIONS *
1473 **************************/
1475 r00 = _mm256_mul_pd(rsq00,rinv00);
1477 /* EWALD ELECTROSTATICS */
1479 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1480 ewrt = _mm256_mul_pd(r00,ewtabscale);
1481 ewitab = _mm256_cvttpd_epi32(ewrt);
1482 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1483 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1484 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1486 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1487 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1491 /* Calculate temporary vectorial force */
1492 tx = _mm256_mul_pd(fscal,dx00);
1493 ty = _mm256_mul_pd(fscal,dy00);
1494 tz = _mm256_mul_pd(fscal,dz00);
1496 /* Update vectorial force */
1497 fix0 = _mm256_add_pd(fix0,tx);
1498 fiy0 = _mm256_add_pd(fiy0,ty);
1499 fiz0 = _mm256_add_pd(fiz0,tz);
1501 fjx0 = _mm256_add_pd(fjx0,tx);
1502 fjy0 = _mm256_add_pd(fjy0,ty);
1503 fjz0 = _mm256_add_pd(fjz0,tz);
1505 /**************************
1506 * CALCULATE INTERACTIONS *
1507 **************************/
1509 r01 = _mm256_mul_pd(rsq01,rinv01);
1511 /* EWALD ELECTROSTATICS */
1513 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1514 ewrt = _mm256_mul_pd(r01,ewtabscale);
1515 ewitab = _mm256_cvttpd_epi32(ewrt);
1516 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1517 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1518 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1520 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1521 felec = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
1525 /* Calculate temporary vectorial force */
1526 tx = _mm256_mul_pd(fscal,dx01);
1527 ty = _mm256_mul_pd(fscal,dy01);
1528 tz = _mm256_mul_pd(fscal,dz01);
1530 /* Update vectorial force */
1531 fix0 = _mm256_add_pd(fix0,tx);
1532 fiy0 = _mm256_add_pd(fiy0,ty);
1533 fiz0 = _mm256_add_pd(fiz0,tz);
1535 fjx1 = _mm256_add_pd(fjx1,tx);
1536 fjy1 = _mm256_add_pd(fjy1,ty);
1537 fjz1 = _mm256_add_pd(fjz1,tz);
1539 /**************************
1540 * CALCULATE INTERACTIONS *
1541 **************************/
1543 r02 = _mm256_mul_pd(rsq02,rinv02);
1545 /* EWALD ELECTROSTATICS */
1547 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1548 ewrt = _mm256_mul_pd(r02,ewtabscale);
1549 ewitab = _mm256_cvttpd_epi32(ewrt);
1550 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1551 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1552 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1554 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1555 felec = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
1559 /* Calculate temporary vectorial force */
1560 tx = _mm256_mul_pd(fscal,dx02);
1561 ty = _mm256_mul_pd(fscal,dy02);
1562 tz = _mm256_mul_pd(fscal,dz02);
1564 /* Update vectorial force */
1565 fix0 = _mm256_add_pd(fix0,tx);
1566 fiy0 = _mm256_add_pd(fiy0,ty);
1567 fiz0 = _mm256_add_pd(fiz0,tz);
1569 fjx2 = _mm256_add_pd(fjx2,tx);
1570 fjy2 = _mm256_add_pd(fjy2,ty);
1571 fjz2 = _mm256_add_pd(fjz2,tz);
1573 /**************************
1574 * CALCULATE INTERACTIONS *
1575 **************************/
1577 r10 = _mm256_mul_pd(rsq10,rinv10);
1579 /* EWALD ELECTROSTATICS */
1581 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1582 ewrt = _mm256_mul_pd(r10,ewtabscale);
1583 ewitab = _mm256_cvttpd_epi32(ewrt);
1584 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1585 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1586 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1588 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1589 felec = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1593 /* Calculate temporary vectorial force */
1594 tx = _mm256_mul_pd(fscal,dx10);
1595 ty = _mm256_mul_pd(fscal,dy10);
1596 tz = _mm256_mul_pd(fscal,dz10);
1598 /* Update vectorial force */
1599 fix1 = _mm256_add_pd(fix1,tx);
1600 fiy1 = _mm256_add_pd(fiy1,ty);
1601 fiz1 = _mm256_add_pd(fiz1,tz);
1603 fjx0 = _mm256_add_pd(fjx0,tx);
1604 fjy0 = _mm256_add_pd(fjy0,ty);
1605 fjz0 = _mm256_add_pd(fjz0,tz);
1607 /**************************
1608 * CALCULATE INTERACTIONS *
1609 **************************/
1611 r11 = _mm256_mul_pd(rsq11,rinv11);
1613 /* EWALD ELECTROSTATICS */
1615 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1616 ewrt = _mm256_mul_pd(r11,ewtabscale);
1617 ewitab = _mm256_cvttpd_epi32(ewrt);
1618 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1619 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1620 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1622 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1623 felec = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1627 /* Calculate temporary vectorial force */
1628 tx = _mm256_mul_pd(fscal,dx11);
1629 ty = _mm256_mul_pd(fscal,dy11);
1630 tz = _mm256_mul_pd(fscal,dz11);
1632 /* Update vectorial force */
1633 fix1 = _mm256_add_pd(fix1,tx);
1634 fiy1 = _mm256_add_pd(fiy1,ty);
1635 fiz1 = _mm256_add_pd(fiz1,tz);
1637 fjx1 = _mm256_add_pd(fjx1,tx);
1638 fjy1 = _mm256_add_pd(fjy1,ty);
1639 fjz1 = _mm256_add_pd(fjz1,tz);
1641 /**************************
1642 * CALCULATE INTERACTIONS *
1643 **************************/
1645 r12 = _mm256_mul_pd(rsq12,rinv12);
1647 /* EWALD ELECTROSTATICS */
1649 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1650 ewrt = _mm256_mul_pd(r12,ewtabscale);
1651 ewitab = _mm256_cvttpd_epi32(ewrt);
1652 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1653 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1654 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1656 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1657 felec = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1661 /* Calculate temporary vectorial force */
1662 tx = _mm256_mul_pd(fscal,dx12);
1663 ty = _mm256_mul_pd(fscal,dy12);
1664 tz = _mm256_mul_pd(fscal,dz12);
1666 /* Update vectorial force */
1667 fix1 = _mm256_add_pd(fix1,tx);
1668 fiy1 = _mm256_add_pd(fiy1,ty);
1669 fiz1 = _mm256_add_pd(fiz1,tz);
1671 fjx2 = _mm256_add_pd(fjx2,tx);
1672 fjy2 = _mm256_add_pd(fjy2,ty);
1673 fjz2 = _mm256_add_pd(fjz2,tz);
1675 /**************************
1676 * CALCULATE INTERACTIONS *
1677 **************************/
1679 r20 = _mm256_mul_pd(rsq20,rinv20);
1681 /* EWALD ELECTROSTATICS */
1683 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1684 ewrt = _mm256_mul_pd(r20,ewtabscale);
1685 ewitab = _mm256_cvttpd_epi32(ewrt);
1686 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1687 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1688 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1690 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1691 felec = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1695 /* Calculate temporary vectorial force */
1696 tx = _mm256_mul_pd(fscal,dx20);
1697 ty = _mm256_mul_pd(fscal,dy20);
1698 tz = _mm256_mul_pd(fscal,dz20);
1700 /* Update vectorial force */
1701 fix2 = _mm256_add_pd(fix2,tx);
1702 fiy2 = _mm256_add_pd(fiy2,ty);
1703 fiz2 = _mm256_add_pd(fiz2,tz);
1705 fjx0 = _mm256_add_pd(fjx0,tx);
1706 fjy0 = _mm256_add_pd(fjy0,ty);
1707 fjz0 = _mm256_add_pd(fjz0,tz);
1709 /**************************
1710 * CALCULATE INTERACTIONS *
1711 **************************/
1713 r21 = _mm256_mul_pd(rsq21,rinv21);
1715 /* EWALD ELECTROSTATICS */
1717 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1718 ewrt = _mm256_mul_pd(r21,ewtabscale);
1719 ewitab = _mm256_cvttpd_epi32(ewrt);
1720 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1721 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1722 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1724 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1725 felec = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1729 /* Calculate temporary vectorial force */
1730 tx = _mm256_mul_pd(fscal,dx21);
1731 ty = _mm256_mul_pd(fscal,dy21);
1732 tz = _mm256_mul_pd(fscal,dz21);
1734 /* Update vectorial force */
1735 fix2 = _mm256_add_pd(fix2,tx);
1736 fiy2 = _mm256_add_pd(fiy2,ty);
1737 fiz2 = _mm256_add_pd(fiz2,tz);
1739 fjx1 = _mm256_add_pd(fjx1,tx);
1740 fjy1 = _mm256_add_pd(fjy1,ty);
1741 fjz1 = _mm256_add_pd(fjz1,tz);
1743 /**************************
1744 * CALCULATE INTERACTIONS *
1745 **************************/
1747 r22 = _mm256_mul_pd(rsq22,rinv22);
1749 /* EWALD ELECTROSTATICS */
1751 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1752 ewrt = _mm256_mul_pd(r22,ewtabscale);
1753 ewitab = _mm256_cvttpd_epi32(ewrt);
1754 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1755 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1756 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1758 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1759 felec = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1763 /* Calculate temporary vectorial force */
1764 tx = _mm256_mul_pd(fscal,dx22);
1765 ty = _mm256_mul_pd(fscal,dy22);
1766 tz = _mm256_mul_pd(fscal,dz22);
1768 /* Update vectorial force */
1769 fix2 = _mm256_add_pd(fix2,tx);
1770 fiy2 = _mm256_add_pd(fiy2,ty);
1771 fiz2 = _mm256_add_pd(fiz2,tz);
1773 fjx2 = _mm256_add_pd(fjx2,tx);
1774 fjy2 = _mm256_add_pd(fjy2,ty);
1775 fjz2 = _mm256_add_pd(fjz2,tz);
1777 fjptrA = f+j_coord_offsetA;
1778 fjptrB = f+j_coord_offsetB;
1779 fjptrC = f+j_coord_offsetC;
1780 fjptrD = f+j_coord_offsetD;
1782 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1783 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1785 /* Inner loop uses 324 flops */
1788 if(jidx<j_index_end)
1791 /* Get j neighbor index, and coordinate index */
1792 jnrlistA = jjnr[jidx];
1793 jnrlistB = jjnr[jidx+1];
1794 jnrlistC = jjnr[jidx+2];
1795 jnrlistD = jjnr[jidx+3];
1796 /* Sign of each element will be negative for non-real atoms.
1797 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1798 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1800 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1802 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1803 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1804 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1806 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1807 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1808 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1809 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1810 j_coord_offsetA = DIM*jnrA;
1811 j_coord_offsetB = DIM*jnrB;
1812 j_coord_offsetC = DIM*jnrC;
1813 j_coord_offsetD = DIM*jnrD;
1815 /* load j atom coordinates */
1816 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1817 x+j_coord_offsetC,x+j_coord_offsetD,
1818 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1820 /* Calculate displacement vector */
1821 dx00 = _mm256_sub_pd(ix0,jx0);
1822 dy00 = _mm256_sub_pd(iy0,jy0);
1823 dz00 = _mm256_sub_pd(iz0,jz0);
1824 dx01 = _mm256_sub_pd(ix0,jx1);
1825 dy01 = _mm256_sub_pd(iy0,jy1);
1826 dz01 = _mm256_sub_pd(iz0,jz1);
1827 dx02 = _mm256_sub_pd(ix0,jx2);
1828 dy02 = _mm256_sub_pd(iy0,jy2);
1829 dz02 = _mm256_sub_pd(iz0,jz2);
1830 dx10 = _mm256_sub_pd(ix1,jx0);
1831 dy10 = _mm256_sub_pd(iy1,jy0);
1832 dz10 = _mm256_sub_pd(iz1,jz0);
1833 dx11 = _mm256_sub_pd(ix1,jx1);
1834 dy11 = _mm256_sub_pd(iy1,jy1);
1835 dz11 = _mm256_sub_pd(iz1,jz1);
1836 dx12 = _mm256_sub_pd(ix1,jx2);
1837 dy12 = _mm256_sub_pd(iy1,jy2);
1838 dz12 = _mm256_sub_pd(iz1,jz2);
1839 dx20 = _mm256_sub_pd(ix2,jx0);
1840 dy20 = _mm256_sub_pd(iy2,jy0);
1841 dz20 = _mm256_sub_pd(iz2,jz0);
1842 dx21 = _mm256_sub_pd(ix2,jx1);
1843 dy21 = _mm256_sub_pd(iy2,jy1);
1844 dz21 = _mm256_sub_pd(iz2,jz1);
1845 dx22 = _mm256_sub_pd(ix2,jx2);
1846 dy22 = _mm256_sub_pd(iy2,jy2);
1847 dz22 = _mm256_sub_pd(iz2,jz2);
1849 /* Calculate squared distance and things based on it */
1850 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1851 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1852 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1853 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1854 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1855 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1856 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1857 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1858 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1860 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1861 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1862 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1863 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1864 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1865 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1866 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1867 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1868 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1870 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
1871 rinvsq01 = _mm256_mul_pd(rinv01,rinv01);
1872 rinvsq02 = _mm256_mul_pd(rinv02,rinv02);
1873 rinvsq10 = _mm256_mul_pd(rinv10,rinv10);
1874 rinvsq11 = _mm256_mul_pd(rinv11,rinv11);
1875 rinvsq12 = _mm256_mul_pd(rinv12,rinv12);
1876 rinvsq20 = _mm256_mul_pd(rinv20,rinv20);
1877 rinvsq21 = _mm256_mul_pd(rinv21,rinv21);
1878 rinvsq22 = _mm256_mul_pd(rinv22,rinv22);
1880 fjx0 = _mm256_setzero_pd();
1881 fjy0 = _mm256_setzero_pd();
1882 fjz0 = _mm256_setzero_pd();
1883 fjx1 = _mm256_setzero_pd();
1884 fjy1 = _mm256_setzero_pd();
1885 fjz1 = _mm256_setzero_pd();
1886 fjx2 = _mm256_setzero_pd();
1887 fjy2 = _mm256_setzero_pd();
1888 fjz2 = _mm256_setzero_pd();
1890 /**************************
1891 * CALCULATE INTERACTIONS *
1892 **************************/
1894 r00 = _mm256_mul_pd(rsq00,rinv00);
1895 r00 = _mm256_andnot_pd(dummy_mask,r00);
1897 /* EWALD ELECTROSTATICS */
1899 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1900 ewrt = _mm256_mul_pd(r00,ewtabscale);
1901 ewitab = _mm256_cvttpd_epi32(ewrt);
1902 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1903 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1904 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1906 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1907 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1911 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1913 /* Calculate temporary vectorial force */
1914 tx = _mm256_mul_pd(fscal,dx00);
1915 ty = _mm256_mul_pd(fscal,dy00);
1916 tz = _mm256_mul_pd(fscal,dz00);
1918 /* Update vectorial force */
1919 fix0 = _mm256_add_pd(fix0,tx);
1920 fiy0 = _mm256_add_pd(fiy0,ty);
1921 fiz0 = _mm256_add_pd(fiz0,tz);
1923 fjx0 = _mm256_add_pd(fjx0,tx);
1924 fjy0 = _mm256_add_pd(fjy0,ty);
1925 fjz0 = _mm256_add_pd(fjz0,tz);
1927 /**************************
1928 * CALCULATE INTERACTIONS *
1929 **************************/
1931 r01 = _mm256_mul_pd(rsq01,rinv01);
1932 r01 = _mm256_andnot_pd(dummy_mask,r01);
1934 /* EWALD ELECTROSTATICS */
1936 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1937 ewrt = _mm256_mul_pd(r01,ewtabscale);
1938 ewitab = _mm256_cvttpd_epi32(ewrt);
1939 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1940 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1941 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1943 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1944 felec = _mm256_mul_pd(_mm256_mul_pd(qq01,rinv01),_mm256_sub_pd(rinvsq01,felec));
1948 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1950 /* Calculate temporary vectorial force */
1951 tx = _mm256_mul_pd(fscal,dx01);
1952 ty = _mm256_mul_pd(fscal,dy01);
1953 tz = _mm256_mul_pd(fscal,dz01);
1955 /* Update vectorial force */
1956 fix0 = _mm256_add_pd(fix0,tx);
1957 fiy0 = _mm256_add_pd(fiy0,ty);
1958 fiz0 = _mm256_add_pd(fiz0,tz);
1960 fjx1 = _mm256_add_pd(fjx1,tx);
1961 fjy1 = _mm256_add_pd(fjy1,ty);
1962 fjz1 = _mm256_add_pd(fjz1,tz);
1964 /**************************
1965 * CALCULATE INTERACTIONS *
1966 **************************/
1968 r02 = _mm256_mul_pd(rsq02,rinv02);
1969 r02 = _mm256_andnot_pd(dummy_mask,r02);
1971 /* EWALD ELECTROSTATICS */
1973 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1974 ewrt = _mm256_mul_pd(r02,ewtabscale);
1975 ewitab = _mm256_cvttpd_epi32(ewrt);
1976 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1977 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1978 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1980 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1981 felec = _mm256_mul_pd(_mm256_mul_pd(qq02,rinv02),_mm256_sub_pd(rinvsq02,felec));
1985 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1987 /* Calculate temporary vectorial force */
1988 tx = _mm256_mul_pd(fscal,dx02);
1989 ty = _mm256_mul_pd(fscal,dy02);
1990 tz = _mm256_mul_pd(fscal,dz02);
1992 /* Update vectorial force */
1993 fix0 = _mm256_add_pd(fix0,tx);
1994 fiy0 = _mm256_add_pd(fiy0,ty);
1995 fiz0 = _mm256_add_pd(fiz0,tz);
1997 fjx2 = _mm256_add_pd(fjx2,tx);
1998 fjy2 = _mm256_add_pd(fjy2,ty);
1999 fjz2 = _mm256_add_pd(fjz2,tz);
2001 /**************************
2002 * CALCULATE INTERACTIONS *
2003 **************************/
2005 r10 = _mm256_mul_pd(rsq10,rinv10);
2006 r10 = _mm256_andnot_pd(dummy_mask,r10);
2008 /* EWALD ELECTROSTATICS */
2010 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2011 ewrt = _mm256_mul_pd(r10,ewtabscale);
2012 ewitab = _mm256_cvttpd_epi32(ewrt);
2013 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2014 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2015 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2017 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2018 felec = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
2022 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2024 /* Calculate temporary vectorial force */
2025 tx = _mm256_mul_pd(fscal,dx10);
2026 ty = _mm256_mul_pd(fscal,dy10);
2027 tz = _mm256_mul_pd(fscal,dz10);
2029 /* Update vectorial force */
2030 fix1 = _mm256_add_pd(fix1,tx);
2031 fiy1 = _mm256_add_pd(fiy1,ty);
2032 fiz1 = _mm256_add_pd(fiz1,tz);
2034 fjx0 = _mm256_add_pd(fjx0,tx);
2035 fjy0 = _mm256_add_pd(fjy0,ty);
2036 fjz0 = _mm256_add_pd(fjz0,tz);
2038 /**************************
2039 * CALCULATE INTERACTIONS *
2040 **************************/
2042 r11 = _mm256_mul_pd(rsq11,rinv11);
2043 r11 = _mm256_andnot_pd(dummy_mask,r11);
2045 /* EWALD ELECTROSTATICS */
2047 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2048 ewrt = _mm256_mul_pd(r11,ewtabscale);
2049 ewitab = _mm256_cvttpd_epi32(ewrt);
2050 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2051 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2052 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2054 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2055 felec = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2059 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2061 /* Calculate temporary vectorial force */
2062 tx = _mm256_mul_pd(fscal,dx11);
2063 ty = _mm256_mul_pd(fscal,dy11);
2064 tz = _mm256_mul_pd(fscal,dz11);
2066 /* Update vectorial force */
2067 fix1 = _mm256_add_pd(fix1,tx);
2068 fiy1 = _mm256_add_pd(fiy1,ty);
2069 fiz1 = _mm256_add_pd(fiz1,tz);
2071 fjx1 = _mm256_add_pd(fjx1,tx);
2072 fjy1 = _mm256_add_pd(fjy1,ty);
2073 fjz1 = _mm256_add_pd(fjz1,tz);
2075 /**************************
2076 * CALCULATE INTERACTIONS *
2077 **************************/
2079 r12 = _mm256_mul_pd(rsq12,rinv12);
2080 r12 = _mm256_andnot_pd(dummy_mask,r12);
2082 /* EWALD ELECTROSTATICS */
2084 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2085 ewrt = _mm256_mul_pd(r12,ewtabscale);
2086 ewitab = _mm256_cvttpd_epi32(ewrt);
2087 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2088 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2089 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2091 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2092 felec = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2096 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2098 /* Calculate temporary vectorial force */
2099 tx = _mm256_mul_pd(fscal,dx12);
2100 ty = _mm256_mul_pd(fscal,dy12);
2101 tz = _mm256_mul_pd(fscal,dz12);
2103 /* Update vectorial force */
2104 fix1 = _mm256_add_pd(fix1,tx);
2105 fiy1 = _mm256_add_pd(fiy1,ty);
2106 fiz1 = _mm256_add_pd(fiz1,tz);
2108 fjx2 = _mm256_add_pd(fjx2,tx);
2109 fjy2 = _mm256_add_pd(fjy2,ty);
2110 fjz2 = _mm256_add_pd(fjz2,tz);
2112 /**************************
2113 * CALCULATE INTERACTIONS *
2114 **************************/
2116 r20 = _mm256_mul_pd(rsq20,rinv20);
2117 r20 = _mm256_andnot_pd(dummy_mask,r20);
2119 /* EWALD ELECTROSTATICS */
2121 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2122 ewrt = _mm256_mul_pd(r20,ewtabscale);
2123 ewitab = _mm256_cvttpd_epi32(ewrt);
2124 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2125 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2126 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2128 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2129 felec = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
2133 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2135 /* Calculate temporary vectorial force */
2136 tx = _mm256_mul_pd(fscal,dx20);
2137 ty = _mm256_mul_pd(fscal,dy20);
2138 tz = _mm256_mul_pd(fscal,dz20);
2140 /* Update vectorial force */
2141 fix2 = _mm256_add_pd(fix2,tx);
2142 fiy2 = _mm256_add_pd(fiy2,ty);
2143 fiz2 = _mm256_add_pd(fiz2,tz);
2145 fjx0 = _mm256_add_pd(fjx0,tx);
2146 fjy0 = _mm256_add_pd(fjy0,ty);
2147 fjz0 = _mm256_add_pd(fjz0,tz);
2149 /**************************
2150 * CALCULATE INTERACTIONS *
2151 **************************/
2153 r21 = _mm256_mul_pd(rsq21,rinv21);
2154 r21 = _mm256_andnot_pd(dummy_mask,r21);
2156 /* EWALD ELECTROSTATICS */
2158 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2159 ewrt = _mm256_mul_pd(r21,ewtabscale);
2160 ewitab = _mm256_cvttpd_epi32(ewrt);
2161 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2162 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2163 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2165 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2166 felec = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2170 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2172 /* Calculate temporary vectorial force */
2173 tx = _mm256_mul_pd(fscal,dx21);
2174 ty = _mm256_mul_pd(fscal,dy21);
2175 tz = _mm256_mul_pd(fscal,dz21);
2177 /* Update vectorial force */
2178 fix2 = _mm256_add_pd(fix2,tx);
2179 fiy2 = _mm256_add_pd(fiy2,ty);
2180 fiz2 = _mm256_add_pd(fiz2,tz);
2182 fjx1 = _mm256_add_pd(fjx1,tx);
2183 fjy1 = _mm256_add_pd(fjy1,ty);
2184 fjz1 = _mm256_add_pd(fjz1,tz);
2186 /**************************
2187 * CALCULATE INTERACTIONS *
2188 **************************/
2190 r22 = _mm256_mul_pd(rsq22,rinv22);
2191 r22 = _mm256_andnot_pd(dummy_mask,r22);
2193 /* EWALD ELECTROSTATICS */
2195 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2196 ewrt = _mm256_mul_pd(r22,ewtabscale);
2197 ewitab = _mm256_cvttpd_epi32(ewrt);
2198 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2199 gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2200 ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2202 felec = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2203 felec = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2207 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2209 /* Calculate temporary vectorial force */
2210 tx = _mm256_mul_pd(fscal,dx22);
2211 ty = _mm256_mul_pd(fscal,dy22);
2212 tz = _mm256_mul_pd(fscal,dz22);
2214 /* Update vectorial force */
2215 fix2 = _mm256_add_pd(fix2,tx);
2216 fiy2 = _mm256_add_pd(fiy2,ty);
2217 fiz2 = _mm256_add_pd(fiz2,tz);
2219 fjx2 = _mm256_add_pd(fjx2,tx);
2220 fjy2 = _mm256_add_pd(fjy2,ty);
2221 fjz2 = _mm256_add_pd(fjz2,tz);
2223 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2224 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2225 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2226 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2228 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2229 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2231 /* Inner loop uses 333 flops */
2234 /* End of innermost loop */
2236 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2237 f+i_coord_offset,fshift+i_shift_offset);
2239 /* Increment number of inner iterations */
2240 inneriter += j_index_end - j_index_start;
2242 /* Outer loop uses 18 flops */
2245 /* Increment number of outer iterations */
2248 /* Update outer/inner flops */
2250 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*333);