2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_256_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_256_double.h"
50 #include "kernelutil_x86_avx_256_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW3W3_VF_avx_256_double
54 * Electrostatics interaction: CubicSplineTable
55 * VdW interaction: None
56 * Geometry: Water3-Water3
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCSTab_VdwNone_GeomW3W3_VF_avx_256_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
79 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
85 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr0;
87 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 real * vdwioffsetptr1;
89 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 real * vdwioffsetptr2;
91 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
93 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
94 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
95 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
96 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
97 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
98 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
99 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
100 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
101 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
102 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
103 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
104 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
105 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
106 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
107 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
110 __m128i ifour = _mm_set1_epi32(4);
111 __m256d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
113 __m256d dummy_mask,cutoff_mask;
114 __m128 tmpmask0,tmpmask1;
115 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
116 __m256d one = _mm256_set1_pd(1.0);
117 __m256d two = _mm256_set1_pd(2.0);
123 jindex = nlist->jindex;
125 shiftidx = nlist->shift;
127 shiftvec = fr->shift_vec[0];
128 fshift = fr->fshift[0];
129 facel = _mm256_set1_pd(fr->epsfac);
130 charge = mdatoms->chargeA;
132 vftab = kernel_data->table_elec->data;
133 vftabscale = _mm256_set1_pd(kernel_data->table_elec->scale);
135 /* Setup water-specific parameters */
136 inr = nlist->iinr[0];
137 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
138 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
139 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
141 jq0 = _mm256_set1_pd(charge[inr+0]);
142 jq1 = _mm256_set1_pd(charge[inr+1]);
143 jq2 = _mm256_set1_pd(charge[inr+2]);
144 qq00 = _mm256_mul_pd(iq0,jq0);
145 qq01 = _mm256_mul_pd(iq0,jq1);
146 qq02 = _mm256_mul_pd(iq0,jq2);
147 qq10 = _mm256_mul_pd(iq1,jq0);
148 qq11 = _mm256_mul_pd(iq1,jq1);
149 qq12 = _mm256_mul_pd(iq1,jq2);
150 qq20 = _mm256_mul_pd(iq2,jq0);
151 qq21 = _mm256_mul_pd(iq2,jq1);
152 qq22 = _mm256_mul_pd(iq2,jq2);
154 /* Avoid stupid compiler warnings */
155 jnrA = jnrB = jnrC = jnrD = 0;
164 for(iidx=0;iidx<4*DIM;iidx++)
169 /* Start outer loop over neighborlists */
170 for(iidx=0; iidx<nri; iidx++)
172 /* Load shift vector for this list */
173 i_shift_offset = DIM*shiftidx[iidx];
175 /* Load limits for loop over neighbors */
176 j_index_start = jindex[iidx];
177 j_index_end = jindex[iidx+1];
179 /* Get outer coordinate index */
181 i_coord_offset = DIM*inr;
183 /* Load i particle coords and add shift vector */
184 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
185 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
187 fix0 = _mm256_setzero_pd();
188 fiy0 = _mm256_setzero_pd();
189 fiz0 = _mm256_setzero_pd();
190 fix1 = _mm256_setzero_pd();
191 fiy1 = _mm256_setzero_pd();
192 fiz1 = _mm256_setzero_pd();
193 fix2 = _mm256_setzero_pd();
194 fiy2 = _mm256_setzero_pd();
195 fiz2 = _mm256_setzero_pd();
197 /* Reset potential sums */
198 velecsum = _mm256_setzero_pd();
200 /* Start inner kernel loop */
201 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
204 /* Get j neighbor index, and coordinate index */
209 j_coord_offsetA = DIM*jnrA;
210 j_coord_offsetB = DIM*jnrB;
211 j_coord_offsetC = DIM*jnrC;
212 j_coord_offsetD = DIM*jnrD;
214 /* load j atom coordinates */
215 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
216 x+j_coord_offsetC,x+j_coord_offsetD,
217 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
219 /* Calculate displacement vector */
220 dx00 = _mm256_sub_pd(ix0,jx0);
221 dy00 = _mm256_sub_pd(iy0,jy0);
222 dz00 = _mm256_sub_pd(iz0,jz0);
223 dx01 = _mm256_sub_pd(ix0,jx1);
224 dy01 = _mm256_sub_pd(iy0,jy1);
225 dz01 = _mm256_sub_pd(iz0,jz1);
226 dx02 = _mm256_sub_pd(ix0,jx2);
227 dy02 = _mm256_sub_pd(iy0,jy2);
228 dz02 = _mm256_sub_pd(iz0,jz2);
229 dx10 = _mm256_sub_pd(ix1,jx0);
230 dy10 = _mm256_sub_pd(iy1,jy0);
231 dz10 = _mm256_sub_pd(iz1,jz0);
232 dx11 = _mm256_sub_pd(ix1,jx1);
233 dy11 = _mm256_sub_pd(iy1,jy1);
234 dz11 = _mm256_sub_pd(iz1,jz1);
235 dx12 = _mm256_sub_pd(ix1,jx2);
236 dy12 = _mm256_sub_pd(iy1,jy2);
237 dz12 = _mm256_sub_pd(iz1,jz2);
238 dx20 = _mm256_sub_pd(ix2,jx0);
239 dy20 = _mm256_sub_pd(iy2,jy0);
240 dz20 = _mm256_sub_pd(iz2,jz0);
241 dx21 = _mm256_sub_pd(ix2,jx1);
242 dy21 = _mm256_sub_pd(iy2,jy1);
243 dz21 = _mm256_sub_pd(iz2,jz1);
244 dx22 = _mm256_sub_pd(ix2,jx2);
245 dy22 = _mm256_sub_pd(iy2,jy2);
246 dz22 = _mm256_sub_pd(iz2,jz2);
248 /* Calculate squared distance and things based on it */
249 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
250 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
251 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
252 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
253 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
254 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
255 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
256 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
257 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
259 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
260 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
261 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
262 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
263 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
264 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
265 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
266 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
267 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
269 fjx0 = _mm256_setzero_pd();
270 fjy0 = _mm256_setzero_pd();
271 fjz0 = _mm256_setzero_pd();
272 fjx1 = _mm256_setzero_pd();
273 fjy1 = _mm256_setzero_pd();
274 fjz1 = _mm256_setzero_pd();
275 fjx2 = _mm256_setzero_pd();
276 fjy2 = _mm256_setzero_pd();
277 fjz2 = _mm256_setzero_pd();
279 /**************************
280 * CALCULATE INTERACTIONS *
281 **************************/
283 r00 = _mm256_mul_pd(rsq00,rinv00);
285 /* Calculate table index by multiplying r with table scale and truncate to integer */
286 rt = _mm256_mul_pd(r00,vftabscale);
287 vfitab = _mm256_cvttpd_epi32(rt);
288 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
289 vfitab = _mm_slli_epi32(vfitab,2);
291 /* CUBIC SPLINE TABLE ELECTROSTATICS */
292 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
293 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
294 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
295 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
296 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
297 Heps = _mm256_mul_pd(vfeps,H);
298 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
299 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
300 velec = _mm256_mul_pd(qq00,VV);
301 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
302 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
304 /* Update potential sum for this i atom from the interaction with this j atom. */
305 velecsum = _mm256_add_pd(velecsum,velec);
309 /* Calculate temporary vectorial force */
310 tx = _mm256_mul_pd(fscal,dx00);
311 ty = _mm256_mul_pd(fscal,dy00);
312 tz = _mm256_mul_pd(fscal,dz00);
314 /* Update vectorial force */
315 fix0 = _mm256_add_pd(fix0,tx);
316 fiy0 = _mm256_add_pd(fiy0,ty);
317 fiz0 = _mm256_add_pd(fiz0,tz);
319 fjx0 = _mm256_add_pd(fjx0,tx);
320 fjy0 = _mm256_add_pd(fjy0,ty);
321 fjz0 = _mm256_add_pd(fjz0,tz);
323 /**************************
324 * CALCULATE INTERACTIONS *
325 **************************/
327 r01 = _mm256_mul_pd(rsq01,rinv01);
329 /* Calculate table index by multiplying r with table scale and truncate to integer */
330 rt = _mm256_mul_pd(r01,vftabscale);
331 vfitab = _mm256_cvttpd_epi32(rt);
332 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
333 vfitab = _mm_slli_epi32(vfitab,2);
335 /* CUBIC SPLINE TABLE ELECTROSTATICS */
336 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
337 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
338 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
339 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
340 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
341 Heps = _mm256_mul_pd(vfeps,H);
342 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
343 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
344 velec = _mm256_mul_pd(qq01,VV);
345 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
346 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
348 /* Update potential sum for this i atom from the interaction with this j atom. */
349 velecsum = _mm256_add_pd(velecsum,velec);
353 /* Calculate temporary vectorial force */
354 tx = _mm256_mul_pd(fscal,dx01);
355 ty = _mm256_mul_pd(fscal,dy01);
356 tz = _mm256_mul_pd(fscal,dz01);
358 /* Update vectorial force */
359 fix0 = _mm256_add_pd(fix0,tx);
360 fiy0 = _mm256_add_pd(fiy0,ty);
361 fiz0 = _mm256_add_pd(fiz0,tz);
363 fjx1 = _mm256_add_pd(fjx1,tx);
364 fjy1 = _mm256_add_pd(fjy1,ty);
365 fjz1 = _mm256_add_pd(fjz1,tz);
367 /**************************
368 * CALCULATE INTERACTIONS *
369 **************************/
371 r02 = _mm256_mul_pd(rsq02,rinv02);
373 /* Calculate table index by multiplying r with table scale and truncate to integer */
374 rt = _mm256_mul_pd(r02,vftabscale);
375 vfitab = _mm256_cvttpd_epi32(rt);
376 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
377 vfitab = _mm_slli_epi32(vfitab,2);
379 /* CUBIC SPLINE TABLE ELECTROSTATICS */
380 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
381 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
382 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
383 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
384 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
385 Heps = _mm256_mul_pd(vfeps,H);
386 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
387 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
388 velec = _mm256_mul_pd(qq02,VV);
389 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
390 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
392 /* Update potential sum for this i atom from the interaction with this j atom. */
393 velecsum = _mm256_add_pd(velecsum,velec);
397 /* Calculate temporary vectorial force */
398 tx = _mm256_mul_pd(fscal,dx02);
399 ty = _mm256_mul_pd(fscal,dy02);
400 tz = _mm256_mul_pd(fscal,dz02);
402 /* Update vectorial force */
403 fix0 = _mm256_add_pd(fix0,tx);
404 fiy0 = _mm256_add_pd(fiy0,ty);
405 fiz0 = _mm256_add_pd(fiz0,tz);
407 fjx2 = _mm256_add_pd(fjx2,tx);
408 fjy2 = _mm256_add_pd(fjy2,ty);
409 fjz2 = _mm256_add_pd(fjz2,tz);
411 /**************************
412 * CALCULATE INTERACTIONS *
413 **************************/
415 r10 = _mm256_mul_pd(rsq10,rinv10);
417 /* Calculate table index by multiplying r with table scale and truncate to integer */
418 rt = _mm256_mul_pd(r10,vftabscale);
419 vfitab = _mm256_cvttpd_epi32(rt);
420 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
421 vfitab = _mm_slli_epi32(vfitab,2);
423 /* CUBIC SPLINE TABLE ELECTROSTATICS */
424 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
425 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
426 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
427 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
428 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
429 Heps = _mm256_mul_pd(vfeps,H);
430 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
431 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
432 velec = _mm256_mul_pd(qq10,VV);
433 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
434 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
436 /* Update potential sum for this i atom from the interaction with this j atom. */
437 velecsum = _mm256_add_pd(velecsum,velec);
441 /* Calculate temporary vectorial force */
442 tx = _mm256_mul_pd(fscal,dx10);
443 ty = _mm256_mul_pd(fscal,dy10);
444 tz = _mm256_mul_pd(fscal,dz10);
446 /* Update vectorial force */
447 fix1 = _mm256_add_pd(fix1,tx);
448 fiy1 = _mm256_add_pd(fiy1,ty);
449 fiz1 = _mm256_add_pd(fiz1,tz);
451 fjx0 = _mm256_add_pd(fjx0,tx);
452 fjy0 = _mm256_add_pd(fjy0,ty);
453 fjz0 = _mm256_add_pd(fjz0,tz);
455 /**************************
456 * CALCULATE INTERACTIONS *
457 **************************/
459 r11 = _mm256_mul_pd(rsq11,rinv11);
461 /* Calculate table index by multiplying r with table scale and truncate to integer */
462 rt = _mm256_mul_pd(r11,vftabscale);
463 vfitab = _mm256_cvttpd_epi32(rt);
464 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
465 vfitab = _mm_slli_epi32(vfitab,2);
467 /* CUBIC SPLINE TABLE ELECTROSTATICS */
468 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
469 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
470 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
471 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
472 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
473 Heps = _mm256_mul_pd(vfeps,H);
474 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
475 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
476 velec = _mm256_mul_pd(qq11,VV);
477 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
478 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
480 /* Update potential sum for this i atom from the interaction with this j atom. */
481 velecsum = _mm256_add_pd(velecsum,velec);
485 /* Calculate temporary vectorial force */
486 tx = _mm256_mul_pd(fscal,dx11);
487 ty = _mm256_mul_pd(fscal,dy11);
488 tz = _mm256_mul_pd(fscal,dz11);
490 /* Update vectorial force */
491 fix1 = _mm256_add_pd(fix1,tx);
492 fiy1 = _mm256_add_pd(fiy1,ty);
493 fiz1 = _mm256_add_pd(fiz1,tz);
495 fjx1 = _mm256_add_pd(fjx1,tx);
496 fjy1 = _mm256_add_pd(fjy1,ty);
497 fjz1 = _mm256_add_pd(fjz1,tz);
499 /**************************
500 * CALCULATE INTERACTIONS *
501 **************************/
503 r12 = _mm256_mul_pd(rsq12,rinv12);
505 /* Calculate table index by multiplying r with table scale and truncate to integer */
506 rt = _mm256_mul_pd(r12,vftabscale);
507 vfitab = _mm256_cvttpd_epi32(rt);
508 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
509 vfitab = _mm_slli_epi32(vfitab,2);
511 /* CUBIC SPLINE TABLE ELECTROSTATICS */
512 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
513 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
514 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
515 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
516 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
517 Heps = _mm256_mul_pd(vfeps,H);
518 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
519 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
520 velec = _mm256_mul_pd(qq12,VV);
521 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
522 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
524 /* Update potential sum for this i atom from the interaction with this j atom. */
525 velecsum = _mm256_add_pd(velecsum,velec);
529 /* Calculate temporary vectorial force */
530 tx = _mm256_mul_pd(fscal,dx12);
531 ty = _mm256_mul_pd(fscal,dy12);
532 tz = _mm256_mul_pd(fscal,dz12);
534 /* Update vectorial force */
535 fix1 = _mm256_add_pd(fix1,tx);
536 fiy1 = _mm256_add_pd(fiy1,ty);
537 fiz1 = _mm256_add_pd(fiz1,tz);
539 fjx2 = _mm256_add_pd(fjx2,tx);
540 fjy2 = _mm256_add_pd(fjy2,ty);
541 fjz2 = _mm256_add_pd(fjz2,tz);
543 /**************************
544 * CALCULATE INTERACTIONS *
545 **************************/
547 r20 = _mm256_mul_pd(rsq20,rinv20);
549 /* Calculate table index by multiplying r with table scale and truncate to integer */
550 rt = _mm256_mul_pd(r20,vftabscale);
551 vfitab = _mm256_cvttpd_epi32(rt);
552 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
553 vfitab = _mm_slli_epi32(vfitab,2);
555 /* CUBIC SPLINE TABLE ELECTROSTATICS */
556 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
557 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
558 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
559 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
560 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
561 Heps = _mm256_mul_pd(vfeps,H);
562 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
563 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
564 velec = _mm256_mul_pd(qq20,VV);
565 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
566 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
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 /* Calculate table index by multiplying r with table scale and truncate to integer */
594 rt = _mm256_mul_pd(r21,vftabscale);
595 vfitab = _mm256_cvttpd_epi32(rt);
596 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
597 vfitab = _mm_slli_epi32(vfitab,2);
599 /* CUBIC SPLINE TABLE ELECTROSTATICS */
600 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
601 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
602 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
603 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
604 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
605 Heps = _mm256_mul_pd(vfeps,H);
606 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
607 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
608 velec = _mm256_mul_pd(qq21,VV);
609 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
610 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
612 /* Update potential sum for this i atom from the interaction with this j atom. */
613 velecsum = _mm256_add_pd(velecsum,velec);
617 /* Calculate temporary vectorial force */
618 tx = _mm256_mul_pd(fscal,dx21);
619 ty = _mm256_mul_pd(fscal,dy21);
620 tz = _mm256_mul_pd(fscal,dz21);
622 /* Update vectorial force */
623 fix2 = _mm256_add_pd(fix2,tx);
624 fiy2 = _mm256_add_pd(fiy2,ty);
625 fiz2 = _mm256_add_pd(fiz2,tz);
627 fjx1 = _mm256_add_pd(fjx1,tx);
628 fjy1 = _mm256_add_pd(fjy1,ty);
629 fjz1 = _mm256_add_pd(fjz1,tz);
631 /**************************
632 * CALCULATE INTERACTIONS *
633 **************************/
635 r22 = _mm256_mul_pd(rsq22,rinv22);
637 /* Calculate table index by multiplying r with table scale and truncate to integer */
638 rt = _mm256_mul_pd(r22,vftabscale);
639 vfitab = _mm256_cvttpd_epi32(rt);
640 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
641 vfitab = _mm_slli_epi32(vfitab,2);
643 /* CUBIC SPLINE TABLE ELECTROSTATICS */
644 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
645 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
646 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
647 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
648 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
649 Heps = _mm256_mul_pd(vfeps,H);
650 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
651 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
652 velec = _mm256_mul_pd(qq22,VV);
653 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
654 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
656 /* Update potential sum for this i atom from the interaction with this j atom. */
657 velecsum = _mm256_add_pd(velecsum,velec);
661 /* Calculate temporary vectorial force */
662 tx = _mm256_mul_pd(fscal,dx22);
663 ty = _mm256_mul_pd(fscal,dy22);
664 tz = _mm256_mul_pd(fscal,dz22);
666 /* Update vectorial force */
667 fix2 = _mm256_add_pd(fix2,tx);
668 fiy2 = _mm256_add_pd(fiy2,ty);
669 fiz2 = _mm256_add_pd(fiz2,tz);
671 fjx2 = _mm256_add_pd(fjx2,tx);
672 fjy2 = _mm256_add_pd(fjy2,ty);
673 fjz2 = _mm256_add_pd(fjz2,tz);
675 fjptrA = f+j_coord_offsetA;
676 fjptrB = f+j_coord_offsetB;
677 fjptrC = f+j_coord_offsetC;
678 fjptrD = f+j_coord_offsetD;
680 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
681 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
683 /* Inner loop uses 387 flops */
689 /* Get j neighbor index, and coordinate index */
690 jnrlistA = jjnr[jidx];
691 jnrlistB = jjnr[jidx+1];
692 jnrlistC = jjnr[jidx+2];
693 jnrlistD = jjnr[jidx+3];
694 /* Sign of each element will be negative for non-real atoms.
695 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
696 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
698 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
700 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
701 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
702 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
704 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
705 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
706 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
707 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
708 j_coord_offsetA = DIM*jnrA;
709 j_coord_offsetB = DIM*jnrB;
710 j_coord_offsetC = DIM*jnrC;
711 j_coord_offsetD = DIM*jnrD;
713 /* load j atom coordinates */
714 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
715 x+j_coord_offsetC,x+j_coord_offsetD,
716 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
718 /* Calculate displacement vector */
719 dx00 = _mm256_sub_pd(ix0,jx0);
720 dy00 = _mm256_sub_pd(iy0,jy0);
721 dz00 = _mm256_sub_pd(iz0,jz0);
722 dx01 = _mm256_sub_pd(ix0,jx1);
723 dy01 = _mm256_sub_pd(iy0,jy1);
724 dz01 = _mm256_sub_pd(iz0,jz1);
725 dx02 = _mm256_sub_pd(ix0,jx2);
726 dy02 = _mm256_sub_pd(iy0,jy2);
727 dz02 = _mm256_sub_pd(iz0,jz2);
728 dx10 = _mm256_sub_pd(ix1,jx0);
729 dy10 = _mm256_sub_pd(iy1,jy0);
730 dz10 = _mm256_sub_pd(iz1,jz0);
731 dx11 = _mm256_sub_pd(ix1,jx1);
732 dy11 = _mm256_sub_pd(iy1,jy1);
733 dz11 = _mm256_sub_pd(iz1,jz1);
734 dx12 = _mm256_sub_pd(ix1,jx2);
735 dy12 = _mm256_sub_pd(iy1,jy2);
736 dz12 = _mm256_sub_pd(iz1,jz2);
737 dx20 = _mm256_sub_pd(ix2,jx0);
738 dy20 = _mm256_sub_pd(iy2,jy0);
739 dz20 = _mm256_sub_pd(iz2,jz0);
740 dx21 = _mm256_sub_pd(ix2,jx1);
741 dy21 = _mm256_sub_pd(iy2,jy1);
742 dz21 = _mm256_sub_pd(iz2,jz1);
743 dx22 = _mm256_sub_pd(ix2,jx2);
744 dy22 = _mm256_sub_pd(iy2,jy2);
745 dz22 = _mm256_sub_pd(iz2,jz2);
747 /* Calculate squared distance and things based on it */
748 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
749 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
750 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
751 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
752 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
753 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
754 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
755 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
756 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
758 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
759 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
760 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
761 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
762 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
763 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
764 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
765 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
766 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
768 fjx0 = _mm256_setzero_pd();
769 fjy0 = _mm256_setzero_pd();
770 fjz0 = _mm256_setzero_pd();
771 fjx1 = _mm256_setzero_pd();
772 fjy1 = _mm256_setzero_pd();
773 fjz1 = _mm256_setzero_pd();
774 fjx2 = _mm256_setzero_pd();
775 fjy2 = _mm256_setzero_pd();
776 fjz2 = _mm256_setzero_pd();
778 /**************************
779 * CALCULATE INTERACTIONS *
780 **************************/
782 r00 = _mm256_mul_pd(rsq00,rinv00);
783 r00 = _mm256_andnot_pd(dummy_mask,r00);
785 /* Calculate table index by multiplying r with table scale and truncate to integer */
786 rt = _mm256_mul_pd(r00,vftabscale);
787 vfitab = _mm256_cvttpd_epi32(rt);
788 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
789 vfitab = _mm_slli_epi32(vfitab,2);
791 /* CUBIC SPLINE TABLE ELECTROSTATICS */
792 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
793 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
794 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
795 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
796 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
797 Heps = _mm256_mul_pd(vfeps,H);
798 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
799 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
800 velec = _mm256_mul_pd(qq00,VV);
801 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
802 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
804 /* Update potential sum for this i atom from the interaction with this j atom. */
805 velec = _mm256_andnot_pd(dummy_mask,velec);
806 velecsum = _mm256_add_pd(velecsum,velec);
810 fscal = _mm256_andnot_pd(dummy_mask,fscal);
812 /* Calculate temporary vectorial force */
813 tx = _mm256_mul_pd(fscal,dx00);
814 ty = _mm256_mul_pd(fscal,dy00);
815 tz = _mm256_mul_pd(fscal,dz00);
817 /* Update vectorial force */
818 fix0 = _mm256_add_pd(fix0,tx);
819 fiy0 = _mm256_add_pd(fiy0,ty);
820 fiz0 = _mm256_add_pd(fiz0,tz);
822 fjx0 = _mm256_add_pd(fjx0,tx);
823 fjy0 = _mm256_add_pd(fjy0,ty);
824 fjz0 = _mm256_add_pd(fjz0,tz);
826 /**************************
827 * CALCULATE INTERACTIONS *
828 **************************/
830 r01 = _mm256_mul_pd(rsq01,rinv01);
831 r01 = _mm256_andnot_pd(dummy_mask,r01);
833 /* Calculate table index by multiplying r with table scale and truncate to integer */
834 rt = _mm256_mul_pd(r01,vftabscale);
835 vfitab = _mm256_cvttpd_epi32(rt);
836 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
837 vfitab = _mm_slli_epi32(vfitab,2);
839 /* CUBIC SPLINE TABLE ELECTROSTATICS */
840 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
841 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
842 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
843 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
844 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
845 Heps = _mm256_mul_pd(vfeps,H);
846 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
847 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
848 velec = _mm256_mul_pd(qq01,VV);
849 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
850 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
852 /* Update potential sum for this i atom from the interaction with this j atom. */
853 velec = _mm256_andnot_pd(dummy_mask,velec);
854 velecsum = _mm256_add_pd(velecsum,velec);
858 fscal = _mm256_andnot_pd(dummy_mask,fscal);
860 /* Calculate temporary vectorial force */
861 tx = _mm256_mul_pd(fscal,dx01);
862 ty = _mm256_mul_pd(fscal,dy01);
863 tz = _mm256_mul_pd(fscal,dz01);
865 /* Update vectorial force */
866 fix0 = _mm256_add_pd(fix0,tx);
867 fiy0 = _mm256_add_pd(fiy0,ty);
868 fiz0 = _mm256_add_pd(fiz0,tz);
870 fjx1 = _mm256_add_pd(fjx1,tx);
871 fjy1 = _mm256_add_pd(fjy1,ty);
872 fjz1 = _mm256_add_pd(fjz1,tz);
874 /**************************
875 * CALCULATE INTERACTIONS *
876 **************************/
878 r02 = _mm256_mul_pd(rsq02,rinv02);
879 r02 = _mm256_andnot_pd(dummy_mask,r02);
881 /* Calculate table index by multiplying r with table scale and truncate to integer */
882 rt = _mm256_mul_pd(r02,vftabscale);
883 vfitab = _mm256_cvttpd_epi32(rt);
884 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
885 vfitab = _mm_slli_epi32(vfitab,2);
887 /* CUBIC SPLINE TABLE ELECTROSTATICS */
888 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
889 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
890 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
891 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
892 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
893 Heps = _mm256_mul_pd(vfeps,H);
894 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
895 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
896 velec = _mm256_mul_pd(qq02,VV);
897 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
898 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
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 /* Calculate table index by multiplying r with table scale and truncate to integer */
930 rt = _mm256_mul_pd(r10,vftabscale);
931 vfitab = _mm256_cvttpd_epi32(rt);
932 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
933 vfitab = _mm_slli_epi32(vfitab,2);
935 /* CUBIC SPLINE TABLE ELECTROSTATICS */
936 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
937 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
938 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
939 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
940 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
941 Heps = _mm256_mul_pd(vfeps,H);
942 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
943 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
944 velec = _mm256_mul_pd(qq10,VV);
945 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
946 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
948 /* Update potential sum for this i atom from the interaction with this j atom. */
949 velec = _mm256_andnot_pd(dummy_mask,velec);
950 velecsum = _mm256_add_pd(velecsum,velec);
954 fscal = _mm256_andnot_pd(dummy_mask,fscal);
956 /* Calculate temporary vectorial force */
957 tx = _mm256_mul_pd(fscal,dx10);
958 ty = _mm256_mul_pd(fscal,dy10);
959 tz = _mm256_mul_pd(fscal,dz10);
961 /* Update vectorial force */
962 fix1 = _mm256_add_pd(fix1,tx);
963 fiy1 = _mm256_add_pd(fiy1,ty);
964 fiz1 = _mm256_add_pd(fiz1,tz);
966 fjx0 = _mm256_add_pd(fjx0,tx);
967 fjy0 = _mm256_add_pd(fjy0,ty);
968 fjz0 = _mm256_add_pd(fjz0,tz);
970 /**************************
971 * CALCULATE INTERACTIONS *
972 **************************/
974 r11 = _mm256_mul_pd(rsq11,rinv11);
975 r11 = _mm256_andnot_pd(dummy_mask,r11);
977 /* Calculate table index by multiplying r with table scale and truncate to integer */
978 rt = _mm256_mul_pd(r11,vftabscale);
979 vfitab = _mm256_cvttpd_epi32(rt);
980 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
981 vfitab = _mm_slli_epi32(vfitab,2);
983 /* CUBIC SPLINE TABLE ELECTROSTATICS */
984 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
985 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
986 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
987 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
988 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
989 Heps = _mm256_mul_pd(vfeps,H);
990 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
991 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
992 velec = _mm256_mul_pd(qq11,VV);
993 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
994 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
996 /* Update potential sum for this i atom from the interaction with this j atom. */
997 velec = _mm256_andnot_pd(dummy_mask,velec);
998 velecsum = _mm256_add_pd(velecsum,velec);
1002 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1004 /* Calculate temporary vectorial force */
1005 tx = _mm256_mul_pd(fscal,dx11);
1006 ty = _mm256_mul_pd(fscal,dy11);
1007 tz = _mm256_mul_pd(fscal,dz11);
1009 /* Update vectorial force */
1010 fix1 = _mm256_add_pd(fix1,tx);
1011 fiy1 = _mm256_add_pd(fiy1,ty);
1012 fiz1 = _mm256_add_pd(fiz1,tz);
1014 fjx1 = _mm256_add_pd(fjx1,tx);
1015 fjy1 = _mm256_add_pd(fjy1,ty);
1016 fjz1 = _mm256_add_pd(fjz1,tz);
1018 /**************************
1019 * CALCULATE INTERACTIONS *
1020 **************************/
1022 r12 = _mm256_mul_pd(rsq12,rinv12);
1023 r12 = _mm256_andnot_pd(dummy_mask,r12);
1025 /* Calculate table index by multiplying r with table scale and truncate to integer */
1026 rt = _mm256_mul_pd(r12,vftabscale);
1027 vfitab = _mm256_cvttpd_epi32(rt);
1028 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1029 vfitab = _mm_slli_epi32(vfitab,2);
1031 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1032 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1033 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1034 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1035 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1036 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1037 Heps = _mm256_mul_pd(vfeps,H);
1038 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1039 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1040 velec = _mm256_mul_pd(qq12,VV);
1041 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1042 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1044 /* Update potential sum for this i atom from the interaction with this j atom. */
1045 velec = _mm256_andnot_pd(dummy_mask,velec);
1046 velecsum = _mm256_add_pd(velecsum,velec);
1050 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1052 /* Calculate temporary vectorial force */
1053 tx = _mm256_mul_pd(fscal,dx12);
1054 ty = _mm256_mul_pd(fscal,dy12);
1055 tz = _mm256_mul_pd(fscal,dz12);
1057 /* Update vectorial force */
1058 fix1 = _mm256_add_pd(fix1,tx);
1059 fiy1 = _mm256_add_pd(fiy1,ty);
1060 fiz1 = _mm256_add_pd(fiz1,tz);
1062 fjx2 = _mm256_add_pd(fjx2,tx);
1063 fjy2 = _mm256_add_pd(fjy2,ty);
1064 fjz2 = _mm256_add_pd(fjz2,tz);
1066 /**************************
1067 * CALCULATE INTERACTIONS *
1068 **************************/
1070 r20 = _mm256_mul_pd(rsq20,rinv20);
1071 r20 = _mm256_andnot_pd(dummy_mask,r20);
1073 /* Calculate table index by multiplying r with table scale and truncate to integer */
1074 rt = _mm256_mul_pd(r20,vftabscale);
1075 vfitab = _mm256_cvttpd_epi32(rt);
1076 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1077 vfitab = _mm_slli_epi32(vfitab,2);
1079 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1080 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1081 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1082 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1083 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1084 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1085 Heps = _mm256_mul_pd(vfeps,H);
1086 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1087 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1088 velec = _mm256_mul_pd(qq20,VV);
1089 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1090 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1092 /* Update potential sum for this i atom from the interaction with this j atom. */
1093 velec = _mm256_andnot_pd(dummy_mask,velec);
1094 velecsum = _mm256_add_pd(velecsum,velec);
1098 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1100 /* Calculate temporary vectorial force */
1101 tx = _mm256_mul_pd(fscal,dx20);
1102 ty = _mm256_mul_pd(fscal,dy20);
1103 tz = _mm256_mul_pd(fscal,dz20);
1105 /* Update vectorial force */
1106 fix2 = _mm256_add_pd(fix2,tx);
1107 fiy2 = _mm256_add_pd(fiy2,ty);
1108 fiz2 = _mm256_add_pd(fiz2,tz);
1110 fjx0 = _mm256_add_pd(fjx0,tx);
1111 fjy0 = _mm256_add_pd(fjy0,ty);
1112 fjz0 = _mm256_add_pd(fjz0,tz);
1114 /**************************
1115 * CALCULATE INTERACTIONS *
1116 **************************/
1118 r21 = _mm256_mul_pd(rsq21,rinv21);
1119 r21 = _mm256_andnot_pd(dummy_mask,r21);
1121 /* Calculate table index by multiplying r with table scale and truncate to integer */
1122 rt = _mm256_mul_pd(r21,vftabscale);
1123 vfitab = _mm256_cvttpd_epi32(rt);
1124 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1125 vfitab = _mm_slli_epi32(vfitab,2);
1127 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1128 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1129 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1130 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1131 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1132 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1133 Heps = _mm256_mul_pd(vfeps,H);
1134 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1135 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1136 velec = _mm256_mul_pd(qq21,VV);
1137 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1138 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1140 /* Update potential sum for this i atom from the interaction with this j atom. */
1141 velec = _mm256_andnot_pd(dummy_mask,velec);
1142 velecsum = _mm256_add_pd(velecsum,velec);
1146 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1148 /* Calculate temporary vectorial force */
1149 tx = _mm256_mul_pd(fscal,dx21);
1150 ty = _mm256_mul_pd(fscal,dy21);
1151 tz = _mm256_mul_pd(fscal,dz21);
1153 /* Update vectorial force */
1154 fix2 = _mm256_add_pd(fix2,tx);
1155 fiy2 = _mm256_add_pd(fiy2,ty);
1156 fiz2 = _mm256_add_pd(fiz2,tz);
1158 fjx1 = _mm256_add_pd(fjx1,tx);
1159 fjy1 = _mm256_add_pd(fjy1,ty);
1160 fjz1 = _mm256_add_pd(fjz1,tz);
1162 /**************************
1163 * CALCULATE INTERACTIONS *
1164 **************************/
1166 r22 = _mm256_mul_pd(rsq22,rinv22);
1167 r22 = _mm256_andnot_pd(dummy_mask,r22);
1169 /* Calculate table index by multiplying r with table scale and truncate to integer */
1170 rt = _mm256_mul_pd(r22,vftabscale);
1171 vfitab = _mm256_cvttpd_epi32(rt);
1172 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1173 vfitab = _mm_slli_epi32(vfitab,2);
1175 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1176 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1177 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1178 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1179 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1180 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1181 Heps = _mm256_mul_pd(vfeps,H);
1182 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1183 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1184 velec = _mm256_mul_pd(qq22,VV);
1185 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1186 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1188 /* Update potential sum for this i atom from the interaction with this j atom. */
1189 velec = _mm256_andnot_pd(dummy_mask,velec);
1190 velecsum = _mm256_add_pd(velecsum,velec);
1194 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1196 /* Calculate temporary vectorial force */
1197 tx = _mm256_mul_pd(fscal,dx22);
1198 ty = _mm256_mul_pd(fscal,dy22);
1199 tz = _mm256_mul_pd(fscal,dz22);
1201 /* Update vectorial force */
1202 fix2 = _mm256_add_pd(fix2,tx);
1203 fiy2 = _mm256_add_pd(fiy2,ty);
1204 fiz2 = _mm256_add_pd(fiz2,tz);
1206 fjx2 = _mm256_add_pd(fjx2,tx);
1207 fjy2 = _mm256_add_pd(fjy2,ty);
1208 fjz2 = _mm256_add_pd(fjz2,tz);
1210 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1211 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1212 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1213 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1215 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1216 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1218 /* Inner loop uses 396 flops */
1221 /* End of innermost loop */
1223 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1224 f+i_coord_offset,fshift+i_shift_offset);
1227 /* Update potential energies */
1228 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1230 /* Increment number of inner iterations */
1231 inneriter += j_index_end - j_index_start;
1233 /* Outer loop uses 19 flops */
1236 /* Increment number of outer iterations */
1239 /* Update outer/inner flops */
1241 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*396);
1244 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW3W3_F_avx_256_double
1245 * Electrostatics interaction: CubicSplineTable
1246 * VdW interaction: None
1247 * Geometry: Water3-Water3
1248 * Calculate force/pot: Force
1251 nb_kernel_ElecCSTab_VdwNone_GeomW3W3_F_avx_256_double
1252 (t_nblist * gmx_restrict nlist,
1253 rvec * gmx_restrict xx,
1254 rvec * gmx_restrict ff,
1255 t_forcerec * gmx_restrict fr,
1256 t_mdatoms * gmx_restrict mdatoms,
1257 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1258 t_nrnb * gmx_restrict nrnb)
1260 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1261 * just 0 for non-waters.
1262 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1263 * jnr indices corresponding to data put in the four positions in the SIMD register.
1265 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1266 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1267 int jnrA,jnrB,jnrC,jnrD;
1268 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1269 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1270 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1271 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1272 real rcutoff_scalar;
1273 real *shiftvec,*fshift,*x,*f;
1274 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1275 real scratch[4*DIM];
1276 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1277 real * vdwioffsetptr0;
1278 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1279 real * vdwioffsetptr1;
1280 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1281 real * vdwioffsetptr2;
1282 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1283 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1284 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1285 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1286 __m256d jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1287 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1288 __m256d jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1289 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1290 __m256d dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1291 __m256d dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1292 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1293 __m256d dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1294 __m256d dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1295 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1296 __m256d dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1297 __m256d dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1298 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
1301 __m128i ifour = _mm_set1_epi32(4);
1302 __m256d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1304 __m256d dummy_mask,cutoff_mask;
1305 __m128 tmpmask0,tmpmask1;
1306 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1307 __m256d one = _mm256_set1_pd(1.0);
1308 __m256d two = _mm256_set1_pd(2.0);
1314 jindex = nlist->jindex;
1316 shiftidx = nlist->shift;
1318 shiftvec = fr->shift_vec[0];
1319 fshift = fr->fshift[0];
1320 facel = _mm256_set1_pd(fr->epsfac);
1321 charge = mdatoms->chargeA;
1323 vftab = kernel_data->table_elec->data;
1324 vftabscale = _mm256_set1_pd(kernel_data->table_elec->scale);
1326 /* Setup water-specific parameters */
1327 inr = nlist->iinr[0];
1328 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1329 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1330 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1332 jq0 = _mm256_set1_pd(charge[inr+0]);
1333 jq1 = _mm256_set1_pd(charge[inr+1]);
1334 jq2 = _mm256_set1_pd(charge[inr+2]);
1335 qq00 = _mm256_mul_pd(iq0,jq0);
1336 qq01 = _mm256_mul_pd(iq0,jq1);
1337 qq02 = _mm256_mul_pd(iq0,jq2);
1338 qq10 = _mm256_mul_pd(iq1,jq0);
1339 qq11 = _mm256_mul_pd(iq1,jq1);
1340 qq12 = _mm256_mul_pd(iq1,jq2);
1341 qq20 = _mm256_mul_pd(iq2,jq0);
1342 qq21 = _mm256_mul_pd(iq2,jq1);
1343 qq22 = _mm256_mul_pd(iq2,jq2);
1345 /* Avoid stupid compiler warnings */
1346 jnrA = jnrB = jnrC = jnrD = 0;
1347 j_coord_offsetA = 0;
1348 j_coord_offsetB = 0;
1349 j_coord_offsetC = 0;
1350 j_coord_offsetD = 0;
1355 for(iidx=0;iidx<4*DIM;iidx++)
1357 scratch[iidx] = 0.0;
1360 /* Start outer loop over neighborlists */
1361 for(iidx=0; iidx<nri; iidx++)
1363 /* Load shift vector for this list */
1364 i_shift_offset = DIM*shiftidx[iidx];
1366 /* Load limits for loop over neighbors */
1367 j_index_start = jindex[iidx];
1368 j_index_end = jindex[iidx+1];
1370 /* Get outer coordinate index */
1372 i_coord_offset = DIM*inr;
1374 /* Load i particle coords and add shift vector */
1375 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1376 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1378 fix0 = _mm256_setzero_pd();
1379 fiy0 = _mm256_setzero_pd();
1380 fiz0 = _mm256_setzero_pd();
1381 fix1 = _mm256_setzero_pd();
1382 fiy1 = _mm256_setzero_pd();
1383 fiz1 = _mm256_setzero_pd();
1384 fix2 = _mm256_setzero_pd();
1385 fiy2 = _mm256_setzero_pd();
1386 fiz2 = _mm256_setzero_pd();
1388 /* Start inner kernel loop */
1389 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1392 /* Get j neighbor index, and coordinate index */
1394 jnrB = jjnr[jidx+1];
1395 jnrC = jjnr[jidx+2];
1396 jnrD = jjnr[jidx+3];
1397 j_coord_offsetA = DIM*jnrA;
1398 j_coord_offsetB = DIM*jnrB;
1399 j_coord_offsetC = DIM*jnrC;
1400 j_coord_offsetD = DIM*jnrD;
1402 /* load j atom coordinates */
1403 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1404 x+j_coord_offsetC,x+j_coord_offsetD,
1405 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1407 /* Calculate displacement vector */
1408 dx00 = _mm256_sub_pd(ix0,jx0);
1409 dy00 = _mm256_sub_pd(iy0,jy0);
1410 dz00 = _mm256_sub_pd(iz0,jz0);
1411 dx01 = _mm256_sub_pd(ix0,jx1);
1412 dy01 = _mm256_sub_pd(iy0,jy1);
1413 dz01 = _mm256_sub_pd(iz0,jz1);
1414 dx02 = _mm256_sub_pd(ix0,jx2);
1415 dy02 = _mm256_sub_pd(iy0,jy2);
1416 dz02 = _mm256_sub_pd(iz0,jz2);
1417 dx10 = _mm256_sub_pd(ix1,jx0);
1418 dy10 = _mm256_sub_pd(iy1,jy0);
1419 dz10 = _mm256_sub_pd(iz1,jz0);
1420 dx11 = _mm256_sub_pd(ix1,jx1);
1421 dy11 = _mm256_sub_pd(iy1,jy1);
1422 dz11 = _mm256_sub_pd(iz1,jz1);
1423 dx12 = _mm256_sub_pd(ix1,jx2);
1424 dy12 = _mm256_sub_pd(iy1,jy2);
1425 dz12 = _mm256_sub_pd(iz1,jz2);
1426 dx20 = _mm256_sub_pd(ix2,jx0);
1427 dy20 = _mm256_sub_pd(iy2,jy0);
1428 dz20 = _mm256_sub_pd(iz2,jz0);
1429 dx21 = _mm256_sub_pd(ix2,jx1);
1430 dy21 = _mm256_sub_pd(iy2,jy1);
1431 dz21 = _mm256_sub_pd(iz2,jz1);
1432 dx22 = _mm256_sub_pd(ix2,jx2);
1433 dy22 = _mm256_sub_pd(iy2,jy2);
1434 dz22 = _mm256_sub_pd(iz2,jz2);
1436 /* Calculate squared distance and things based on it */
1437 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1438 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1439 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1440 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1441 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1442 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1443 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1444 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1445 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1447 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1448 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1449 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1450 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1451 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1452 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1453 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1454 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1455 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1457 fjx0 = _mm256_setzero_pd();
1458 fjy0 = _mm256_setzero_pd();
1459 fjz0 = _mm256_setzero_pd();
1460 fjx1 = _mm256_setzero_pd();
1461 fjy1 = _mm256_setzero_pd();
1462 fjz1 = _mm256_setzero_pd();
1463 fjx2 = _mm256_setzero_pd();
1464 fjy2 = _mm256_setzero_pd();
1465 fjz2 = _mm256_setzero_pd();
1467 /**************************
1468 * CALCULATE INTERACTIONS *
1469 **************************/
1471 r00 = _mm256_mul_pd(rsq00,rinv00);
1473 /* Calculate table index by multiplying r with table scale and truncate to integer */
1474 rt = _mm256_mul_pd(r00,vftabscale);
1475 vfitab = _mm256_cvttpd_epi32(rt);
1476 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1477 vfitab = _mm_slli_epi32(vfitab,2);
1479 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1480 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1481 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1482 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1483 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1484 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1485 Heps = _mm256_mul_pd(vfeps,H);
1486 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1487 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1488 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1492 /* Calculate temporary vectorial force */
1493 tx = _mm256_mul_pd(fscal,dx00);
1494 ty = _mm256_mul_pd(fscal,dy00);
1495 tz = _mm256_mul_pd(fscal,dz00);
1497 /* Update vectorial force */
1498 fix0 = _mm256_add_pd(fix0,tx);
1499 fiy0 = _mm256_add_pd(fiy0,ty);
1500 fiz0 = _mm256_add_pd(fiz0,tz);
1502 fjx0 = _mm256_add_pd(fjx0,tx);
1503 fjy0 = _mm256_add_pd(fjy0,ty);
1504 fjz0 = _mm256_add_pd(fjz0,tz);
1506 /**************************
1507 * CALCULATE INTERACTIONS *
1508 **************************/
1510 r01 = _mm256_mul_pd(rsq01,rinv01);
1512 /* Calculate table index by multiplying r with table scale and truncate to integer */
1513 rt = _mm256_mul_pd(r01,vftabscale);
1514 vfitab = _mm256_cvttpd_epi32(rt);
1515 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1516 vfitab = _mm_slli_epi32(vfitab,2);
1518 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1519 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1520 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1521 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1522 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1523 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1524 Heps = _mm256_mul_pd(vfeps,H);
1525 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1526 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1527 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
1531 /* Calculate temporary vectorial force */
1532 tx = _mm256_mul_pd(fscal,dx01);
1533 ty = _mm256_mul_pd(fscal,dy01);
1534 tz = _mm256_mul_pd(fscal,dz01);
1536 /* Update vectorial force */
1537 fix0 = _mm256_add_pd(fix0,tx);
1538 fiy0 = _mm256_add_pd(fiy0,ty);
1539 fiz0 = _mm256_add_pd(fiz0,tz);
1541 fjx1 = _mm256_add_pd(fjx1,tx);
1542 fjy1 = _mm256_add_pd(fjy1,ty);
1543 fjz1 = _mm256_add_pd(fjz1,tz);
1545 /**************************
1546 * CALCULATE INTERACTIONS *
1547 **************************/
1549 r02 = _mm256_mul_pd(rsq02,rinv02);
1551 /* Calculate table index by multiplying r with table scale and truncate to integer */
1552 rt = _mm256_mul_pd(r02,vftabscale);
1553 vfitab = _mm256_cvttpd_epi32(rt);
1554 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1555 vfitab = _mm_slli_epi32(vfitab,2);
1557 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1558 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1559 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1560 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1561 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1562 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1563 Heps = _mm256_mul_pd(vfeps,H);
1564 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1565 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1566 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
1570 /* Calculate temporary vectorial force */
1571 tx = _mm256_mul_pd(fscal,dx02);
1572 ty = _mm256_mul_pd(fscal,dy02);
1573 tz = _mm256_mul_pd(fscal,dz02);
1575 /* Update vectorial force */
1576 fix0 = _mm256_add_pd(fix0,tx);
1577 fiy0 = _mm256_add_pd(fiy0,ty);
1578 fiz0 = _mm256_add_pd(fiz0,tz);
1580 fjx2 = _mm256_add_pd(fjx2,tx);
1581 fjy2 = _mm256_add_pd(fjy2,ty);
1582 fjz2 = _mm256_add_pd(fjz2,tz);
1584 /**************************
1585 * CALCULATE INTERACTIONS *
1586 **************************/
1588 r10 = _mm256_mul_pd(rsq10,rinv10);
1590 /* Calculate table index by multiplying r with table scale and truncate to integer */
1591 rt = _mm256_mul_pd(r10,vftabscale);
1592 vfitab = _mm256_cvttpd_epi32(rt);
1593 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1594 vfitab = _mm_slli_epi32(vfitab,2);
1596 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1597 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1598 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1599 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1600 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1601 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1602 Heps = _mm256_mul_pd(vfeps,H);
1603 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1604 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1605 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1609 /* Calculate temporary vectorial force */
1610 tx = _mm256_mul_pd(fscal,dx10);
1611 ty = _mm256_mul_pd(fscal,dy10);
1612 tz = _mm256_mul_pd(fscal,dz10);
1614 /* Update vectorial force */
1615 fix1 = _mm256_add_pd(fix1,tx);
1616 fiy1 = _mm256_add_pd(fiy1,ty);
1617 fiz1 = _mm256_add_pd(fiz1,tz);
1619 fjx0 = _mm256_add_pd(fjx0,tx);
1620 fjy0 = _mm256_add_pd(fjy0,ty);
1621 fjz0 = _mm256_add_pd(fjz0,tz);
1623 /**************************
1624 * CALCULATE INTERACTIONS *
1625 **************************/
1627 r11 = _mm256_mul_pd(rsq11,rinv11);
1629 /* Calculate table index by multiplying r with table scale and truncate to integer */
1630 rt = _mm256_mul_pd(r11,vftabscale);
1631 vfitab = _mm256_cvttpd_epi32(rt);
1632 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1633 vfitab = _mm_slli_epi32(vfitab,2);
1635 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1636 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1637 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1638 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1639 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1640 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1641 Heps = _mm256_mul_pd(vfeps,H);
1642 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1643 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1644 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1648 /* Calculate temporary vectorial force */
1649 tx = _mm256_mul_pd(fscal,dx11);
1650 ty = _mm256_mul_pd(fscal,dy11);
1651 tz = _mm256_mul_pd(fscal,dz11);
1653 /* Update vectorial force */
1654 fix1 = _mm256_add_pd(fix1,tx);
1655 fiy1 = _mm256_add_pd(fiy1,ty);
1656 fiz1 = _mm256_add_pd(fiz1,tz);
1658 fjx1 = _mm256_add_pd(fjx1,tx);
1659 fjy1 = _mm256_add_pd(fjy1,ty);
1660 fjz1 = _mm256_add_pd(fjz1,tz);
1662 /**************************
1663 * CALCULATE INTERACTIONS *
1664 **************************/
1666 r12 = _mm256_mul_pd(rsq12,rinv12);
1668 /* Calculate table index by multiplying r with table scale and truncate to integer */
1669 rt = _mm256_mul_pd(r12,vftabscale);
1670 vfitab = _mm256_cvttpd_epi32(rt);
1671 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1672 vfitab = _mm_slli_epi32(vfitab,2);
1674 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1675 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1676 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1677 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1678 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1679 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1680 Heps = _mm256_mul_pd(vfeps,H);
1681 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1682 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1683 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1687 /* Calculate temporary vectorial force */
1688 tx = _mm256_mul_pd(fscal,dx12);
1689 ty = _mm256_mul_pd(fscal,dy12);
1690 tz = _mm256_mul_pd(fscal,dz12);
1692 /* Update vectorial force */
1693 fix1 = _mm256_add_pd(fix1,tx);
1694 fiy1 = _mm256_add_pd(fiy1,ty);
1695 fiz1 = _mm256_add_pd(fiz1,tz);
1697 fjx2 = _mm256_add_pd(fjx2,tx);
1698 fjy2 = _mm256_add_pd(fjy2,ty);
1699 fjz2 = _mm256_add_pd(fjz2,tz);
1701 /**************************
1702 * CALCULATE INTERACTIONS *
1703 **************************/
1705 r20 = _mm256_mul_pd(rsq20,rinv20);
1707 /* Calculate table index by multiplying r with table scale and truncate to integer */
1708 rt = _mm256_mul_pd(r20,vftabscale);
1709 vfitab = _mm256_cvttpd_epi32(rt);
1710 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1711 vfitab = _mm_slli_epi32(vfitab,2);
1713 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1714 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1715 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1716 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1717 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1718 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1719 Heps = _mm256_mul_pd(vfeps,H);
1720 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1721 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1722 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1726 /* Calculate temporary vectorial force */
1727 tx = _mm256_mul_pd(fscal,dx20);
1728 ty = _mm256_mul_pd(fscal,dy20);
1729 tz = _mm256_mul_pd(fscal,dz20);
1731 /* Update vectorial force */
1732 fix2 = _mm256_add_pd(fix2,tx);
1733 fiy2 = _mm256_add_pd(fiy2,ty);
1734 fiz2 = _mm256_add_pd(fiz2,tz);
1736 fjx0 = _mm256_add_pd(fjx0,tx);
1737 fjy0 = _mm256_add_pd(fjy0,ty);
1738 fjz0 = _mm256_add_pd(fjz0,tz);
1740 /**************************
1741 * CALCULATE INTERACTIONS *
1742 **************************/
1744 r21 = _mm256_mul_pd(rsq21,rinv21);
1746 /* Calculate table index by multiplying r with table scale and truncate to integer */
1747 rt = _mm256_mul_pd(r21,vftabscale);
1748 vfitab = _mm256_cvttpd_epi32(rt);
1749 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1750 vfitab = _mm_slli_epi32(vfitab,2);
1752 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1753 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1754 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1755 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1756 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1757 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1758 Heps = _mm256_mul_pd(vfeps,H);
1759 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1760 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1761 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1765 /* Calculate temporary vectorial force */
1766 tx = _mm256_mul_pd(fscal,dx21);
1767 ty = _mm256_mul_pd(fscal,dy21);
1768 tz = _mm256_mul_pd(fscal,dz21);
1770 /* Update vectorial force */
1771 fix2 = _mm256_add_pd(fix2,tx);
1772 fiy2 = _mm256_add_pd(fiy2,ty);
1773 fiz2 = _mm256_add_pd(fiz2,tz);
1775 fjx1 = _mm256_add_pd(fjx1,tx);
1776 fjy1 = _mm256_add_pd(fjy1,ty);
1777 fjz1 = _mm256_add_pd(fjz1,tz);
1779 /**************************
1780 * CALCULATE INTERACTIONS *
1781 **************************/
1783 r22 = _mm256_mul_pd(rsq22,rinv22);
1785 /* Calculate table index by multiplying r with table scale and truncate to integer */
1786 rt = _mm256_mul_pd(r22,vftabscale);
1787 vfitab = _mm256_cvttpd_epi32(rt);
1788 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1789 vfitab = _mm_slli_epi32(vfitab,2);
1791 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1792 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1793 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1794 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1795 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1796 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1797 Heps = _mm256_mul_pd(vfeps,H);
1798 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1799 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1800 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1804 /* Calculate temporary vectorial force */
1805 tx = _mm256_mul_pd(fscal,dx22);
1806 ty = _mm256_mul_pd(fscal,dy22);
1807 tz = _mm256_mul_pd(fscal,dz22);
1809 /* Update vectorial force */
1810 fix2 = _mm256_add_pd(fix2,tx);
1811 fiy2 = _mm256_add_pd(fiy2,ty);
1812 fiz2 = _mm256_add_pd(fiz2,tz);
1814 fjx2 = _mm256_add_pd(fjx2,tx);
1815 fjy2 = _mm256_add_pd(fjy2,ty);
1816 fjz2 = _mm256_add_pd(fjz2,tz);
1818 fjptrA = f+j_coord_offsetA;
1819 fjptrB = f+j_coord_offsetB;
1820 fjptrC = f+j_coord_offsetC;
1821 fjptrD = f+j_coord_offsetD;
1823 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1824 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1826 /* Inner loop uses 351 flops */
1829 if(jidx<j_index_end)
1832 /* Get j neighbor index, and coordinate index */
1833 jnrlistA = jjnr[jidx];
1834 jnrlistB = jjnr[jidx+1];
1835 jnrlistC = jjnr[jidx+2];
1836 jnrlistD = jjnr[jidx+3];
1837 /* Sign of each element will be negative for non-real atoms.
1838 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1839 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1841 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1843 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1844 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1845 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1847 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1848 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1849 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1850 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1851 j_coord_offsetA = DIM*jnrA;
1852 j_coord_offsetB = DIM*jnrB;
1853 j_coord_offsetC = DIM*jnrC;
1854 j_coord_offsetD = DIM*jnrD;
1856 /* load j atom coordinates */
1857 gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1858 x+j_coord_offsetC,x+j_coord_offsetD,
1859 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1861 /* Calculate displacement vector */
1862 dx00 = _mm256_sub_pd(ix0,jx0);
1863 dy00 = _mm256_sub_pd(iy0,jy0);
1864 dz00 = _mm256_sub_pd(iz0,jz0);
1865 dx01 = _mm256_sub_pd(ix0,jx1);
1866 dy01 = _mm256_sub_pd(iy0,jy1);
1867 dz01 = _mm256_sub_pd(iz0,jz1);
1868 dx02 = _mm256_sub_pd(ix0,jx2);
1869 dy02 = _mm256_sub_pd(iy0,jy2);
1870 dz02 = _mm256_sub_pd(iz0,jz2);
1871 dx10 = _mm256_sub_pd(ix1,jx0);
1872 dy10 = _mm256_sub_pd(iy1,jy0);
1873 dz10 = _mm256_sub_pd(iz1,jz0);
1874 dx11 = _mm256_sub_pd(ix1,jx1);
1875 dy11 = _mm256_sub_pd(iy1,jy1);
1876 dz11 = _mm256_sub_pd(iz1,jz1);
1877 dx12 = _mm256_sub_pd(ix1,jx2);
1878 dy12 = _mm256_sub_pd(iy1,jy2);
1879 dz12 = _mm256_sub_pd(iz1,jz2);
1880 dx20 = _mm256_sub_pd(ix2,jx0);
1881 dy20 = _mm256_sub_pd(iy2,jy0);
1882 dz20 = _mm256_sub_pd(iz2,jz0);
1883 dx21 = _mm256_sub_pd(ix2,jx1);
1884 dy21 = _mm256_sub_pd(iy2,jy1);
1885 dz21 = _mm256_sub_pd(iz2,jz1);
1886 dx22 = _mm256_sub_pd(ix2,jx2);
1887 dy22 = _mm256_sub_pd(iy2,jy2);
1888 dz22 = _mm256_sub_pd(iz2,jz2);
1890 /* Calculate squared distance and things based on it */
1891 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1892 rsq01 = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1893 rsq02 = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1894 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1895 rsq11 = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1896 rsq12 = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1897 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1898 rsq21 = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1899 rsq22 = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1901 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
1902 rinv01 = gmx_mm256_invsqrt_pd(rsq01);
1903 rinv02 = gmx_mm256_invsqrt_pd(rsq02);
1904 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
1905 rinv11 = gmx_mm256_invsqrt_pd(rsq11);
1906 rinv12 = gmx_mm256_invsqrt_pd(rsq12);
1907 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
1908 rinv21 = gmx_mm256_invsqrt_pd(rsq21);
1909 rinv22 = gmx_mm256_invsqrt_pd(rsq22);
1911 fjx0 = _mm256_setzero_pd();
1912 fjy0 = _mm256_setzero_pd();
1913 fjz0 = _mm256_setzero_pd();
1914 fjx1 = _mm256_setzero_pd();
1915 fjy1 = _mm256_setzero_pd();
1916 fjz1 = _mm256_setzero_pd();
1917 fjx2 = _mm256_setzero_pd();
1918 fjy2 = _mm256_setzero_pd();
1919 fjz2 = _mm256_setzero_pd();
1921 /**************************
1922 * CALCULATE INTERACTIONS *
1923 **************************/
1925 r00 = _mm256_mul_pd(rsq00,rinv00);
1926 r00 = _mm256_andnot_pd(dummy_mask,r00);
1928 /* Calculate table index by multiplying r with table scale and truncate to integer */
1929 rt = _mm256_mul_pd(r00,vftabscale);
1930 vfitab = _mm256_cvttpd_epi32(rt);
1931 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1932 vfitab = _mm_slli_epi32(vfitab,2);
1934 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1935 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1936 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1937 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1938 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1939 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1940 Heps = _mm256_mul_pd(vfeps,H);
1941 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1942 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1943 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1947 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1949 /* Calculate temporary vectorial force */
1950 tx = _mm256_mul_pd(fscal,dx00);
1951 ty = _mm256_mul_pd(fscal,dy00);
1952 tz = _mm256_mul_pd(fscal,dz00);
1954 /* Update vectorial force */
1955 fix0 = _mm256_add_pd(fix0,tx);
1956 fiy0 = _mm256_add_pd(fiy0,ty);
1957 fiz0 = _mm256_add_pd(fiz0,tz);
1959 fjx0 = _mm256_add_pd(fjx0,tx);
1960 fjy0 = _mm256_add_pd(fjy0,ty);
1961 fjz0 = _mm256_add_pd(fjz0,tz);
1963 /**************************
1964 * CALCULATE INTERACTIONS *
1965 **************************/
1967 r01 = _mm256_mul_pd(rsq01,rinv01);
1968 r01 = _mm256_andnot_pd(dummy_mask,r01);
1970 /* Calculate table index by multiplying r with table scale and truncate to integer */
1971 rt = _mm256_mul_pd(r01,vftabscale);
1972 vfitab = _mm256_cvttpd_epi32(rt);
1973 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1974 vfitab = _mm_slli_epi32(vfitab,2);
1976 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1977 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1978 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1979 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1980 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1981 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1982 Heps = _mm256_mul_pd(vfeps,H);
1983 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1984 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1985 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
1989 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1991 /* Calculate temporary vectorial force */
1992 tx = _mm256_mul_pd(fscal,dx01);
1993 ty = _mm256_mul_pd(fscal,dy01);
1994 tz = _mm256_mul_pd(fscal,dz01);
1996 /* Update vectorial force */
1997 fix0 = _mm256_add_pd(fix0,tx);
1998 fiy0 = _mm256_add_pd(fiy0,ty);
1999 fiz0 = _mm256_add_pd(fiz0,tz);
2001 fjx1 = _mm256_add_pd(fjx1,tx);
2002 fjy1 = _mm256_add_pd(fjy1,ty);
2003 fjz1 = _mm256_add_pd(fjz1,tz);
2005 /**************************
2006 * CALCULATE INTERACTIONS *
2007 **************************/
2009 r02 = _mm256_mul_pd(rsq02,rinv02);
2010 r02 = _mm256_andnot_pd(dummy_mask,r02);
2012 /* Calculate table index by multiplying r with table scale and truncate to integer */
2013 rt = _mm256_mul_pd(r02,vftabscale);
2014 vfitab = _mm256_cvttpd_epi32(rt);
2015 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2016 vfitab = _mm_slli_epi32(vfitab,2);
2018 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2019 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2020 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2021 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2022 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2023 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2024 Heps = _mm256_mul_pd(vfeps,H);
2025 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2026 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2027 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
2031 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2033 /* Calculate temporary vectorial force */
2034 tx = _mm256_mul_pd(fscal,dx02);
2035 ty = _mm256_mul_pd(fscal,dy02);
2036 tz = _mm256_mul_pd(fscal,dz02);
2038 /* Update vectorial force */
2039 fix0 = _mm256_add_pd(fix0,tx);
2040 fiy0 = _mm256_add_pd(fiy0,ty);
2041 fiz0 = _mm256_add_pd(fiz0,tz);
2043 fjx2 = _mm256_add_pd(fjx2,tx);
2044 fjy2 = _mm256_add_pd(fjy2,ty);
2045 fjz2 = _mm256_add_pd(fjz2,tz);
2047 /**************************
2048 * CALCULATE INTERACTIONS *
2049 **************************/
2051 r10 = _mm256_mul_pd(rsq10,rinv10);
2052 r10 = _mm256_andnot_pd(dummy_mask,r10);
2054 /* Calculate table index by multiplying r with table scale and truncate to integer */
2055 rt = _mm256_mul_pd(r10,vftabscale);
2056 vfitab = _mm256_cvttpd_epi32(rt);
2057 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2058 vfitab = _mm_slli_epi32(vfitab,2);
2060 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2061 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2062 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2063 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2064 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2065 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2066 Heps = _mm256_mul_pd(vfeps,H);
2067 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2068 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2069 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
2073 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2075 /* Calculate temporary vectorial force */
2076 tx = _mm256_mul_pd(fscal,dx10);
2077 ty = _mm256_mul_pd(fscal,dy10);
2078 tz = _mm256_mul_pd(fscal,dz10);
2080 /* Update vectorial force */
2081 fix1 = _mm256_add_pd(fix1,tx);
2082 fiy1 = _mm256_add_pd(fiy1,ty);
2083 fiz1 = _mm256_add_pd(fiz1,tz);
2085 fjx0 = _mm256_add_pd(fjx0,tx);
2086 fjy0 = _mm256_add_pd(fjy0,ty);
2087 fjz0 = _mm256_add_pd(fjz0,tz);
2089 /**************************
2090 * CALCULATE INTERACTIONS *
2091 **************************/
2093 r11 = _mm256_mul_pd(rsq11,rinv11);
2094 r11 = _mm256_andnot_pd(dummy_mask,r11);
2096 /* Calculate table index by multiplying r with table scale and truncate to integer */
2097 rt = _mm256_mul_pd(r11,vftabscale);
2098 vfitab = _mm256_cvttpd_epi32(rt);
2099 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2100 vfitab = _mm_slli_epi32(vfitab,2);
2102 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2103 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2104 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2105 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2106 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2107 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2108 Heps = _mm256_mul_pd(vfeps,H);
2109 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2110 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2111 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
2115 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2117 /* Calculate temporary vectorial force */
2118 tx = _mm256_mul_pd(fscal,dx11);
2119 ty = _mm256_mul_pd(fscal,dy11);
2120 tz = _mm256_mul_pd(fscal,dz11);
2122 /* Update vectorial force */
2123 fix1 = _mm256_add_pd(fix1,tx);
2124 fiy1 = _mm256_add_pd(fiy1,ty);
2125 fiz1 = _mm256_add_pd(fiz1,tz);
2127 fjx1 = _mm256_add_pd(fjx1,tx);
2128 fjy1 = _mm256_add_pd(fjy1,ty);
2129 fjz1 = _mm256_add_pd(fjz1,tz);
2131 /**************************
2132 * CALCULATE INTERACTIONS *
2133 **************************/
2135 r12 = _mm256_mul_pd(rsq12,rinv12);
2136 r12 = _mm256_andnot_pd(dummy_mask,r12);
2138 /* Calculate table index by multiplying r with table scale and truncate to integer */
2139 rt = _mm256_mul_pd(r12,vftabscale);
2140 vfitab = _mm256_cvttpd_epi32(rt);
2141 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2142 vfitab = _mm_slli_epi32(vfitab,2);
2144 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2145 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2146 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2147 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2148 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2149 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2150 Heps = _mm256_mul_pd(vfeps,H);
2151 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2152 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2153 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
2157 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2159 /* Calculate temporary vectorial force */
2160 tx = _mm256_mul_pd(fscal,dx12);
2161 ty = _mm256_mul_pd(fscal,dy12);
2162 tz = _mm256_mul_pd(fscal,dz12);
2164 /* Update vectorial force */
2165 fix1 = _mm256_add_pd(fix1,tx);
2166 fiy1 = _mm256_add_pd(fiy1,ty);
2167 fiz1 = _mm256_add_pd(fiz1,tz);
2169 fjx2 = _mm256_add_pd(fjx2,tx);
2170 fjy2 = _mm256_add_pd(fjy2,ty);
2171 fjz2 = _mm256_add_pd(fjz2,tz);
2173 /**************************
2174 * CALCULATE INTERACTIONS *
2175 **************************/
2177 r20 = _mm256_mul_pd(rsq20,rinv20);
2178 r20 = _mm256_andnot_pd(dummy_mask,r20);
2180 /* Calculate table index by multiplying r with table scale and truncate to integer */
2181 rt = _mm256_mul_pd(r20,vftabscale);
2182 vfitab = _mm256_cvttpd_epi32(rt);
2183 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2184 vfitab = _mm_slli_epi32(vfitab,2);
2186 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2187 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2188 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2189 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2190 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2191 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2192 Heps = _mm256_mul_pd(vfeps,H);
2193 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2194 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2195 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
2199 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2201 /* Calculate temporary vectorial force */
2202 tx = _mm256_mul_pd(fscal,dx20);
2203 ty = _mm256_mul_pd(fscal,dy20);
2204 tz = _mm256_mul_pd(fscal,dz20);
2206 /* Update vectorial force */
2207 fix2 = _mm256_add_pd(fix2,tx);
2208 fiy2 = _mm256_add_pd(fiy2,ty);
2209 fiz2 = _mm256_add_pd(fiz2,tz);
2211 fjx0 = _mm256_add_pd(fjx0,tx);
2212 fjy0 = _mm256_add_pd(fjy0,ty);
2213 fjz0 = _mm256_add_pd(fjz0,tz);
2215 /**************************
2216 * CALCULATE INTERACTIONS *
2217 **************************/
2219 r21 = _mm256_mul_pd(rsq21,rinv21);
2220 r21 = _mm256_andnot_pd(dummy_mask,r21);
2222 /* Calculate table index by multiplying r with table scale and truncate to integer */
2223 rt = _mm256_mul_pd(r21,vftabscale);
2224 vfitab = _mm256_cvttpd_epi32(rt);
2225 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2226 vfitab = _mm_slli_epi32(vfitab,2);
2228 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2229 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2230 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2231 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2232 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2233 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2234 Heps = _mm256_mul_pd(vfeps,H);
2235 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2236 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2237 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
2241 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2243 /* Calculate temporary vectorial force */
2244 tx = _mm256_mul_pd(fscal,dx21);
2245 ty = _mm256_mul_pd(fscal,dy21);
2246 tz = _mm256_mul_pd(fscal,dz21);
2248 /* Update vectorial force */
2249 fix2 = _mm256_add_pd(fix2,tx);
2250 fiy2 = _mm256_add_pd(fiy2,ty);
2251 fiz2 = _mm256_add_pd(fiz2,tz);
2253 fjx1 = _mm256_add_pd(fjx1,tx);
2254 fjy1 = _mm256_add_pd(fjy1,ty);
2255 fjz1 = _mm256_add_pd(fjz1,tz);
2257 /**************************
2258 * CALCULATE INTERACTIONS *
2259 **************************/
2261 r22 = _mm256_mul_pd(rsq22,rinv22);
2262 r22 = _mm256_andnot_pd(dummy_mask,r22);
2264 /* Calculate table index by multiplying r with table scale and truncate to integer */
2265 rt = _mm256_mul_pd(r22,vftabscale);
2266 vfitab = _mm256_cvttpd_epi32(rt);
2267 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2268 vfitab = _mm_slli_epi32(vfitab,2);
2270 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2271 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2272 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2273 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2274 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2275 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2276 Heps = _mm256_mul_pd(vfeps,H);
2277 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2278 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2279 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
2283 fscal = _mm256_andnot_pd(dummy_mask,fscal);
2285 /* Calculate temporary vectorial force */
2286 tx = _mm256_mul_pd(fscal,dx22);
2287 ty = _mm256_mul_pd(fscal,dy22);
2288 tz = _mm256_mul_pd(fscal,dz22);
2290 /* Update vectorial force */
2291 fix2 = _mm256_add_pd(fix2,tx);
2292 fiy2 = _mm256_add_pd(fiy2,ty);
2293 fiz2 = _mm256_add_pd(fiz2,tz);
2295 fjx2 = _mm256_add_pd(fjx2,tx);
2296 fjy2 = _mm256_add_pd(fjy2,ty);
2297 fjz2 = _mm256_add_pd(fjz2,tz);
2299 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2300 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2301 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2302 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2304 gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2305 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2307 /* Inner loop uses 360 flops */
2310 /* End of innermost loop */
2312 gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2313 f+i_coord_offset,fshift+i_shift_offset);
2315 /* Increment number of inner iterations */
2316 inneriter += j_index_end - j_index_start;
2318 /* Outer loop uses 18 flops */
2321 /* Increment number of outer iterations */
2324 /* Update outer/inner flops */
2326 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*360);