2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: CubicSplineTable
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
86 int vdwjidx0A,vdwjidx0B;
87 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
88 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
89 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
90 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
91 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
94 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
97 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
98 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
100 __m128i ifour = _mm_set1_epi32(4);
101 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
103 __m128d dummy_mask,cutoff_mask;
104 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
105 __m128d one = _mm_set1_pd(1.0);
106 __m128d two = _mm_set1_pd(2.0);
112 jindex = nlist->jindex;
114 shiftidx = nlist->shift;
116 shiftvec = fr->shift_vec[0];
117 fshift = fr->fshift[0];
118 facel = _mm_set1_pd(fr->epsfac);
119 charge = mdatoms->chargeA;
120 krf = _mm_set1_pd(fr->ic->k_rf);
121 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
122 crf = _mm_set1_pd(fr->ic->c_rf);
123 nvdwtype = fr->ntype;
125 vdwtype = mdatoms->typeA;
127 vftab = kernel_data->table_vdw->data;
128 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
130 /* Setup water-specific parameters */
131 inr = nlist->iinr[0];
132 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
133 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
134 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
135 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
137 /* Avoid stupid compiler warnings */
145 /* Start outer loop over neighborlists */
146 for(iidx=0; iidx<nri; iidx++)
148 /* Load shift vector for this list */
149 i_shift_offset = DIM*shiftidx[iidx];
151 /* Load limits for loop over neighbors */
152 j_index_start = jindex[iidx];
153 j_index_end = jindex[iidx+1];
155 /* Get outer coordinate index */
157 i_coord_offset = DIM*inr;
159 /* Load i particle coords and add shift vector */
160 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
161 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
163 fix0 = _mm_setzero_pd();
164 fiy0 = _mm_setzero_pd();
165 fiz0 = _mm_setzero_pd();
166 fix1 = _mm_setzero_pd();
167 fiy1 = _mm_setzero_pd();
168 fiz1 = _mm_setzero_pd();
169 fix2 = _mm_setzero_pd();
170 fiy2 = _mm_setzero_pd();
171 fiz2 = _mm_setzero_pd();
173 /* Reset potential sums */
174 velecsum = _mm_setzero_pd();
175 vvdwsum = _mm_setzero_pd();
177 /* Start inner kernel loop */
178 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
181 /* Get j neighbor index, and coordinate index */
184 j_coord_offsetA = DIM*jnrA;
185 j_coord_offsetB = DIM*jnrB;
187 /* load j atom coordinates */
188 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
191 /* Calculate displacement vector */
192 dx00 = _mm_sub_pd(ix0,jx0);
193 dy00 = _mm_sub_pd(iy0,jy0);
194 dz00 = _mm_sub_pd(iz0,jz0);
195 dx10 = _mm_sub_pd(ix1,jx0);
196 dy10 = _mm_sub_pd(iy1,jy0);
197 dz10 = _mm_sub_pd(iz1,jz0);
198 dx20 = _mm_sub_pd(ix2,jx0);
199 dy20 = _mm_sub_pd(iy2,jy0);
200 dz20 = _mm_sub_pd(iz2,jz0);
202 /* Calculate squared distance and things based on it */
203 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
204 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
205 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
207 rinv00 = gmx_mm_invsqrt_pd(rsq00);
208 rinv10 = gmx_mm_invsqrt_pd(rsq10);
209 rinv20 = gmx_mm_invsqrt_pd(rsq20);
211 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
212 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
213 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
215 /* Load parameters for j particles */
216 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
217 vdwjidx0A = 2*vdwtype[jnrA+0];
218 vdwjidx0B = 2*vdwtype[jnrB+0];
220 fjx0 = _mm_setzero_pd();
221 fjy0 = _mm_setzero_pd();
222 fjz0 = _mm_setzero_pd();
224 /**************************
225 * CALCULATE INTERACTIONS *
226 **************************/
228 r00 = _mm_mul_pd(rsq00,rinv00);
230 /* Compute parameters for interactions between i and j atoms */
231 qq00 = _mm_mul_pd(iq0,jq0);
232 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
233 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
235 /* Calculate table index by multiplying r with table scale and truncate to integer */
236 rt = _mm_mul_pd(r00,vftabscale);
237 vfitab = _mm_cvttpd_epi32(rt);
239 vfeps = _mm_frcz_pd(rt);
241 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
243 twovfeps = _mm_add_pd(vfeps,vfeps);
244 vfitab = _mm_slli_epi32(vfitab,3);
246 /* REACTION-FIELD ELECTROSTATICS */
247 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
248 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
250 /* CUBIC SPLINE TABLE DISPERSION */
251 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
252 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
253 GMX_MM_TRANSPOSE2_PD(Y,F);
254 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
255 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
256 GMX_MM_TRANSPOSE2_PD(G,H);
257 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
258 VV = _mm_macc_pd(vfeps,Fp,Y);
259 vvdw6 = _mm_mul_pd(c6_00,VV);
260 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
261 fvdw6 = _mm_mul_pd(c6_00,FF);
263 /* CUBIC SPLINE TABLE REPULSION */
264 vfitab = _mm_add_epi32(vfitab,ifour);
265 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
266 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
267 GMX_MM_TRANSPOSE2_PD(Y,F);
268 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
269 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
270 GMX_MM_TRANSPOSE2_PD(G,H);
271 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
272 VV = _mm_macc_pd(vfeps,Fp,Y);
273 vvdw12 = _mm_mul_pd(c12_00,VV);
274 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
275 fvdw12 = _mm_mul_pd(c12_00,FF);
276 vvdw = _mm_add_pd(vvdw12,vvdw6);
277 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
279 /* Update potential sum for this i atom from the interaction with this j atom. */
280 velecsum = _mm_add_pd(velecsum,velec);
281 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
283 fscal = _mm_add_pd(felec,fvdw);
285 /* Update vectorial force */
286 fix0 = _mm_macc_pd(dx00,fscal,fix0);
287 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
288 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
290 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
291 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
292 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
294 /**************************
295 * CALCULATE INTERACTIONS *
296 **************************/
298 /* Compute parameters for interactions between i and j atoms */
299 qq10 = _mm_mul_pd(iq1,jq0);
301 /* REACTION-FIELD ELECTROSTATICS */
302 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
303 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
305 /* Update potential sum for this i atom from the interaction with this j atom. */
306 velecsum = _mm_add_pd(velecsum,velec);
310 /* Update vectorial force */
311 fix1 = _mm_macc_pd(dx10,fscal,fix1);
312 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
313 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
315 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
316 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
317 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
319 /**************************
320 * CALCULATE INTERACTIONS *
321 **************************/
323 /* Compute parameters for interactions between i and j atoms */
324 qq20 = _mm_mul_pd(iq2,jq0);
326 /* REACTION-FIELD ELECTROSTATICS */
327 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
328 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
330 /* Update potential sum for this i atom from the interaction with this j atom. */
331 velecsum = _mm_add_pd(velecsum,velec);
335 /* Update vectorial force */
336 fix2 = _mm_macc_pd(dx20,fscal,fix2);
337 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
338 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
340 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
341 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
342 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
344 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
346 /* Inner loop uses 143 flops */
353 j_coord_offsetA = DIM*jnrA;
355 /* load j atom coordinates */
356 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
359 /* Calculate displacement vector */
360 dx00 = _mm_sub_pd(ix0,jx0);
361 dy00 = _mm_sub_pd(iy0,jy0);
362 dz00 = _mm_sub_pd(iz0,jz0);
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);
370 /* Calculate squared distance and things based on it */
371 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
372 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
373 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
375 rinv00 = gmx_mm_invsqrt_pd(rsq00);
376 rinv10 = gmx_mm_invsqrt_pd(rsq10);
377 rinv20 = gmx_mm_invsqrt_pd(rsq20);
379 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
380 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
381 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
383 /* Load parameters for j particles */
384 jq0 = _mm_load_sd(charge+jnrA+0);
385 vdwjidx0A = 2*vdwtype[jnrA+0];
387 fjx0 = _mm_setzero_pd();
388 fjy0 = _mm_setzero_pd();
389 fjz0 = _mm_setzero_pd();
391 /**************************
392 * CALCULATE INTERACTIONS *
393 **************************/
395 r00 = _mm_mul_pd(rsq00,rinv00);
397 /* Compute parameters for interactions between i and j atoms */
398 qq00 = _mm_mul_pd(iq0,jq0);
399 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
401 /* Calculate table index by multiplying r with table scale and truncate to integer */
402 rt = _mm_mul_pd(r00,vftabscale);
403 vfitab = _mm_cvttpd_epi32(rt);
405 vfeps = _mm_frcz_pd(rt);
407 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
409 twovfeps = _mm_add_pd(vfeps,vfeps);
410 vfitab = _mm_slli_epi32(vfitab,3);
412 /* REACTION-FIELD ELECTROSTATICS */
413 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
414 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
416 /* CUBIC SPLINE TABLE DISPERSION */
417 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
418 F = _mm_setzero_pd();
419 GMX_MM_TRANSPOSE2_PD(Y,F);
420 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
421 H = _mm_setzero_pd();
422 GMX_MM_TRANSPOSE2_PD(G,H);
423 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
424 VV = _mm_macc_pd(vfeps,Fp,Y);
425 vvdw6 = _mm_mul_pd(c6_00,VV);
426 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
427 fvdw6 = _mm_mul_pd(c6_00,FF);
429 /* CUBIC SPLINE TABLE REPULSION */
430 vfitab = _mm_add_epi32(vfitab,ifour);
431 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
432 F = _mm_setzero_pd();
433 GMX_MM_TRANSPOSE2_PD(Y,F);
434 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
435 H = _mm_setzero_pd();
436 GMX_MM_TRANSPOSE2_PD(G,H);
437 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
438 VV = _mm_macc_pd(vfeps,Fp,Y);
439 vvdw12 = _mm_mul_pd(c12_00,VV);
440 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
441 fvdw12 = _mm_mul_pd(c12_00,FF);
442 vvdw = _mm_add_pd(vvdw12,vvdw6);
443 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
445 /* Update potential sum for this i atom from the interaction with this j atom. */
446 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
447 velecsum = _mm_add_pd(velecsum,velec);
448 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
449 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
451 fscal = _mm_add_pd(felec,fvdw);
453 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
455 /* Update vectorial force */
456 fix0 = _mm_macc_pd(dx00,fscal,fix0);
457 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
458 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
460 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
461 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
462 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
464 /**************************
465 * CALCULATE INTERACTIONS *
466 **************************/
468 /* Compute parameters for interactions between i and j atoms */
469 qq10 = _mm_mul_pd(iq1,jq0);
471 /* REACTION-FIELD ELECTROSTATICS */
472 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
473 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
475 /* Update potential sum for this i atom from the interaction with this j atom. */
476 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
477 velecsum = _mm_add_pd(velecsum,velec);
481 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
483 /* Update vectorial force */
484 fix1 = _mm_macc_pd(dx10,fscal,fix1);
485 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
486 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
488 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
489 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
490 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
492 /**************************
493 * CALCULATE INTERACTIONS *
494 **************************/
496 /* Compute parameters for interactions between i and j atoms */
497 qq20 = _mm_mul_pd(iq2,jq0);
499 /* REACTION-FIELD ELECTROSTATICS */
500 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
501 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
503 /* Update potential sum for this i atom from the interaction with this j atom. */
504 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
505 velecsum = _mm_add_pd(velecsum,velec);
509 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
511 /* Update vectorial force */
512 fix2 = _mm_macc_pd(dx20,fscal,fix2);
513 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
514 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
516 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
517 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
518 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
520 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
522 /* Inner loop uses 143 flops */
525 /* End of innermost loop */
527 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
528 f+i_coord_offset,fshift+i_shift_offset);
531 /* Update potential energies */
532 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
533 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
535 /* Increment number of inner iterations */
536 inneriter += j_index_end - j_index_start;
538 /* Outer loop uses 20 flops */
541 /* Increment number of outer iterations */
544 /* Update outer/inner flops */
546 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*143);
549 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_double
550 * Electrostatics interaction: ReactionField
551 * VdW interaction: CubicSplineTable
552 * Geometry: Water3-Particle
553 * Calculate force/pot: Force
556 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_double
557 (t_nblist * gmx_restrict nlist,
558 rvec * gmx_restrict xx,
559 rvec * gmx_restrict ff,
560 t_forcerec * gmx_restrict fr,
561 t_mdatoms * gmx_restrict mdatoms,
562 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
563 t_nrnb * gmx_restrict nrnb)
565 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
566 * just 0 for non-waters.
567 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
568 * jnr indices corresponding to data put in the four positions in the SIMD register.
570 int i_shift_offset,i_coord_offset,outeriter,inneriter;
571 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
573 int j_coord_offsetA,j_coord_offsetB;
574 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
576 real *shiftvec,*fshift,*x,*f;
577 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
579 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
581 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
583 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
584 int vdwjidx0A,vdwjidx0B;
585 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
586 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
587 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
588 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
589 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
592 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
595 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
596 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
598 __m128i ifour = _mm_set1_epi32(4);
599 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
601 __m128d dummy_mask,cutoff_mask;
602 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
603 __m128d one = _mm_set1_pd(1.0);
604 __m128d two = _mm_set1_pd(2.0);
610 jindex = nlist->jindex;
612 shiftidx = nlist->shift;
614 shiftvec = fr->shift_vec[0];
615 fshift = fr->fshift[0];
616 facel = _mm_set1_pd(fr->epsfac);
617 charge = mdatoms->chargeA;
618 krf = _mm_set1_pd(fr->ic->k_rf);
619 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
620 crf = _mm_set1_pd(fr->ic->c_rf);
621 nvdwtype = fr->ntype;
623 vdwtype = mdatoms->typeA;
625 vftab = kernel_data->table_vdw->data;
626 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
628 /* Setup water-specific parameters */
629 inr = nlist->iinr[0];
630 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
631 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
632 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
633 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
635 /* Avoid stupid compiler warnings */
643 /* Start outer loop over neighborlists */
644 for(iidx=0; iidx<nri; iidx++)
646 /* Load shift vector for this list */
647 i_shift_offset = DIM*shiftidx[iidx];
649 /* Load limits for loop over neighbors */
650 j_index_start = jindex[iidx];
651 j_index_end = jindex[iidx+1];
653 /* Get outer coordinate index */
655 i_coord_offset = DIM*inr;
657 /* Load i particle coords and add shift vector */
658 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
659 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
661 fix0 = _mm_setzero_pd();
662 fiy0 = _mm_setzero_pd();
663 fiz0 = _mm_setzero_pd();
664 fix1 = _mm_setzero_pd();
665 fiy1 = _mm_setzero_pd();
666 fiz1 = _mm_setzero_pd();
667 fix2 = _mm_setzero_pd();
668 fiy2 = _mm_setzero_pd();
669 fiz2 = _mm_setzero_pd();
671 /* Start inner kernel loop */
672 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
675 /* Get j neighbor index, and coordinate index */
678 j_coord_offsetA = DIM*jnrA;
679 j_coord_offsetB = DIM*jnrB;
681 /* load j atom coordinates */
682 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
685 /* Calculate displacement vector */
686 dx00 = _mm_sub_pd(ix0,jx0);
687 dy00 = _mm_sub_pd(iy0,jy0);
688 dz00 = _mm_sub_pd(iz0,jz0);
689 dx10 = _mm_sub_pd(ix1,jx0);
690 dy10 = _mm_sub_pd(iy1,jy0);
691 dz10 = _mm_sub_pd(iz1,jz0);
692 dx20 = _mm_sub_pd(ix2,jx0);
693 dy20 = _mm_sub_pd(iy2,jy0);
694 dz20 = _mm_sub_pd(iz2,jz0);
696 /* Calculate squared distance and things based on it */
697 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
698 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
699 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
701 rinv00 = gmx_mm_invsqrt_pd(rsq00);
702 rinv10 = gmx_mm_invsqrt_pd(rsq10);
703 rinv20 = gmx_mm_invsqrt_pd(rsq20);
705 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
706 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
707 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
709 /* Load parameters for j particles */
710 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
711 vdwjidx0A = 2*vdwtype[jnrA+0];
712 vdwjidx0B = 2*vdwtype[jnrB+0];
714 fjx0 = _mm_setzero_pd();
715 fjy0 = _mm_setzero_pd();
716 fjz0 = _mm_setzero_pd();
718 /**************************
719 * CALCULATE INTERACTIONS *
720 **************************/
722 r00 = _mm_mul_pd(rsq00,rinv00);
724 /* Compute parameters for interactions between i and j atoms */
725 qq00 = _mm_mul_pd(iq0,jq0);
726 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
727 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
729 /* Calculate table index by multiplying r with table scale and truncate to integer */
730 rt = _mm_mul_pd(r00,vftabscale);
731 vfitab = _mm_cvttpd_epi32(rt);
733 vfeps = _mm_frcz_pd(rt);
735 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
737 twovfeps = _mm_add_pd(vfeps,vfeps);
738 vfitab = _mm_slli_epi32(vfitab,3);
740 /* REACTION-FIELD ELECTROSTATICS */
741 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
743 /* CUBIC SPLINE TABLE DISPERSION */
744 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
745 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
746 GMX_MM_TRANSPOSE2_PD(Y,F);
747 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
748 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
749 GMX_MM_TRANSPOSE2_PD(G,H);
750 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
751 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
752 fvdw6 = _mm_mul_pd(c6_00,FF);
754 /* CUBIC SPLINE TABLE REPULSION */
755 vfitab = _mm_add_epi32(vfitab,ifour);
756 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
757 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
758 GMX_MM_TRANSPOSE2_PD(Y,F);
759 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
760 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
761 GMX_MM_TRANSPOSE2_PD(G,H);
762 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
763 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
764 fvdw12 = _mm_mul_pd(c12_00,FF);
765 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
767 fscal = _mm_add_pd(felec,fvdw);
769 /* Update vectorial force */
770 fix0 = _mm_macc_pd(dx00,fscal,fix0);
771 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
772 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
774 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
775 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
776 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
778 /**************************
779 * CALCULATE INTERACTIONS *
780 **************************/
782 /* Compute parameters for interactions between i and j atoms */
783 qq10 = _mm_mul_pd(iq1,jq0);
785 /* REACTION-FIELD ELECTROSTATICS */
786 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
790 /* Update vectorial force */
791 fix1 = _mm_macc_pd(dx10,fscal,fix1);
792 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
793 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
795 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
796 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
797 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
799 /**************************
800 * CALCULATE INTERACTIONS *
801 **************************/
803 /* Compute parameters for interactions between i and j atoms */
804 qq20 = _mm_mul_pd(iq2,jq0);
806 /* REACTION-FIELD ELECTROSTATICS */
807 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
811 /* Update vectorial force */
812 fix2 = _mm_macc_pd(dx20,fscal,fix2);
813 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
814 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
816 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
817 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
818 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
820 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
822 /* Inner loop uses 120 flops */
829 j_coord_offsetA = DIM*jnrA;
831 /* load j atom coordinates */
832 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
835 /* Calculate displacement vector */
836 dx00 = _mm_sub_pd(ix0,jx0);
837 dy00 = _mm_sub_pd(iy0,jy0);
838 dz00 = _mm_sub_pd(iz0,jz0);
839 dx10 = _mm_sub_pd(ix1,jx0);
840 dy10 = _mm_sub_pd(iy1,jy0);
841 dz10 = _mm_sub_pd(iz1,jz0);
842 dx20 = _mm_sub_pd(ix2,jx0);
843 dy20 = _mm_sub_pd(iy2,jy0);
844 dz20 = _mm_sub_pd(iz2,jz0);
846 /* Calculate squared distance and things based on it */
847 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
848 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
849 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
851 rinv00 = gmx_mm_invsqrt_pd(rsq00);
852 rinv10 = gmx_mm_invsqrt_pd(rsq10);
853 rinv20 = gmx_mm_invsqrt_pd(rsq20);
855 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
856 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
857 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
859 /* Load parameters for j particles */
860 jq0 = _mm_load_sd(charge+jnrA+0);
861 vdwjidx0A = 2*vdwtype[jnrA+0];
863 fjx0 = _mm_setzero_pd();
864 fjy0 = _mm_setzero_pd();
865 fjz0 = _mm_setzero_pd();
867 /**************************
868 * CALCULATE INTERACTIONS *
869 **************************/
871 r00 = _mm_mul_pd(rsq00,rinv00);
873 /* Compute parameters for interactions between i and j atoms */
874 qq00 = _mm_mul_pd(iq0,jq0);
875 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
877 /* Calculate table index by multiplying r with table scale and truncate to integer */
878 rt = _mm_mul_pd(r00,vftabscale);
879 vfitab = _mm_cvttpd_epi32(rt);
881 vfeps = _mm_frcz_pd(rt);
883 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
885 twovfeps = _mm_add_pd(vfeps,vfeps);
886 vfitab = _mm_slli_epi32(vfitab,3);
888 /* REACTION-FIELD ELECTROSTATICS */
889 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
891 /* CUBIC SPLINE TABLE DISPERSION */
892 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
893 F = _mm_setzero_pd();
894 GMX_MM_TRANSPOSE2_PD(Y,F);
895 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
896 H = _mm_setzero_pd();
897 GMX_MM_TRANSPOSE2_PD(G,H);
898 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
899 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
900 fvdw6 = _mm_mul_pd(c6_00,FF);
902 /* CUBIC SPLINE TABLE REPULSION */
903 vfitab = _mm_add_epi32(vfitab,ifour);
904 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
905 F = _mm_setzero_pd();
906 GMX_MM_TRANSPOSE2_PD(Y,F);
907 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
908 H = _mm_setzero_pd();
909 GMX_MM_TRANSPOSE2_PD(G,H);
910 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
911 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
912 fvdw12 = _mm_mul_pd(c12_00,FF);
913 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
915 fscal = _mm_add_pd(felec,fvdw);
917 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
919 /* Update vectorial force */
920 fix0 = _mm_macc_pd(dx00,fscal,fix0);
921 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
922 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
924 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
925 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
926 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
928 /**************************
929 * CALCULATE INTERACTIONS *
930 **************************/
932 /* Compute parameters for interactions between i and j atoms */
933 qq10 = _mm_mul_pd(iq1,jq0);
935 /* REACTION-FIELD ELECTROSTATICS */
936 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
940 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
942 /* Update vectorial force */
943 fix1 = _mm_macc_pd(dx10,fscal,fix1);
944 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
945 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
947 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
948 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
949 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
955 /* Compute parameters for interactions between i and j atoms */
956 qq20 = _mm_mul_pd(iq2,jq0);
958 /* REACTION-FIELD ELECTROSTATICS */
959 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
963 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
965 /* Update vectorial force */
966 fix2 = _mm_macc_pd(dx20,fscal,fix2);
967 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
968 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
970 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
971 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
972 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
974 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
976 /* Inner loop uses 120 flops */
979 /* End of innermost loop */
981 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
982 f+i_coord_offset,fshift+i_shift_offset);
984 /* Increment number of inner iterations */
985 inneriter += j_index_end - j_index_start;
987 /* Outer loop uses 18 flops */
990 /* Increment number of outer iterations */
993 /* Update outer/inner flops */
995 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*120);