2 * Note: this file was generated by the Gromacs avx_256_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_avx_256_double.h"
34 #include "kernelutil_x86_avx_256_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_256_double
38 * Electrostatics interaction: CubicSplineTable
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_256_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
63 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
64 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
66 real *shiftvec,*fshift,*x,*f;
67 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
69 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 real * vdwioffsetptr1;
71 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
72 real * vdwioffsetptr2;
73 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
74 real * vdwioffsetptr3;
75 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
76 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
77 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
78 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
79 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
80 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
81 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
84 __m128i ifour = _mm_set1_epi32(4);
85 __m256d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
87 __m256d dummy_mask,cutoff_mask;
88 __m128 tmpmask0,tmpmask1;
89 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
90 __m256d one = _mm256_set1_pd(1.0);
91 __m256d two = _mm256_set1_pd(2.0);
97 jindex = nlist->jindex;
99 shiftidx = nlist->shift;
101 shiftvec = fr->shift_vec[0];
102 fshift = fr->fshift[0];
103 facel = _mm256_set1_pd(fr->epsfac);
104 charge = mdatoms->chargeA;
106 vftab = kernel_data->table_elec->data;
107 vftabscale = _mm256_set1_pd(kernel_data->table_elec->scale);
109 /* Setup water-specific parameters */
110 inr = nlist->iinr[0];
111 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
112 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
113 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
115 /* Avoid stupid compiler warnings */
116 jnrA = jnrB = jnrC = jnrD = 0;
125 for(iidx=0;iidx<4*DIM;iidx++)
130 /* Start outer loop over neighborlists */
131 for(iidx=0; iidx<nri; iidx++)
133 /* Load shift vector for this list */
134 i_shift_offset = DIM*shiftidx[iidx];
136 /* Load limits for loop over neighbors */
137 j_index_start = jindex[iidx];
138 j_index_end = jindex[iidx+1];
140 /* Get outer coordinate index */
142 i_coord_offset = DIM*inr;
144 /* Load i particle coords and add shift vector */
145 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
146 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
148 fix1 = _mm256_setzero_pd();
149 fiy1 = _mm256_setzero_pd();
150 fiz1 = _mm256_setzero_pd();
151 fix2 = _mm256_setzero_pd();
152 fiy2 = _mm256_setzero_pd();
153 fiz2 = _mm256_setzero_pd();
154 fix3 = _mm256_setzero_pd();
155 fiy3 = _mm256_setzero_pd();
156 fiz3 = _mm256_setzero_pd();
158 /* Reset potential sums */
159 velecsum = _mm256_setzero_pd();
161 /* Start inner kernel loop */
162 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
165 /* Get j neighbor index, and coordinate index */
170 j_coord_offsetA = DIM*jnrA;
171 j_coord_offsetB = DIM*jnrB;
172 j_coord_offsetC = DIM*jnrC;
173 j_coord_offsetD = DIM*jnrD;
175 /* load j atom coordinates */
176 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
177 x+j_coord_offsetC,x+j_coord_offsetD,
180 /* Calculate displacement vector */
181 dx10 = _mm256_sub_pd(ix1,jx0);
182 dy10 = _mm256_sub_pd(iy1,jy0);
183 dz10 = _mm256_sub_pd(iz1,jz0);
184 dx20 = _mm256_sub_pd(ix2,jx0);
185 dy20 = _mm256_sub_pd(iy2,jy0);
186 dz20 = _mm256_sub_pd(iz2,jz0);
187 dx30 = _mm256_sub_pd(ix3,jx0);
188 dy30 = _mm256_sub_pd(iy3,jy0);
189 dz30 = _mm256_sub_pd(iz3,jz0);
191 /* Calculate squared distance and things based on it */
192 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
193 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
194 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
196 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
197 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
198 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
200 /* Load parameters for j particles */
201 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
202 charge+jnrC+0,charge+jnrD+0);
204 fjx0 = _mm256_setzero_pd();
205 fjy0 = _mm256_setzero_pd();
206 fjz0 = _mm256_setzero_pd();
208 /**************************
209 * CALCULATE INTERACTIONS *
210 **************************/
212 r10 = _mm256_mul_pd(rsq10,rinv10);
214 /* Compute parameters for interactions between i and j atoms */
215 qq10 = _mm256_mul_pd(iq1,jq0);
217 /* Calculate table index by multiplying r with table scale and truncate to integer */
218 rt = _mm256_mul_pd(r10,vftabscale);
219 vfitab = _mm256_cvttpd_epi32(rt);
220 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
221 vfitab = _mm_slli_epi32(vfitab,2);
223 /* CUBIC SPLINE TABLE ELECTROSTATICS */
224 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
225 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
226 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
227 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
228 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
229 Heps = _mm256_mul_pd(vfeps,H);
230 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
231 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
232 velec = _mm256_mul_pd(qq10,VV);
233 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
234 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
236 /* Update potential sum for this i atom from the interaction with this j atom. */
237 velecsum = _mm256_add_pd(velecsum,velec);
241 /* Calculate temporary vectorial force */
242 tx = _mm256_mul_pd(fscal,dx10);
243 ty = _mm256_mul_pd(fscal,dy10);
244 tz = _mm256_mul_pd(fscal,dz10);
246 /* Update vectorial force */
247 fix1 = _mm256_add_pd(fix1,tx);
248 fiy1 = _mm256_add_pd(fiy1,ty);
249 fiz1 = _mm256_add_pd(fiz1,tz);
251 fjx0 = _mm256_add_pd(fjx0,tx);
252 fjy0 = _mm256_add_pd(fjy0,ty);
253 fjz0 = _mm256_add_pd(fjz0,tz);
255 /**************************
256 * CALCULATE INTERACTIONS *
257 **************************/
259 r20 = _mm256_mul_pd(rsq20,rinv20);
261 /* Compute parameters for interactions between i and j atoms */
262 qq20 = _mm256_mul_pd(iq2,jq0);
264 /* Calculate table index by multiplying r with table scale and truncate to integer */
265 rt = _mm256_mul_pd(r20,vftabscale);
266 vfitab = _mm256_cvttpd_epi32(rt);
267 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
268 vfitab = _mm_slli_epi32(vfitab,2);
270 /* CUBIC SPLINE TABLE ELECTROSTATICS */
271 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
272 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
273 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
274 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
275 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
276 Heps = _mm256_mul_pd(vfeps,H);
277 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
278 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
279 velec = _mm256_mul_pd(qq20,VV);
280 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
281 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
283 /* Update potential sum for this i atom from the interaction with this j atom. */
284 velecsum = _mm256_add_pd(velecsum,velec);
288 /* Calculate temporary vectorial force */
289 tx = _mm256_mul_pd(fscal,dx20);
290 ty = _mm256_mul_pd(fscal,dy20);
291 tz = _mm256_mul_pd(fscal,dz20);
293 /* Update vectorial force */
294 fix2 = _mm256_add_pd(fix2,tx);
295 fiy2 = _mm256_add_pd(fiy2,ty);
296 fiz2 = _mm256_add_pd(fiz2,tz);
298 fjx0 = _mm256_add_pd(fjx0,tx);
299 fjy0 = _mm256_add_pd(fjy0,ty);
300 fjz0 = _mm256_add_pd(fjz0,tz);
302 /**************************
303 * CALCULATE INTERACTIONS *
304 **************************/
306 r30 = _mm256_mul_pd(rsq30,rinv30);
308 /* Compute parameters for interactions between i and j atoms */
309 qq30 = _mm256_mul_pd(iq3,jq0);
311 /* Calculate table index by multiplying r with table scale and truncate to integer */
312 rt = _mm256_mul_pd(r30,vftabscale);
313 vfitab = _mm256_cvttpd_epi32(rt);
314 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
315 vfitab = _mm_slli_epi32(vfitab,2);
317 /* CUBIC SPLINE TABLE ELECTROSTATICS */
318 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
319 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
320 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
321 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
322 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
323 Heps = _mm256_mul_pd(vfeps,H);
324 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
325 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
326 velec = _mm256_mul_pd(qq30,VV);
327 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
328 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
330 /* Update potential sum for this i atom from the interaction with this j atom. */
331 velecsum = _mm256_add_pd(velecsum,velec);
335 /* Calculate temporary vectorial force */
336 tx = _mm256_mul_pd(fscal,dx30);
337 ty = _mm256_mul_pd(fscal,dy30);
338 tz = _mm256_mul_pd(fscal,dz30);
340 /* Update vectorial force */
341 fix3 = _mm256_add_pd(fix3,tx);
342 fiy3 = _mm256_add_pd(fiy3,ty);
343 fiz3 = _mm256_add_pd(fiz3,tz);
345 fjx0 = _mm256_add_pd(fjx0,tx);
346 fjy0 = _mm256_add_pd(fjy0,ty);
347 fjz0 = _mm256_add_pd(fjz0,tz);
349 fjptrA = f+j_coord_offsetA;
350 fjptrB = f+j_coord_offsetB;
351 fjptrC = f+j_coord_offsetC;
352 fjptrD = f+j_coord_offsetD;
354 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
356 /* Inner loop uses 132 flops */
362 /* Get j neighbor index, and coordinate index */
363 jnrlistA = jjnr[jidx];
364 jnrlistB = jjnr[jidx+1];
365 jnrlistC = jjnr[jidx+2];
366 jnrlistD = jjnr[jidx+3];
367 /* Sign of each element will be negative for non-real atoms.
368 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
369 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
371 tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
373 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
374 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
375 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
377 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
378 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
379 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
380 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
381 j_coord_offsetA = DIM*jnrA;
382 j_coord_offsetB = DIM*jnrB;
383 j_coord_offsetC = DIM*jnrC;
384 j_coord_offsetD = DIM*jnrD;
386 /* load j atom coordinates */
387 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
388 x+j_coord_offsetC,x+j_coord_offsetD,
391 /* Calculate displacement vector */
392 dx10 = _mm256_sub_pd(ix1,jx0);
393 dy10 = _mm256_sub_pd(iy1,jy0);
394 dz10 = _mm256_sub_pd(iz1,jz0);
395 dx20 = _mm256_sub_pd(ix2,jx0);
396 dy20 = _mm256_sub_pd(iy2,jy0);
397 dz20 = _mm256_sub_pd(iz2,jz0);
398 dx30 = _mm256_sub_pd(ix3,jx0);
399 dy30 = _mm256_sub_pd(iy3,jy0);
400 dz30 = _mm256_sub_pd(iz3,jz0);
402 /* Calculate squared distance and things based on it */
403 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
404 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
405 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
407 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
408 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
409 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
411 /* Load parameters for j particles */
412 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
413 charge+jnrC+0,charge+jnrD+0);
415 fjx0 = _mm256_setzero_pd();
416 fjy0 = _mm256_setzero_pd();
417 fjz0 = _mm256_setzero_pd();
419 /**************************
420 * CALCULATE INTERACTIONS *
421 **************************/
423 r10 = _mm256_mul_pd(rsq10,rinv10);
424 r10 = _mm256_andnot_pd(dummy_mask,r10);
426 /* Compute parameters for interactions between i and j atoms */
427 qq10 = _mm256_mul_pd(iq1,jq0);
429 /* Calculate table index by multiplying r with table scale and truncate to integer */
430 rt = _mm256_mul_pd(r10,vftabscale);
431 vfitab = _mm256_cvttpd_epi32(rt);
432 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
433 vfitab = _mm_slli_epi32(vfitab,2);
435 /* CUBIC SPLINE TABLE ELECTROSTATICS */
436 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
437 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
438 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
439 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
440 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
441 Heps = _mm256_mul_pd(vfeps,H);
442 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
443 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
444 velec = _mm256_mul_pd(qq10,VV);
445 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
446 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
448 /* Update potential sum for this i atom from the interaction with this j atom. */
449 velec = _mm256_andnot_pd(dummy_mask,velec);
450 velecsum = _mm256_add_pd(velecsum,velec);
454 fscal = _mm256_andnot_pd(dummy_mask,fscal);
456 /* Calculate temporary vectorial force */
457 tx = _mm256_mul_pd(fscal,dx10);
458 ty = _mm256_mul_pd(fscal,dy10);
459 tz = _mm256_mul_pd(fscal,dz10);
461 /* Update vectorial force */
462 fix1 = _mm256_add_pd(fix1,tx);
463 fiy1 = _mm256_add_pd(fiy1,ty);
464 fiz1 = _mm256_add_pd(fiz1,tz);
466 fjx0 = _mm256_add_pd(fjx0,tx);
467 fjy0 = _mm256_add_pd(fjy0,ty);
468 fjz0 = _mm256_add_pd(fjz0,tz);
470 /**************************
471 * CALCULATE INTERACTIONS *
472 **************************/
474 r20 = _mm256_mul_pd(rsq20,rinv20);
475 r20 = _mm256_andnot_pd(dummy_mask,r20);
477 /* Compute parameters for interactions between i and j atoms */
478 qq20 = _mm256_mul_pd(iq2,jq0);
480 /* Calculate table index by multiplying r with table scale and truncate to integer */
481 rt = _mm256_mul_pd(r20,vftabscale);
482 vfitab = _mm256_cvttpd_epi32(rt);
483 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
484 vfitab = _mm_slli_epi32(vfitab,2);
486 /* CUBIC SPLINE TABLE ELECTROSTATICS */
487 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
488 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
489 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
490 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
491 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
492 Heps = _mm256_mul_pd(vfeps,H);
493 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
494 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
495 velec = _mm256_mul_pd(qq20,VV);
496 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
497 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
499 /* Update potential sum for this i atom from the interaction with this j atom. */
500 velec = _mm256_andnot_pd(dummy_mask,velec);
501 velecsum = _mm256_add_pd(velecsum,velec);
505 fscal = _mm256_andnot_pd(dummy_mask,fscal);
507 /* Calculate temporary vectorial force */
508 tx = _mm256_mul_pd(fscal,dx20);
509 ty = _mm256_mul_pd(fscal,dy20);
510 tz = _mm256_mul_pd(fscal,dz20);
512 /* Update vectorial force */
513 fix2 = _mm256_add_pd(fix2,tx);
514 fiy2 = _mm256_add_pd(fiy2,ty);
515 fiz2 = _mm256_add_pd(fiz2,tz);
517 fjx0 = _mm256_add_pd(fjx0,tx);
518 fjy0 = _mm256_add_pd(fjy0,ty);
519 fjz0 = _mm256_add_pd(fjz0,tz);
521 /**************************
522 * CALCULATE INTERACTIONS *
523 **************************/
525 r30 = _mm256_mul_pd(rsq30,rinv30);
526 r30 = _mm256_andnot_pd(dummy_mask,r30);
528 /* Compute parameters for interactions between i and j atoms */
529 qq30 = _mm256_mul_pd(iq3,jq0);
531 /* Calculate table index by multiplying r with table scale and truncate to integer */
532 rt = _mm256_mul_pd(r30,vftabscale);
533 vfitab = _mm256_cvttpd_epi32(rt);
534 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
535 vfitab = _mm_slli_epi32(vfitab,2);
537 /* CUBIC SPLINE TABLE ELECTROSTATICS */
538 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
539 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
540 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
541 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
542 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
543 Heps = _mm256_mul_pd(vfeps,H);
544 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
545 VV = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
546 velec = _mm256_mul_pd(qq30,VV);
547 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
548 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
550 /* Update potential sum for this i atom from the interaction with this j atom. */
551 velec = _mm256_andnot_pd(dummy_mask,velec);
552 velecsum = _mm256_add_pd(velecsum,velec);
556 fscal = _mm256_andnot_pd(dummy_mask,fscal);
558 /* Calculate temporary vectorial force */
559 tx = _mm256_mul_pd(fscal,dx30);
560 ty = _mm256_mul_pd(fscal,dy30);
561 tz = _mm256_mul_pd(fscal,dz30);
563 /* Update vectorial force */
564 fix3 = _mm256_add_pd(fix3,tx);
565 fiy3 = _mm256_add_pd(fiy3,ty);
566 fiz3 = _mm256_add_pd(fiz3,tz);
568 fjx0 = _mm256_add_pd(fjx0,tx);
569 fjy0 = _mm256_add_pd(fjy0,ty);
570 fjz0 = _mm256_add_pd(fjz0,tz);
572 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
573 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
574 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
575 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
577 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
579 /* Inner loop uses 135 flops */
582 /* End of innermost loop */
584 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
585 f+i_coord_offset+DIM,fshift+i_shift_offset);
588 /* Update potential energies */
589 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
591 /* Increment number of inner iterations */
592 inneriter += j_index_end - j_index_start;
594 /* Outer loop uses 19 flops */
597 /* Increment number of outer iterations */
600 /* Update outer/inner flops */
602 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*135);
605 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_256_double
606 * Electrostatics interaction: CubicSplineTable
607 * VdW interaction: None
608 * Geometry: Water4-Particle
609 * Calculate force/pot: Force
612 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_256_double
613 (t_nblist * gmx_restrict nlist,
614 rvec * gmx_restrict xx,
615 rvec * gmx_restrict ff,
616 t_forcerec * gmx_restrict fr,
617 t_mdatoms * gmx_restrict mdatoms,
618 nb_kernel_data_t * gmx_restrict kernel_data,
619 t_nrnb * gmx_restrict nrnb)
621 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
622 * just 0 for non-waters.
623 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
624 * jnr indices corresponding to data put in the four positions in the SIMD register.
626 int i_shift_offset,i_coord_offset,outeriter,inneriter;
627 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
628 int jnrA,jnrB,jnrC,jnrD;
629 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
630 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
631 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
632 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
634 real *shiftvec,*fshift,*x,*f;
635 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
637 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
638 real * vdwioffsetptr1;
639 __m256d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
640 real * vdwioffsetptr2;
641 __m256d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
642 real * vdwioffsetptr3;
643 __m256d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
644 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
645 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
646 __m256d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
647 __m256d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
648 __m256d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
649 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
652 __m128i ifour = _mm_set1_epi32(4);
653 __m256d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
655 __m256d dummy_mask,cutoff_mask;
656 __m128 tmpmask0,tmpmask1;
657 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
658 __m256d one = _mm256_set1_pd(1.0);
659 __m256d two = _mm256_set1_pd(2.0);
665 jindex = nlist->jindex;
667 shiftidx = nlist->shift;
669 shiftvec = fr->shift_vec[0];
670 fshift = fr->fshift[0];
671 facel = _mm256_set1_pd(fr->epsfac);
672 charge = mdatoms->chargeA;
674 vftab = kernel_data->table_elec->data;
675 vftabscale = _mm256_set1_pd(kernel_data->table_elec->scale);
677 /* Setup water-specific parameters */
678 inr = nlist->iinr[0];
679 iq1 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
680 iq2 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
681 iq3 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
683 /* Avoid stupid compiler warnings */
684 jnrA = jnrB = jnrC = jnrD = 0;
693 for(iidx=0;iidx<4*DIM;iidx++)
698 /* Start outer loop over neighborlists */
699 for(iidx=0; iidx<nri; iidx++)
701 /* Load shift vector for this list */
702 i_shift_offset = DIM*shiftidx[iidx];
704 /* Load limits for loop over neighbors */
705 j_index_start = jindex[iidx];
706 j_index_end = jindex[iidx+1];
708 /* Get outer coordinate index */
710 i_coord_offset = DIM*inr;
712 /* Load i particle coords and add shift vector */
713 gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
714 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
716 fix1 = _mm256_setzero_pd();
717 fiy1 = _mm256_setzero_pd();
718 fiz1 = _mm256_setzero_pd();
719 fix2 = _mm256_setzero_pd();
720 fiy2 = _mm256_setzero_pd();
721 fiz2 = _mm256_setzero_pd();
722 fix3 = _mm256_setzero_pd();
723 fiy3 = _mm256_setzero_pd();
724 fiz3 = _mm256_setzero_pd();
726 /* Start inner kernel loop */
727 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
730 /* Get j neighbor index, and coordinate index */
735 j_coord_offsetA = DIM*jnrA;
736 j_coord_offsetB = DIM*jnrB;
737 j_coord_offsetC = DIM*jnrC;
738 j_coord_offsetD = DIM*jnrD;
740 /* load j atom coordinates */
741 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
742 x+j_coord_offsetC,x+j_coord_offsetD,
745 /* Calculate displacement vector */
746 dx10 = _mm256_sub_pd(ix1,jx0);
747 dy10 = _mm256_sub_pd(iy1,jy0);
748 dz10 = _mm256_sub_pd(iz1,jz0);
749 dx20 = _mm256_sub_pd(ix2,jx0);
750 dy20 = _mm256_sub_pd(iy2,jy0);
751 dz20 = _mm256_sub_pd(iz2,jz0);
752 dx30 = _mm256_sub_pd(ix3,jx0);
753 dy30 = _mm256_sub_pd(iy3,jy0);
754 dz30 = _mm256_sub_pd(iz3,jz0);
756 /* Calculate squared distance and things based on it */
757 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
758 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
759 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
761 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
762 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
763 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
765 /* Load parameters for j particles */
766 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
767 charge+jnrC+0,charge+jnrD+0);
769 fjx0 = _mm256_setzero_pd();
770 fjy0 = _mm256_setzero_pd();
771 fjz0 = _mm256_setzero_pd();
773 /**************************
774 * CALCULATE INTERACTIONS *
775 **************************/
777 r10 = _mm256_mul_pd(rsq10,rinv10);
779 /* Compute parameters for interactions between i and j atoms */
780 qq10 = _mm256_mul_pd(iq1,jq0);
782 /* Calculate table index by multiplying r with table scale and truncate to integer */
783 rt = _mm256_mul_pd(r10,vftabscale);
784 vfitab = _mm256_cvttpd_epi32(rt);
785 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
786 vfitab = _mm_slli_epi32(vfitab,2);
788 /* CUBIC SPLINE TABLE ELECTROSTATICS */
789 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
790 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
791 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
792 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
793 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
794 Heps = _mm256_mul_pd(vfeps,H);
795 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
796 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
797 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
801 /* Calculate temporary vectorial force */
802 tx = _mm256_mul_pd(fscal,dx10);
803 ty = _mm256_mul_pd(fscal,dy10);
804 tz = _mm256_mul_pd(fscal,dz10);
806 /* Update vectorial force */
807 fix1 = _mm256_add_pd(fix1,tx);
808 fiy1 = _mm256_add_pd(fiy1,ty);
809 fiz1 = _mm256_add_pd(fiz1,tz);
811 fjx0 = _mm256_add_pd(fjx0,tx);
812 fjy0 = _mm256_add_pd(fjy0,ty);
813 fjz0 = _mm256_add_pd(fjz0,tz);
815 /**************************
816 * CALCULATE INTERACTIONS *
817 **************************/
819 r20 = _mm256_mul_pd(rsq20,rinv20);
821 /* Compute parameters for interactions between i and j atoms */
822 qq20 = _mm256_mul_pd(iq2,jq0);
824 /* Calculate table index by multiplying r with table scale and truncate to integer */
825 rt = _mm256_mul_pd(r20,vftabscale);
826 vfitab = _mm256_cvttpd_epi32(rt);
827 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
828 vfitab = _mm_slli_epi32(vfitab,2);
830 /* CUBIC SPLINE TABLE ELECTROSTATICS */
831 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
832 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
833 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
834 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
835 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
836 Heps = _mm256_mul_pd(vfeps,H);
837 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
838 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
839 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
843 /* Calculate temporary vectorial force */
844 tx = _mm256_mul_pd(fscal,dx20);
845 ty = _mm256_mul_pd(fscal,dy20);
846 tz = _mm256_mul_pd(fscal,dz20);
848 /* Update vectorial force */
849 fix2 = _mm256_add_pd(fix2,tx);
850 fiy2 = _mm256_add_pd(fiy2,ty);
851 fiz2 = _mm256_add_pd(fiz2,tz);
853 fjx0 = _mm256_add_pd(fjx0,tx);
854 fjy0 = _mm256_add_pd(fjy0,ty);
855 fjz0 = _mm256_add_pd(fjz0,tz);
857 /**************************
858 * CALCULATE INTERACTIONS *
859 **************************/
861 r30 = _mm256_mul_pd(rsq30,rinv30);
863 /* Compute parameters for interactions between i and j atoms */
864 qq30 = _mm256_mul_pd(iq3,jq0);
866 /* Calculate table index by multiplying r with table scale and truncate to integer */
867 rt = _mm256_mul_pd(r30,vftabscale);
868 vfitab = _mm256_cvttpd_epi32(rt);
869 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
870 vfitab = _mm_slli_epi32(vfitab,2);
872 /* CUBIC SPLINE TABLE ELECTROSTATICS */
873 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
874 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
875 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
876 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
877 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
878 Heps = _mm256_mul_pd(vfeps,H);
879 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
880 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
881 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
885 /* Calculate temporary vectorial force */
886 tx = _mm256_mul_pd(fscal,dx30);
887 ty = _mm256_mul_pd(fscal,dy30);
888 tz = _mm256_mul_pd(fscal,dz30);
890 /* Update vectorial force */
891 fix3 = _mm256_add_pd(fix3,tx);
892 fiy3 = _mm256_add_pd(fiy3,ty);
893 fiz3 = _mm256_add_pd(fiz3,tz);
895 fjx0 = _mm256_add_pd(fjx0,tx);
896 fjy0 = _mm256_add_pd(fjy0,ty);
897 fjz0 = _mm256_add_pd(fjz0,tz);
899 fjptrA = f+j_coord_offsetA;
900 fjptrB = f+j_coord_offsetB;
901 fjptrC = f+j_coord_offsetC;
902 fjptrD = f+j_coord_offsetD;
904 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
906 /* Inner loop uses 120 flops */
912 /* Get j neighbor index, and coordinate index */
913 jnrlistA = jjnr[jidx];
914 jnrlistB = jjnr[jidx+1];
915 jnrlistC = jjnr[jidx+2];
916 jnrlistD = jjnr[jidx+3];
917 /* Sign of each element will be negative for non-real atoms.
918 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
919 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
921 tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
923 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
924 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
925 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
927 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
928 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
929 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
930 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
931 j_coord_offsetA = DIM*jnrA;
932 j_coord_offsetB = DIM*jnrB;
933 j_coord_offsetC = DIM*jnrC;
934 j_coord_offsetD = DIM*jnrD;
936 /* load j atom coordinates */
937 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
938 x+j_coord_offsetC,x+j_coord_offsetD,
941 /* Calculate displacement vector */
942 dx10 = _mm256_sub_pd(ix1,jx0);
943 dy10 = _mm256_sub_pd(iy1,jy0);
944 dz10 = _mm256_sub_pd(iz1,jz0);
945 dx20 = _mm256_sub_pd(ix2,jx0);
946 dy20 = _mm256_sub_pd(iy2,jy0);
947 dz20 = _mm256_sub_pd(iz2,jz0);
948 dx30 = _mm256_sub_pd(ix3,jx0);
949 dy30 = _mm256_sub_pd(iy3,jy0);
950 dz30 = _mm256_sub_pd(iz3,jz0);
952 /* Calculate squared distance and things based on it */
953 rsq10 = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
954 rsq20 = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
955 rsq30 = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
957 rinv10 = gmx_mm256_invsqrt_pd(rsq10);
958 rinv20 = gmx_mm256_invsqrt_pd(rsq20);
959 rinv30 = gmx_mm256_invsqrt_pd(rsq30);
961 /* Load parameters for j particles */
962 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
963 charge+jnrC+0,charge+jnrD+0);
965 fjx0 = _mm256_setzero_pd();
966 fjy0 = _mm256_setzero_pd();
967 fjz0 = _mm256_setzero_pd();
969 /**************************
970 * CALCULATE INTERACTIONS *
971 **************************/
973 r10 = _mm256_mul_pd(rsq10,rinv10);
974 r10 = _mm256_andnot_pd(dummy_mask,r10);
976 /* Compute parameters for interactions between i and j atoms */
977 qq10 = _mm256_mul_pd(iq1,jq0);
979 /* Calculate table index by multiplying r with table scale and truncate to integer */
980 rt = _mm256_mul_pd(r10,vftabscale);
981 vfitab = _mm256_cvttpd_epi32(rt);
982 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
983 vfitab = _mm_slli_epi32(vfitab,2);
985 /* CUBIC SPLINE TABLE ELECTROSTATICS */
986 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
987 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
988 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
989 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
990 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
991 Heps = _mm256_mul_pd(vfeps,H);
992 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
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(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
998 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1000 /* Calculate temporary vectorial force */
1001 tx = _mm256_mul_pd(fscal,dx10);
1002 ty = _mm256_mul_pd(fscal,dy10);
1003 tz = _mm256_mul_pd(fscal,dz10);
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 fjx0 = _mm256_add_pd(fjx0,tx);
1011 fjy0 = _mm256_add_pd(fjy0,ty);
1012 fjz0 = _mm256_add_pd(fjz0,tz);
1014 /**************************
1015 * CALCULATE INTERACTIONS *
1016 **************************/
1018 r20 = _mm256_mul_pd(rsq20,rinv20);
1019 r20 = _mm256_andnot_pd(dummy_mask,r20);
1021 /* Compute parameters for interactions between i and j atoms */
1022 qq20 = _mm256_mul_pd(iq2,jq0);
1024 /* Calculate table index by multiplying r with table scale and truncate to integer */
1025 rt = _mm256_mul_pd(r20,vftabscale);
1026 vfitab = _mm256_cvttpd_epi32(rt);
1027 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1028 vfitab = _mm_slli_epi32(vfitab,2);
1030 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1031 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1032 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1033 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1034 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1035 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1036 Heps = _mm256_mul_pd(vfeps,H);
1037 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1038 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1039 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1043 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1045 /* Calculate temporary vectorial force */
1046 tx = _mm256_mul_pd(fscal,dx20);
1047 ty = _mm256_mul_pd(fscal,dy20);
1048 tz = _mm256_mul_pd(fscal,dz20);
1050 /* Update vectorial force */
1051 fix2 = _mm256_add_pd(fix2,tx);
1052 fiy2 = _mm256_add_pd(fiy2,ty);
1053 fiz2 = _mm256_add_pd(fiz2,tz);
1055 fjx0 = _mm256_add_pd(fjx0,tx);
1056 fjy0 = _mm256_add_pd(fjy0,ty);
1057 fjz0 = _mm256_add_pd(fjz0,tz);
1059 /**************************
1060 * CALCULATE INTERACTIONS *
1061 **************************/
1063 r30 = _mm256_mul_pd(rsq30,rinv30);
1064 r30 = _mm256_andnot_pd(dummy_mask,r30);
1066 /* Compute parameters for interactions between i and j atoms */
1067 qq30 = _mm256_mul_pd(iq3,jq0);
1069 /* Calculate table index by multiplying r with table scale and truncate to integer */
1070 rt = _mm256_mul_pd(r30,vftabscale);
1071 vfitab = _mm256_cvttpd_epi32(rt);
1072 vfeps = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1073 vfitab = _mm_slli_epi32(vfitab,2);
1075 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1076 Y = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1077 F = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1078 G = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1079 H = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1080 GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1081 Heps = _mm256_mul_pd(vfeps,H);
1082 Fp = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1083 FF = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1084 felec = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
1088 fscal = _mm256_andnot_pd(dummy_mask,fscal);
1090 /* Calculate temporary vectorial force */
1091 tx = _mm256_mul_pd(fscal,dx30);
1092 ty = _mm256_mul_pd(fscal,dy30);
1093 tz = _mm256_mul_pd(fscal,dz30);
1095 /* Update vectorial force */
1096 fix3 = _mm256_add_pd(fix3,tx);
1097 fiy3 = _mm256_add_pd(fiy3,ty);
1098 fiz3 = _mm256_add_pd(fiz3,tz);
1100 fjx0 = _mm256_add_pd(fjx0,tx);
1101 fjy0 = _mm256_add_pd(fjy0,ty);
1102 fjz0 = _mm256_add_pd(fjz0,tz);
1104 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1105 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1106 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1107 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1109 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1111 /* Inner loop uses 123 flops */
1114 /* End of innermost loop */
1116 gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1117 f+i_coord_offset+DIM,fshift+i_shift_offset);
1119 /* Increment number of inner iterations */
1120 inneriter += j_index_end - j_index_start;
1122 /* Outer loop uses 18 flops */
1125 /* Increment number of outer iterations */
1128 /* Update outer/inner flops */
1130 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*123);