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_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
52 * Electrostatics interaction: Coulomb
53 * VdW interaction: CubicSplineTable
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCoul_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 nvdwtype = fr->ntype;
122 vdwtype = mdatoms->typeA;
124 vftab = kernel_data->table_vdw->data;
125 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
127 /* Setup water-specific parameters */
128 inr = nlist->iinr[0];
129 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
130 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
131 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
132 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
134 /* Avoid stupid compiler warnings */
142 /* Start outer loop over neighborlists */
143 for(iidx=0; iidx<nri; iidx++)
145 /* Load shift vector for this list */
146 i_shift_offset = DIM*shiftidx[iidx];
148 /* Load limits for loop over neighbors */
149 j_index_start = jindex[iidx];
150 j_index_end = jindex[iidx+1];
152 /* Get outer coordinate index */
154 i_coord_offset = DIM*inr;
156 /* Load i particle coords and add shift vector */
157 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
158 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
160 fix0 = _mm_setzero_pd();
161 fiy0 = _mm_setzero_pd();
162 fiz0 = _mm_setzero_pd();
163 fix1 = _mm_setzero_pd();
164 fiy1 = _mm_setzero_pd();
165 fiz1 = _mm_setzero_pd();
166 fix2 = _mm_setzero_pd();
167 fiy2 = _mm_setzero_pd();
168 fiz2 = _mm_setzero_pd();
170 /* Reset potential sums */
171 velecsum = _mm_setzero_pd();
172 vvdwsum = _mm_setzero_pd();
174 /* Start inner kernel loop */
175 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
178 /* Get j neighbor index, and coordinate index */
181 j_coord_offsetA = DIM*jnrA;
182 j_coord_offsetB = DIM*jnrB;
184 /* load j atom coordinates */
185 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
188 /* Calculate displacement vector */
189 dx00 = _mm_sub_pd(ix0,jx0);
190 dy00 = _mm_sub_pd(iy0,jy0);
191 dz00 = _mm_sub_pd(iz0,jz0);
192 dx10 = _mm_sub_pd(ix1,jx0);
193 dy10 = _mm_sub_pd(iy1,jy0);
194 dz10 = _mm_sub_pd(iz1,jz0);
195 dx20 = _mm_sub_pd(ix2,jx0);
196 dy20 = _mm_sub_pd(iy2,jy0);
197 dz20 = _mm_sub_pd(iz2,jz0);
199 /* Calculate squared distance and things based on it */
200 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
201 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
202 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
204 rinv00 = gmx_mm_invsqrt_pd(rsq00);
205 rinv10 = gmx_mm_invsqrt_pd(rsq10);
206 rinv20 = gmx_mm_invsqrt_pd(rsq20);
208 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
209 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
210 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
212 /* Load parameters for j particles */
213 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
214 vdwjidx0A = 2*vdwtype[jnrA+0];
215 vdwjidx0B = 2*vdwtype[jnrB+0];
217 fjx0 = _mm_setzero_pd();
218 fjy0 = _mm_setzero_pd();
219 fjz0 = _mm_setzero_pd();
221 /**************************
222 * CALCULATE INTERACTIONS *
223 **************************/
225 r00 = _mm_mul_pd(rsq00,rinv00);
227 /* Compute parameters for interactions between i and j atoms */
228 qq00 = _mm_mul_pd(iq0,jq0);
229 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
230 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
232 /* Calculate table index by multiplying r with table scale and truncate to integer */
233 rt = _mm_mul_pd(r00,vftabscale);
234 vfitab = _mm_cvttpd_epi32(rt);
236 vfeps = _mm_frcz_pd(rt);
238 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
240 twovfeps = _mm_add_pd(vfeps,vfeps);
241 vfitab = _mm_slli_epi32(vfitab,3);
243 /* COULOMB ELECTROSTATICS */
244 velec = _mm_mul_pd(qq00,rinv00);
245 felec = _mm_mul_pd(velec,rinvsq00);
247 /* CUBIC SPLINE TABLE DISPERSION */
248 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
249 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
250 GMX_MM_TRANSPOSE2_PD(Y,F);
251 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
252 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
253 GMX_MM_TRANSPOSE2_PD(G,H);
254 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
255 VV = _mm_macc_pd(vfeps,Fp,Y);
256 vvdw6 = _mm_mul_pd(c6_00,VV);
257 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
258 fvdw6 = _mm_mul_pd(c6_00,FF);
260 /* CUBIC SPLINE TABLE REPULSION */
261 vfitab = _mm_add_epi32(vfitab,ifour);
262 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
263 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
264 GMX_MM_TRANSPOSE2_PD(Y,F);
265 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
266 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
267 GMX_MM_TRANSPOSE2_PD(G,H);
268 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
269 VV = _mm_macc_pd(vfeps,Fp,Y);
270 vvdw12 = _mm_mul_pd(c12_00,VV);
271 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
272 fvdw12 = _mm_mul_pd(c12_00,FF);
273 vvdw = _mm_add_pd(vvdw12,vvdw6);
274 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
276 /* Update potential sum for this i atom from the interaction with this j atom. */
277 velecsum = _mm_add_pd(velecsum,velec);
278 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
280 fscal = _mm_add_pd(felec,fvdw);
282 /* Update vectorial force */
283 fix0 = _mm_macc_pd(dx00,fscal,fix0);
284 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
285 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
287 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
288 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
289 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
291 /**************************
292 * CALCULATE INTERACTIONS *
293 **************************/
295 /* Compute parameters for interactions between i and j atoms */
296 qq10 = _mm_mul_pd(iq1,jq0);
298 /* COULOMB ELECTROSTATICS */
299 velec = _mm_mul_pd(qq10,rinv10);
300 felec = _mm_mul_pd(velec,rinvsq10);
302 /* Update potential sum for this i atom from the interaction with this j atom. */
303 velecsum = _mm_add_pd(velecsum,velec);
307 /* Update vectorial force */
308 fix1 = _mm_macc_pd(dx10,fscal,fix1);
309 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
310 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
312 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
313 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
314 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
316 /**************************
317 * CALCULATE INTERACTIONS *
318 **************************/
320 /* Compute parameters for interactions between i and j atoms */
321 qq20 = _mm_mul_pd(iq2,jq0);
323 /* COULOMB ELECTROSTATICS */
324 velec = _mm_mul_pd(qq20,rinv20);
325 felec = _mm_mul_pd(velec,rinvsq20);
327 /* Update potential sum for this i atom from the interaction with this j atom. */
328 velecsum = _mm_add_pd(velecsum,velec);
332 /* Update vectorial force */
333 fix2 = _mm_macc_pd(dx20,fscal,fix2);
334 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
335 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
337 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
338 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
339 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
341 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
343 /* Inner loop uses 131 flops */
350 j_coord_offsetA = DIM*jnrA;
352 /* load j atom coordinates */
353 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
356 /* Calculate displacement vector */
357 dx00 = _mm_sub_pd(ix0,jx0);
358 dy00 = _mm_sub_pd(iy0,jy0);
359 dz00 = _mm_sub_pd(iz0,jz0);
360 dx10 = _mm_sub_pd(ix1,jx0);
361 dy10 = _mm_sub_pd(iy1,jy0);
362 dz10 = _mm_sub_pd(iz1,jz0);
363 dx20 = _mm_sub_pd(ix2,jx0);
364 dy20 = _mm_sub_pd(iy2,jy0);
365 dz20 = _mm_sub_pd(iz2,jz0);
367 /* Calculate squared distance and things based on it */
368 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
369 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
370 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
372 rinv00 = gmx_mm_invsqrt_pd(rsq00);
373 rinv10 = gmx_mm_invsqrt_pd(rsq10);
374 rinv20 = gmx_mm_invsqrt_pd(rsq20);
376 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
377 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
378 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
380 /* Load parameters for j particles */
381 jq0 = _mm_load_sd(charge+jnrA+0);
382 vdwjidx0A = 2*vdwtype[jnrA+0];
384 fjx0 = _mm_setzero_pd();
385 fjy0 = _mm_setzero_pd();
386 fjz0 = _mm_setzero_pd();
388 /**************************
389 * CALCULATE INTERACTIONS *
390 **************************/
392 r00 = _mm_mul_pd(rsq00,rinv00);
394 /* Compute parameters for interactions between i and j atoms */
395 qq00 = _mm_mul_pd(iq0,jq0);
396 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
398 /* Calculate table index by multiplying r with table scale and truncate to integer */
399 rt = _mm_mul_pd(r00,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,3);
409 /* COULOMB ELECTROSTATICS */
410 velec = _mm_mul_pd(qq00,rinv00);
411 felec = _mm_mul_pd(velec,rinvsq00);
413 /* CUBIC SPLINE TABLE DISPERSION */
414 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
415 F = _mm_setzero_pd();
416 GMX_MM_TRANSPOSE2_PD(Y,F);
417 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
418 H = _mm_setzero_pd();
419 GMX_MM_TRANSPOSE2_PD(G,H);
420 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
421 VV = _mm_macc_pd(vfeps,Fp,Y);
422 vvdw6 = _mm_mul_pd(c6_00,VV);
423 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
424 fvdw6 = _mm_mul_pd(c6_00,FF);
426 /* CUBIC SPLINE TABLE REPULSION */
427 vfitab = _mm_add_epi32(vfitab,ifour);
428 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
429 F = _mm_setzero_pd();
430 GMX_MM_TRANSPOSE2_PD(Y,F);
431 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
432 H = _mm_setzero_pd();
433 GMX_MM_TRANSPOSE2_PD(G,H);
434 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
435 VV = _mm_macc_pd(vfeps,Fp,Y);
436 vvdw12 = _mm_mul_pd(c12_00,VV);
437 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
438 fvdw12 = _mm_mul_pd(c12_00,FF);
439 vvdw = _mm_add_pd(vvdw12,vvdw6);
440 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
442 /* Update potential sum for this i atom from the interaction with this j atom. */
443 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
444 velecsum = _mm_add_pd(velecsum,velec);
445 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
446 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
448 fscal = _mm_add_pd(felec,fvdw);
450 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
452 /* Update vectorial force */
453 fix0 = _mm_macc_pd(dx00,fscal,fix0);
454 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
455 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
457 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
458 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
459 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
461 /**************************
462 * CALCULATE INTERACTIONS *
463 **************************/
465 /* Compute parameters for interactions between i and j atoms */
466 qq10 = _mm_mul_pd(iq1,jq0);
468 /* COULOMB ELECTROSTATICS */
469 velec = _mm_mul_pd(qq10,rinv10);
470 felec = _mm_mul_pd(velec,rinvsq10);
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 fix1 = _mm_macc_pd(dx10,fscal,fix1);
482 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
483 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
485 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
486 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
487 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
489 /**************************
490 * CALCULATE INTERACTIONS *
491 **************************/
493 /* Compute parameters for interactions between i and j atoms */
494 qq20 = _mm_mul_pd(iq2,jq0);
496 /* COULOMB ELECTROSTATICS */
497 velec = _mm_mul_pd(qq20,rinv20);
498 felec = _mm_mul_pd(velec,rinvsq20);
500 /* Update potential sum for this i atom from the interaction with this j atom. */
501 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
502 velecsum = _mm_add_pd(velecsum,velec);
506 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
508 /* Update vectorial force */
509 fix2 = _mm_macc_pd(dx20,fscal,fix2);
510 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
511 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
513 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
514 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
515 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
517 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
519 /* Inner loop uses 131 flops */
522 /* End of innermost loop */
524 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
525 f+i_coord_offset,fshift+i_shift_offset);
528 /* Update potential energies */
529 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
530 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
532 /* Increment number of inner iterations */
533 inneriter += j_index_end - j_index_start;
535 /* Outer loop uses 20 flops */
538 /* Increment number of outer iterations */
541 /* Update outer/inner flops */
543 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*131);
546 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_double
547 * Electrostatics interaction: Coulomb
548 * VdW interaction: CubicSplineTable
549 * Geometry: Water3-Particle
550 * Calculate force/pot: Force
553 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_double
554 (t_nblist * gmx_restrict nlist,
555 rvec * gmx_restrict xx,
556 rvec * gmx_restrict ff,
557 t_forcerec * gmx_restrict fr,
558 t_mdatoms * gmx_restrict mdatoms,
559 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
560 t_nrnb * gmx_restrict nrnb)
562 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
563 * just 0 for non-waters.
564 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
565 * jnr indices corresponding to data put in the four positions in the SIMD register.
567 int i_shift_offset,i_coord_offset,outeriter,inneriter;
568 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
570 int j_coord_offsetA,j_coord_offsetB;
571 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
573 real *shiftvec,*fshift,*x,*f;
574 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
576 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
578 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
580 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
581 int vdwjidx0A,vdwjidx0B;
582 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
583 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
584 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
585 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
586 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
589 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
592 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
593 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
595 __m128i ifour = _mm_set1_epi32(4);
596 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
598 __m128d dummy_mask,cutoff_mask;
599 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
600 __m128d one = _mm_set1_pd(1.0);
601 __m128d two = _mm_set1_pd(2.0);
607 jindex = nlist->jindex;
609 shiftidx = nlist->shift;
611 shiftvec = fr->shift_vec[0];
612 fshift = fr->fshift[0];
613 facel = _mm_set1_pd(fr->epsfac);
614 charge = mdatoms->chargeA;
615 nvdwtype = fr->ntype;
617 vdwtype = mdatoms->typeA;
619 vftab = kernel_data->table_vdw->data;
620 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
622 /* Setup water-specific parameters */
623 inr = nlist->iinr[0];
624 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
625 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
626 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
627 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
629 /* Avoid stupid compiler warnings */
637 /* Start outer loop over neighborlists */
638 for(iidx=0; iidx<nri; iidx++)
640 /* Load shift vector for this list */
641 i_shift_offset = DIM*shiftidx[iidx];
643 /* Load limits for loop over neighbors */
644 j_index_start = jindex[iidx];
645 j_index_end = jindex[iidx+1];
647 /* Get outer coordinate index */
649 i_coord_offset = DIM*inr;
651 /* Load i particle coords and add shift vector */
652 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
653 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
655 fix0 = _mm_setzero_pd();
656 fiy0 = _mm_setzero_pd();
657 fiz0 = _mm_setzero_pd();
658 fix1 = _mm_setzero_pd();
659 fiy1 = _mm_setzero_pd();
660 fiz1 = _mm_setzero_pd();
661 fix2 = _mm_setzero_pd();
662 fiy2 = _mm_setzero_pd();
663 fiz2 = _mm_setzero_pd();
665 /* Start inner kernel loop */
666 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
669 /* Get j neighbor index, and coordinate index */
672 j_coord_offsetA = DIM*jnrA;
673 j_coord_offsetB = DIM*jnrB;
675 /* load j atom coordinates */
676 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
679 /* Calculate displacement vector */
680 dx00 = _mm_sub_pd(ix0,jx0);
681 dy00 = _mm_sub_pd(iy0,jy0);
682 dz00 = _mm_sub_pd(iz0,jz0);
683 dx10 = _mm_sub_pd(ix1,jx0);
684 dy10 = _mm_sub_pd(iy1,jy0);
685 dz10 = _mm_sub_pd(iz1,jz0);
686 dx20 = _mm_sub_pd(ix2,jx0);
687 dy20 = _mm_sub_pd(iy2,jy0);
688 dz20 = _mm_sub_pd(iz2,jz0);
690 /* Calculate squared distance and things based on it */
691 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
692 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
693 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
695 rinv00 = gmx_mm_invsqrt_pd(rsq00);
696 rinv10 = gmx_mm_invsqrt_pd(rsq10);
697 rinv20 = gmx_mm_invsqrt_pd(rsq20);
699 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
700 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
701 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
703 /* Load parameters for j particles */
704 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
705 vdwjidx0A = 2*vdwtype[jnrA+0];
706 vdwjidx0B = 2*vdwtype[jnrB+0];
708 fjx0 = _mm_setzero_pd();
709 fjy0 = _mm_setzero_pd();
710 fjz0 = _mm_setzero_pd();
712 /**************************
713 * CALCULATE INTERACTIONS *
714 **************************/
716 r00 = _mm_mul_pd(rsq00,rinv00);
718 /* Compute parameters for interactions between i and j atoms */
719 qq00 = _mm_mul_pd(iq0,jq0);
720 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
721 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
723 /* Calculate table index by multiplying r with table scale and truncate to integer */
724 rt = _mm_mul_pd(r00,vftabscale);
725 vfitab = _mm_cvttpd_epi32(rt);
727 vfeps = _mm_frcz_pd(rt);
729 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
731 twovfeps = _mm_add_pd(vfeps,vfeps);
732 vfitab = _mm_slli_epi32(vfitab,3);
734 /* COULOMB ELECTROSTATICS */
735 velec = _mm_mul_pd(qq00,rinv00);
736 felec = _mm_mul_pd(velec,rinvsq00);
738 /* CUBIC SPLINE TABLE DISPERSION */
739 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
740 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
741 GMX_MM_TRANSPOSE2_PD(Y,F);
742 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
743 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
744 GMX_MM_TRANSPOSE2_PD(G,H);
745 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
746 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
747 fvdw6 = _mm_mul_pd(c6_00,FF);
749 /* CUBIC SPLINE TABLE REPULSION */
750 vfitab = _mm_add_epi32(vfitab,ifour);
751 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
752 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
753 GMX_MM_TRANSPOSE2_PD(Y,F);
754 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
755 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
756 GMX_MM_TRANSPOSE2_PD(G,H);
757 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
758 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
759 fvdw12 = _mm_mul_pd(c12_00,FF);
760 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
762 fscal = _mm_add_pd(felec,fvdw);
764 /* Update vectorial force */
765 fix0 = _mm_macc_pd(dx00,fscal,fix0);
766 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
767 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
769 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
770 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
771 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
773 /**************************
774 * CALCULATE INTERACTIONS *
775 **************************/
777 /* Compute parameters for interactions between i and j atoms */
778 qq10 = _mm_mul_pd(iq1,jq0);
780 /* COULOMB ELECTROSTATICS */
781 velec = _mm_mul_pd(qq10,rinv10);
782 felec = _mm_mul_pd(velec,rinvsq10);
786 /* Update vectorial force */
787 fix1 = _mm_macc_pd(dx10,fscal,fix1);
788 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
789 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
791 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
792 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
793 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
795 /**************************
796 * CALCULATE INTERACTIONS *
797 **************************/
799 /* Compute parameters for interactions between i and j atoms */
800 qq20 = _mm_mul_pd(iq2,jq0);
802 /* COULOMB ELECTROSTATICS */
803 velec = _mm_mul_pd(qq20,rinv20);
804 felec = _mm_mul_pd(velec,rinvsq20);
808 /* Update vectorial force */
809 fix2 = _mm_macc_pd(dx20,fscal,fix2);
810 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
811 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
813 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
814 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
815 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
817 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
819 /* Inner loop uses 120 flops */
826 j_coord_offsetA = DIM*jnrA;
828 /* load j atom coordinates */
829 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
832 /* Calculate displacement vector */
833 dx00 = _mm_sub_pd(ix0,jx0);
834 dy00 = _mm_sub_pd(iy0,jy0);
835 dz00 = _mm_sub_pd(iz0,jz0);
836 dx10 = _mm_sub_pd(ix1,jx0);
837 dy10 = _mm_sub_pd(iy1,jy0);
838 dz10 = _mm_sub_pd(iz1,jz0);
839 dx20 = _mm_sub_pd(ix2,jx0);
840 dy20 = _mm_sub_pd(iy2,jy0);
841 dz20 = _mm_sub_pd(iz2,jz0);
843 /* Calculate squared distance and things based on it */
844 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
845 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
846 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
848 rinv00 = gmx_mm_invsqrt_pd(rsq00);
849 rinv10 = gmx_mm_invsqrt_pd(rsq10);
850 rinv20 = gmx_mm_invsqrt_pd(rsq20);
852 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
853 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
854 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
856 /* Load parameters for j particles */
857 jq0 = _mm_load_sd(charge+jnrA+0);
858 vdwjidx0A = 2*vdwtype[jnrA+0];
860 fjx0 = _mm_setzero_pd();
861 fjy0 = _mm_setzero_pd();
862 fjz0 = _mm_setzero_pd();
864 /**************************
865 * CALCULATE INTERACTIONS *
866 **************************/
868 r00 = _mm_mul_pd(rsq00,rinv00);
870 /* Compute parameters for interactions between i and j atoms */
871 qq00 = _mm_mul_pd(iq0,jq0);
872 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
874 /* Calculate table index by multiplying r with table scale and truncate to integer */
875 rt = _mm_mul_pd(r00,vftabscale);
876 vfitab = _mm_cvttpd_epi32(rt);
878 vfeps = _mm_frcz_pd(rt);
880 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
882 twovfeps = _mm_add_pd(vfeps,vfeps);
883 vfitab = _mm_slli_epi32(vfitab,3);
885 /* COULOMB ELECTROSTATICS */
886 velec = _mm_mul_pd(qq00,rinv00);
887 felec = _mm_mul_pd(velec,rinvsq00);
889 /* CUBIC SPLINE TABLE DISPERSION */
890 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
891 F = _mm_setzero_pd();
892 GMX_MM_TRANSPOSE2_PD(Y,F);
893 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
894 H = _mm_setzero_pd();
895 GMX_MM_TRANSPOSE2_PD(G,H);
896 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
897 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
898 fvdw6 = _mm_mul_pd(c6_00,FF);
900 /* CUBIC SPLINE TABLE REPULSION */
901 vfitab = _mm_add_epi32(vfitab,ifour);
902 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
903 F = _mm_setzero_pd();
904 GMX_MM_TRANSPOSE2_PD(Y,F);
905 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
906 H = _mm_setzero_pd();
907 GMX_MM_TRANSPOSE2_PD(G,H);
908 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
909 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
910 fvdw12 = _mm_mul_pd(c12_00,FF);
911 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
913 fscal = _mm_add_pd(felec,fvdw);
915 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
917 /* Update vectorial force */
918 fix0 = _mm_macc_pd(dx00,fscal,fix0);
919 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
920 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
922 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
923 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
924 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
926 /**************************
927 * CALCULATE INTERACTIONS *
928 **************************/
930 /* Compute parameters for interactions between i and j atoms */
931 qq10 = _mm_mul_pd(iq1,jq0);
933 /* COULOMB ELECTROSTATICS */
934 velec = _mm_mul_pd(qq10,rinv10);
935 felec = _mm_mul_pd(velec,rinvsq10);
939 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
941 /* Update vectorial force */
942 fix1 = _mm_macc_pd(dx10,fscal,fix1);
943 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
944 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
946 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
947 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
948 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
950 /**************************
951 * CALCULATE INTERACTIONS *
952 **************************/
954 /* Compute parameters for interactions between i and j atoms */
955 qq20 = _mm_mul_pd(iq2,jq0);
957 /* COULOMB ELECTROSTATICS */
958 velec = _mm_mul_pd(qq20,rinv20);
959 felec = _mm_mul_pd(velec,rinvsq20);
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);