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_128_fma_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
50 #include "kernelutil_x86_avx_128_fma_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_128_fma_double
54 * Electrostatics interaction: CubicSplineTable
55 * VdW interaction: None
56 * Geometry: Water4-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_128_fma_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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
88 int vdwjidx0A,vdwjidx0B;
89 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
91 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
92 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
93 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
96 __m128i ifour = _mm_set1_epi32(4);
97 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
99 __m128d dummy_mask,cutoff_mask;
100 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
101 __m128d one = _mm_set1_pd(1.0);
102 __m128d two = _mm_set1_pd(2.0);
108 jindex = nlist->jindex;
110 shiftidx = nlist->shift;
112 shiftvec = fr->shift_vec[0];
113 fshift = fr->fshift[0];
114 facel = _mm_set1_pd(fr->epsfac);
115 charge = mdatoms->chargeA;
117 vftab = kernel_data->table_elec->data;
118 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
120 /* Setup water-specific parameters */
121 inr = nlist->iinr[0];
122 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
123 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
124 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
126 /* Avoid stupid compiler warnings */
134 /* Start outer loop over neighborlists */
135 for(iidx=0; iidx<nri; iidx++)
137 /* Load shift vector for this list */
138 i_shift_offset = DIM*shiftidx[iidx];
140 /* Load limits for loop over neighbors */
141 j_index_start = jindex[iidx];
142 j_index_end = jindex[iidx+1];
144 /* Get outer coordinate index */
146 i_coord_offset = DIM*inr;
148 /* Load i particle coords and add shift vector */
149 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
150 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
152 fix1 = _mm_setzero_pd();
153 fiy1 = _mm_setzero_pd();
154 fiz1 = _mm_setzero_pd();
155 fix2 = _mm_setzero_pd();
156 fiy2 = _mm_setzero_pd();
157 fiz2 = _mm_setzero_pd();
158 fix3 = _mm_setzero_pd();
159 fiy3 = _mm_setzero_pd();
160 fiz3 = _mm_setzero_pd();
162 /* Reset potential sums */
163 velecsum = _mm_setzero_pd();
165 /* Start inner kernel loop */
166 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
169 /* Get j neighbor index, and coordinate index */
172 j_coord_offsetA = DIM*jnrA;
173 j_coord_offsetB = DIM*jnrB;
175 /* load j atom coordinates */
176 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
179 /* Calculate displacement vector */
180 dx10 = _mm_sub_pd(ix1,jx0);
181 dy10 = _mm_sub_pd(iy1,jy0);
182 dz10 = _mm_sub_pd(iz1,jz0);
183 dx20 = _mm_sub_pd(ix2,jx0);
184 dy20 = _mm_sub_pd(iy2,jy0);
185 dz20 = _mm_sub_pd(iz2,jz0);
186 dx30 = _mm_sub_pd(ix3,jx0);
187 dy30 = _mm_sub_pd(iy3,jy0);
188 dz30 = _mm_sub_pd(iz3,jz0);
190 /* Calculate squared distance and things based on it */
191 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
192 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
193 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
195 rinv10 = gmx_mm_invsqrt_pd(rsq10);
196 rinv20 = gmx_mm_invsqrt_pd(rsq20);
197 rinv30 = gmx_mm_invsqrt_pd(rsq30);
199 /* Load parameters for j particles */
200 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
202 fjx0 = _mm_setzero_pd();
203 fjy0 = _mm_setzero_pd();
204 fjz0 = _mm_setzero_pd();
206 /**************************
207 * CALCULATE INTERACTIONS *
208 **************************/
210 r10 = _mm_mul_pd(rsq10,rinv10);
212 /* Compute parameters for interactions between i and j atoms */
213 qq10 = _mm_mul_pd(iq1,jq0);
215 /* Calculate table index by multiplying r with table scale and truncate to integer */
216 rt = _mm_mul_pd(r10,vftabscale);
217 vfitab = _mm_cvttpd_epi32(rt);
219 vfeps = _mm_frcz_pd(rt);
221 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
223 twovfeps = _mm_add_pd(vfeps,vfeps);
224 vfitab = _mm_slli_epi32(vfitab,2);
226 /* CUBIC SPLINE TABLE ELECTROSTATICS */
227 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
228 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
229 GMX_MM_TRANSPOSE2_PD(Y,F);
230 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
231 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
232 GMX_MM_TRANSPOSE2_PD(G,H);
233 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
234 VV = _mm_macc_pd(vfeps,Fp,Y);
235 velec = _mm_mul_pd(qq10,VV);
236 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
237 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
239 /* Update potential sum for this i atom from the interaction with this j atom. */
240 velecsum = _mm_add_pd(velecsum,velec);
244 /* Update vectorial force */
245 fix1 = _mm_macc_pd(dx10,fscal,fix1);
246 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
247 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
249 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
250 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
251 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
253 /**************************
254 * CALCULATE INTERACTIONS *
255 **************************/
257 r20 = _mm_mul_pd(rsq20,rinv20);
259 /* Compute parameters for interactions between i and j atoms */
260 qq20 = _mm_mul_pd(iq2,jq0);
262 /* Calculate table index by multiplying r with table scale and truncate to integer */
263 rt = _mm_mul_pd(r20,vftabscale);
264 vfitab = _mm_cvttpd_epi32(rt);
266 vfeps = _mm_frcz_pd(rt);
268 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
270 twovfeps = _mm_add_pd(vfeps,vfeps);
271 vfitab = _mm_slli_epi32(vfitab,2);
273 /* CUBIC SPLINE TABLE ELECTROSTATICS */
274 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
275 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
276 GMX_MM_TRANSPOSE2_PD(Y,F);
277 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
278 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
279 GMX_MM_TRANSPOSE2_PD(G,H);
280 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
281 VV = _mm_macc_pd(vfeps,Fp,Y);
282 velec = _mm_mul_pd(qq20,VV);
283 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
284 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
286 /* Update potential sum for this i atom from the interaction with this j atom. */
287 velecsum = _mm_add_pd(velecsum,velec);
291 /* Update vectorial force */
292 fix2 = _mm_macc_pd(dx20,fscal,fix2);
293 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
294 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
296 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
297 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
298 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
300 /**************************
301 * CALCULATE INTERACTIONS *
302 **************************/
304 r30 = _mm_mul_pd(rsq30,rinv30);
306 /* Compute parameters for interactions between i and j atoms */
307 qq30 = _mm_mul_pd(iq3,jq0);
309 /* Calculate table index by multiplying r with table scale and truncate to integer */
310 rt = _mm_mul_pd(r30,vftabscale);
311 vfitab = _mm_cvttpd_epi32(rt);
313 vfeps = _mm_frcz_pd(rt);
315 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
317 twovfeps = _mm_add_pd(vfeps,vfeps);
318 vfitab = _mm_slli_epi32(vfitab,2);
320 /* CUBIC SPLINE TABLE ELECTROSTATICS */
321 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
322 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
323 GMX_MM_TRANSPOSE2_PD(Y,F);
324 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
325 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
326 GMX_MM_TRANSPOSE2_PD(G,H);
327 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
328 VV = _mm_macc_pd(vfeps,Fp,Y);
329 velec = _mm_mul_pd(qq30,VV);
330 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
331 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
333 /* Update potential sum for this i atom from the interaction with this j atom. */
334 velecsum = _mm_add_pd(velecsum,velec);
338 /* Update vectorial force */
339 fix3 = _mm_macc_pd(dx30,fscal,fix3);
340 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
341 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
343 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
344 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
345 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
347 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
349 /* Inner loop uses 141 flops */
356 j_coord_offsetA = DIM*jnrA;
358 /* load j atom coordinates */
359 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
362 /* Calculate displacement vector */
363 dx10 = _mm_sub_pd(ix1,jx0);
364 dy10 = _mm_sub_pd(iy1,jy0);
365 dz10 = _mm_sub_pd(iz1,jz0);
366 dx20 = _mm_sub_pd(ix2,jx0);
367 dy20 = _mm_sub_pd(iy2,jy0);
368 dz20 = _mm_sub_pd(iz2,jz0);
369 dx30 = _mm_sub_pd(ix3,jx0);
370 dy30 = _mm_sub_pd(iy3,jy0);
371 dz30 = _mm_sub_pd(iz3,jz0);
373 /* Calculate squared distance and things based on it */
374 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
375 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
376 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
378 rinv10 = gmx_mm_invsqrt_pd(rsq10);
379 rinv20 = gmx_mm_invsqrt_pd(rsq20);
380 rinv30 = gmx_mm_invsqrt_pd(rsq30);
382 /* Load parameters for j particles */
383 jq0 = _mm_load_sd(charge+jnrA+0);
385 fjx0 = _mm_setzero_pd();
386 fjy0 = _mm_setzero_pd();
387 fjz0 = _mm_setzero_pd();
389 /**************************
390 * CALCULATE INTERACTIONS *
391 **************************/
393 r10 = _mm_mul_pd(rsq10,rinv10);
395 /* Compute parameters for interactions between i and j atoms */
396 qq10 = _mm_mul_pd(iq1,jq0);
398 /* Calculate table index by multiplying r with table scale and truncate to integer */
399 rt = _mm_mul_pd(r10,vftabscale);
400 vfitab = _mm_cvttpd_epi32(rt);
402 vfeps = _mm_frcz_pd(rt);
404 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
406 twovfeps = _mm_add_pd(vfeps,vfeps);
407 vfitab = _mm_slli_epi32(vfitab,2);
409 /* CUBIC SPLINE TABLE ELECTROSTATICS */
410 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
411 F = _mm_setzero_pd();
412 GMX_MM_TRANSPOSE2_PD(Y,F);
413 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
414 H = _mm_setzero_pd();
415 GMX_MM_TRANSPOSE2_PD(G,H);
416 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
417 VV = _mm_macc_pd(vfeps,Fp,Y);
418 velec = _mm_mul_pd(qq10,VV);
419 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
420 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
422 /* Update potential sum for this i atom from the interaction with this j atom. */
423 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
424 velecsum = _mm_add_pd(velecsum,velec);
428 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
430 /* Update vectorial force */
431 fix1 = _mm_macc_pd(dx10,fscal,fix1);
432 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
433 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
435 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
436 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
437 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
439 /**************************
440 * CALCULATE INTERACTIONS *
441 **************************/
443 r20 = _mm_mul_pd(rsq20,rinv20);
445 /* Compute parameters for interactions between i and j atoms */
446 qq20 = _mm_mul_pd(iq2,jq0);
448 /* Calculate table index by multiplying r with table scale and truncate to integer */
449 rt = _mm_mul_pd(r20,vftabscale);
450 vfitab = _mm_cvttpd_epi32(rt);
452 vfeps = _mm_frcz_pd(rt);
454 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
456 twovfeps = _mm_add_pd(vfeps,vfeps);
457 vfitab = _mm_slli_epi32(vfitab,2);
459 /* CUBIC SPLINE TABLE ELECTROSTATICS */
460 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
461 F = _mm_setzero_pd();
462 GMX_MM_TRANSPOSE2_PD(Y,F);
463 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
464 H = _mm_setzero_pd();
465 GMX_MM_TRANSPOSE2_PD(G,H);
466 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
467 VV = _mm_macc_pd(vfeps,Fp,Y);
468 velec = _mm_mul_pd(qq20,VV);
469 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
470 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
472 /* Update potential sum for this i atom from the interaction with this j atom. */
473 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
474 velecsum = _mm_add_pd(velecsum,velec);
478 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
480 /* Update vectorial force */
481 fix2 = _mm_macc_pd(dx20,fscal,fix2);
482 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
483 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
485 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
486 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
487 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
489 /**************************
490 * CALCULATE INTERACTIONS *
491 **************************/
493 r30 = _mm_mul_pd(rsq30,rinv30);
495 /* Compute parameters for interactions between i and j atoms */
496 qq30 = _mm_mul_pd(iq3,jq0);
498 /* Calculate table index by multiplying r with table scale and truncate to integer */
499 rt = _mm_mul_pd(r30,vftabscale);
500 vfitab = _mm_cvttpd_epi32(rt);
502 vfeps = _mm_frcz_pd(rt);
504 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
506 twovfeps = _mm_add_pd(vfeps,vfeps);
507 vfitab = _mm_slli_epi32(vfitab,2);
509 /* CUBIC SPLINE TABLE ELECTROSTATICS */
510 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
511 F = _mm_setzero_pd();
512 GMX_MM_TRANSPOSE2_PD(Y,F);
513 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
514 H = _mm_setzero_pd();
515 GMX_MM_TRANSPOSE2_PD(G,H);
516 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
517 VV = _mm_macc_pd(vfeps,Fp,Y);
518 velec = _mm_mul_pd(qq30,VV);
519 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
520 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
522 /* Update potential sum for this i atom from the interaction with this j atom. */
523 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
524 velecsum = _mm_add_pd(velecsum,velec);
528 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
530 /* Update vectorial force */
531 fix3 = _mm_macc_pd(dx30,fscal,fix3);
532 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
533 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
535 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
536 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
537 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
539 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
541 /* Inner loop uses 141 flops */
544 /* End of innermost loop */
546 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
547 f+i_coord_offset+DIM,fshift+i_shift_offset);
550 /* Update potential energies */
551 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
553 /* Increment number of inner iterations */
554 inneriter += j_index_end - j_index_start;
556 /* Outer loop uses 19 flops */
559 /* Increment number of outer iterations */
562 /* Update outer/inner flops */
564 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*141);
567 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_128_fma_double
568 * Electrostatics interaction: CubicSplineTable
569 * VdW interaction: None
570 * Geometry: Water4-Particle
571 * Calculate force/pot: Force
574 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_128_fma_double
575 (t_nblist * gmx_restrict nlist,
576 rvec * gmx_restrict xx,
577 rvec * gmx_restrict ff,
578 t_forcerec * gmx_restrict fr,
579 t_mdatoms * gmx_restrict mdatoms,
580 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
581 t_nrnb * gmx_restrict nrnb)
583 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
584 * just 0 for non-waters.
585 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
586 * jnr indices corresponding to data put in the four positions in the SIMD register.
588 int i_shift_offset,i_coord_offset,outeriter,inneriter;
589 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
591 int j_coord_offsetA,j_coord_offsetB;
592 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
594 real *shiftvec,*fshift,*x,*f;
595 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
597 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
599 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
601 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
602 int vdwjidx0A,vdwjidx0B;
603 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
604 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
605 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
606 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
607 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
610 __m128i ifour = _mm_set1_epi32(4);
611 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
613 __m128d dummy_mask,cutoff_mask;
614 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
615 __m128d one = _mm_set1_pd(1.0);
616 __m128d two = _mm_set1_pd(2.0);
622 jindex = nlist->jindex;
624 shiftidx = nlist->shift;
626 shiftvec = fr->shift_vec[0];
627 fshift = fr->fshift[0];
628 facel = _mm_set1_pd(fr->epsfac);
629 charge = mdatoms->chargeA;
631 vftab = kernel_data->table_elec->data;
632 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
634 /* Setup water-specific parameters */
635 inr = nlist->iinr[0];
636 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
637 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
638 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
640 /* Avoid stupid compiler warnings */
648 /* Start outer loop over neighborlists */
649 for(iidx=0; iidx<nri; iidx++)
651 /* Load shift vector for this list */
652 i_shift_offset = DIM*shiftidx[iidx];
654 /* Load limits for loop over neighbors */
655 j_index_start = jindex[iidx];
656 j_index_end = jindex[iidx+1];
658 /* Get outer coordinate index */
660 i_coord_offset = DIM*inr;
662 /* Load i particle coords and add shift vector */
663 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
664 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
666 fix1 = _mm_setzero_pd();
667 fiy1 = _mm_setzero_pd();
668 fiz1 = _mm_setzero_pd();
669 fix2 = _mm_setzero_pd();
670 fiy2 = _mm_setzero_pd();
671 fiz2 = _mm_setzero_pd();
672 fix3 = _mm_setzero_pd();
673 fiy3 = _mm_setzero_pd();
674 fiz3 = _mm_setzero_pd();
676 /* Start inner kernel loop */
677 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
680 /* Get j neighbor index, and coordinate index */
683 j_coord_offsetA = DIM*jnrA;
684 j_coord_offsetB = DIM*jnrB;
686 /* load j atom coordinates */
687 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
690 /* Calculate displacement vector */
691 dx10 = _mm_sub_pd(ix1,jx0);
692 dy10 = _mm_sub_pd(iy1,jy0);
693 dz10 = _mm_sub_pd(iz1,jz0);
694 dx20 = _mm_sub_pd(ix2,jx0);
695 dy20 = _mm_sub_pd(iy2,jy0);
696 dz20 = _mm_sub_pd(iz2,jz0);
697 dx30 = _mm_sub_pd(ix3,jx0);
698 dy30 = _mm_sub_pd(iy3,jy0);
699 dz30 = _mm_sub_pd(iz3,jz0);
701 /* Calculate squared distance and things based on it */
702 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
703 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
704 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
706 rinv10 = gmx_mm_invsqrt_pd(rsq10);
707 rinv20 = gmx_mm_invsqrt_pd(rsq20);
708 rinv30 = gmx_mm_invsqrt_pd(rsq30);
710 /* Load parameters for j particles */
711 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
713 fjx0 = _mm_setzero_pd();
714 fjy0 = _mm_setzero_pd();
715 fjz0 = _mm_setzero_pd();
717 /**************************
718 * CALCULATE INTERACTIONS *
719 **************************/
721 r10 = _mm_mul_pd(rsq10,rinv10);
723 /* Compute parameters for interactions between i and j atoms */
724 qq10 = _mm_mul_pd(iq1,jq0);
726 /* Calculate table index by multiplying r with table scale and truncate to integer */
727 rt = _mm_mul_pd(r10,vftabscale);
728 vfitab = _mm_cvttpd_epi32(rt);
730 vfeps = _mm_frcz_pd(rt);
732 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
734 twovfeps = _mm_add_pd(vfeps,vfeps);
735 vfitab = _mm_slli_epi32(vfitab,2);
737 /* CUBIC SPLINE TABLE ELECTROSTATICS */
738 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
739 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
740 GMX_MM_TRANSPOSE2_PD(Y,F);
741 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
742 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
743 GMX_MM_TRANSPOSE2_PD(G,H);
744 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
745 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
746 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
750 /* Update vectorial force */
751 fix1 = _mm_macc_pd(dx10,fscal,fix1);
752 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
753 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
755 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
756 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
757 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
759 /**************************
760 * CALCULATE INTERACTIONS *
761 **************************/
763 r20 = _mm_mul_pd(rsq20,rinv20);
765 /* Compute parameters for interactions between i and j atoms */
766 qq20 = _mm_mul_pd(iq2,jq0);
768 /* Calculate table index by multiplying r with table scale and truncate to integer */
769 rt = _mm_mul_pd(r20,vftabscale);
770 vfitab = _mm_cvttpd_epi32(rt);
772 vfeps = _mm_frcz_pd(rt);
774 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
776 twovfeps = _mm_add_pd(vfeps,vfeps);
777 vfitab = _mm_slli_epi32(vfitab,2);
779 /* CUBIC SPLINE TABLE ELECTROSTATICS */
780 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
781 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
782 GMX_MM_TRANSPOSE2_PD(Y,F);
783 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
784 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
785 GMX_MM_TRANSPOSE2_PD(G,H);
786 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
787 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
788 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
792 /* Update vectorial force */
793 fix2 = _mm_macc_pd(dx20,fscal,fix2);
794 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
795 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
797 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
798 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
799 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
801 /**************************
802 * CALCULATE INTERACTIONS *
803 **************************/
805 r30 = _mm_mul_pd(rsq30,rinv30);
807 /* Compute parameters for interactions between i and j atoms */
808 qq30 = _mm_mul_pd(iq3,jq0);
810 /* Calculate table index by multiplying r with table scale and truncate to integer */
811 rt = _mm_mul_pd(r30,vftabscale);
812 vfitab = _mm_cvttpd_epi32(rt);
814 vfeps = _mm_frcz_pd(rt);
816 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
818 twovfeps = _mm_add_pd(vfeps,vfeps);
819 vfitab = _mm_slli_epi32(vfitab,2);
821 /* CUBIC SPLINE TABLE ELECTROSTATICS */
822 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
823 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
824 GMX_MM_TRANSPOSE2_PD(Y,F);
825 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
826 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
827 GMX_MM_TRANSPOSE2_PD(G,H);
828 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
829 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
830 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
834 /* Update vectorial force */
835 fix3 = _mm_macc_pd(dx30,fscal,fix3);
836 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
837 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
839 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
840 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
841 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
843 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
845 /* Inner loop uses 129 flops */
852 j_coord_offsetA = DIM*jnrA;
854 /* load j atom coordinates */
855 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
858 /* Calculate displacement vector */
859 dx10 = _mm_sub_pd(ix1,jx0);
860 dy10 = _mm_sub_pd(iy1,jy0);
861 dz10 = _mm_sub_pd(iz1,jz0);
862 dx20 = _mm_sub_pd(ix2,jx0);
863 dy20 = _mm_sub_pd(iy2,jy0);
864 dz20 = _mm_sub_pd(iz2,jz0);
865 dx30 = _mm_sub_pd(ix3,jx0);
866 dy30 = _mm_sub_pd(iy3,jy0);
867 dz30 = _mm_sub_pd(iz3,jz0);
869 /* Calculate squared distance and things based on it */
870 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
871 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
872 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
874 rinv10 = gmx_mm_invsqrt_pd(rsq10);
875 rinv20 = gmx_mm_invsqrt_pd(rsq20);
876 rinv30 = gmx_mm_invsqrt_pd(rsq30);
878 /* Load parameters for j particles */
879 jq0 = _mm_load_sd(charge+jnrA+0);
881 fjx0 = _mm_setzero_pd();
882 fjy0 = _mm_setzero_pd();
883 fjz0 = _mm_setzero_pd();
885 /**************************
886 * CALCULATE INTERACTIONS *
887 **************************/
889 r10 = _mm_mul_pd(rsq10,rinv10);
891 /* Compute parameters for interactions between i and j atoms */
892 qq10 = _mm_mul_pd(iq1,jq0);
894 /* Calculate table index by multiplying r with table scale and truncate to integer */
895 rt = _mm_mul_pd(r10,vftabscale);
896 vfitab = _mm_cvttpd_epi32(rt);
898 vfeps = _mm_frcz_pd(rt);
900 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
902 twovfeps = _mm_add_pd(vfeps,vfeps);
903 vfitab = _mm_slli_epi32(vfitab,2);
905 /* CUBIC SPLINE TABLE ELECTROSTATICS */
906 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
907 F = _mm_setzero_pd();
908 GMX_MM_TRANSPOSE2_PD(Y,F);
909 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
910 H = _mm_setzero_pd();
911 GMX_MM_TRANSPOSE2_PD(G,H);
912 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
913 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
914 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
918 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
920 /* Update vectorial force */
921 fix1 = _mm_macc_pd(dx10,fscal,fix1);
922 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
923 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
925 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
926 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
927 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
929 /**************************
930 * CALCULATE INTERACTIONS *
931 **************************/
933 r20 = _mm_mul_pd(rsq20,rinv20);
935 /* Compute parameters for interactions between i and j atoms */
936 qq20 = _mm_mul_pd(iq2,jq0);
938 /* Calculate table index by multiplying r with table scale and truncate to integer */
939 rt = _mm_mul_pd(r20,vftabscale);
940 vfitab = _mm_cvttpd_epi32(rt);
942 vfeps = _mm_frcz_pd(rt);
944 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
946 twovfeps = _mm_add_pd(vfeps,vfeps);
947 vfitab = _mm_slli_epi32(vfitab,2);
949 /* CUBIC SPLINE TABLE ELECTROSTATICS */
950 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
951 F = _mm_setzero_pd();
952 GMX_MM_TRANSPOSE2_PD(Y,F);
953 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
954 H = _mm_setzero_pd();
955 GMX_MM_TRANSPOSE2_PD(G,H);
956 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
957 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
958 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
962 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
964 /* Update vectorial force */
965 fix2 = _mm_macc_pd(dx20,fscal,fix2);
966 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
967 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
969 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
970 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
971 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
973 /**************************
974 * CALCULATE INTERACTIONS *
975 **************************/
977 r30 = _mm_mul_pd(rsq30,rinv30);
979 /* Compute parameters for interactions between i and j atoms */
980 qq30 = _mm_mul_pd(iq3,jq0);
982 /* Calculate table index by multiplying r with table scale and truncate to integer */
983 rt = _mm_mul_pd(r30,vftabscale);
984 vfitab = _mm_cvttpd_epi32(rt);
986 vfeps = _mm_frcz_pd(rt);
988 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
990 twovfeps = _mm_add_pd(vfeps,vfeps);
991 vfitab = _mm_slli_epi32(vfitab,2);
993 /* CUBIC SPLINE TABLE ELECTROSTATICS */
994 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
995 F = _mm_setzero_pd();
996 GMX_MM_TRANSPOSE2_PD(Y,F);
997 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
998 H = _mm_setzero_pd();
999 GMX_MM_TRANSPOSE2_PD(G,H);
1000 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1001 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1002 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1006 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1008 /* Update vectorial force */
1009 fix3 = _mm_macc_pd(dx30,fscal,fix3);
1010 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
1011 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
1013 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
1014 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
1015 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
1017 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1019 /* Inner loop uses 129 flops */
1022 /* End of innermost loop */
1024 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1025 f+i_coord_offset+DIM,fshift+i_shift_offset);
1027 /* Increment number of inner iterations */
1028 inneriter += j_index_end - j_index_start;
1030 /* Outer loop uses 18 flops */
1033 /* Increment number of outer iterations */
1036 /* Update outer/inner flops */
1038 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*129);