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_ElecRF_VdwLJ_GeomP1P1_VF_avx_128_fma_double
54 * Electrostatics interaction: ReactionField
55 * VdW interaction: LennardJones
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecRF_VdwLJ_GeomP1P1_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 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
84 int vdwjidx0A,vdwjidx0B;
85 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
86 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
87 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
90 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
93 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
94 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
95 __m128d dummy_mask,cutoff_mask;
96 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
97 __m128d one = _mm_set1_pd(1.0);
98 __m128d two = _mm_set1_pd(2.0);
104 jindex = nlist->jindex;
106 shiftidx = nlist->shift;
108 shiftvec = fr->shift_vec[0];
109 fshift = fr->fshift[0];
110 facel = _mm_set1_pd(fr->epsfac);
111 charge = mdatoms->chargeA;
112 krf = _mm_set1_pd(fr->ic->k_rf);
113 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
114 crf = _mm_set1_pd(fr->ic->c_rf);
115 nvdwtype = fr->ntype;
117 vdwtype = mdatoms->typeA;
119 /* Avoid stupid compiler warnings */
127 /* Start outer loop over neighborlists */
128 for(iidx=0; iidx<nri; iidx++)
130 /* Load shift vector for this list */
131 i_shift_offset = DIM*shiftidx[iidx];
133 /* Load limits for loop over neighbors */
134 j_index_start = jindex[iidx];
135 j_index_end = jindex[iidx+1];
137 /* Get outer coordinate index */
139 i_coord_offset = DIM*inr;
141 /* Load i particle coords and add shift vector */
142 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
144 fix0 = _mm_setzero_pd();
145 fiy0 = _mm_setzero_pd();
146 fiz0 = _mm_setzero_pd();
148 /* Load parameters for i particles */
149 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
150 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
152 /* Reset potential sums */
153 velecsum = _mm_setzero_pd();
154 vvdwsum = _mm_setzero_pd();
156 /* Start inner kernel loop */
157 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
160 /* Get j neighbor index, and coordinate index */
163 j_coord_offsetA = DIM*jnrA;
164 j_coord_offsetB = DIM*jnrB;
166 /* load j atom coordinates */
167 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
170 /* Calculate displacement vector */
171 dx00 = _mm_sub_pd(ix0,jx0);
172 dy00 = _mm_sub_pd(iy0,jy0);
173 dz00 = _mm_sub_pd(iz0,jz0);
175 /* Calculate squared distance and things based on it */
176 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
178 rinv00 = gmx_mm_invsqrt_pd(rsq00);
180 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
182 /* Load parameters for j particles */
183 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
184 vdwjidx0A = 2*vdwtype[jnrA+0];
185 vdwjidx0B = 2*vdwtype[jnrB+0];
187 /**************************
188 * CALCULATE INTERACTIONS *
189 **************************/
191 /* Compute parameters for interactions between i and j atoms */
192 qq00 = _mm_mul_pd(iq0,jq0);
193 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
194 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
196 /* REACTION-FIELD ELECTROSTATICS */
197 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
198 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
200 /* LENNARD-JONES DISPERSION/REPULSION */
202 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
203 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
204 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
205 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
206 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
208 /* Update potential sum for this i atom from the interaction with this j atom. */
209 velecsum = _mm_add_pd(velecsum,velec);
210 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
212 fscal = _mm_add_pd(felec,fvdw);
214 /* Update vectorial force */
215 fix0 = _mm_macc_pd(dx00,fscal,fix0);
216 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
217 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
219 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
220 _mm_mul_pd(dx00,fscal),
221 _mm_mul_pd(dy00,fscal),
222 _mm_mul_pd(dz00,fscal));
224 /* Inner loop uses 47 flops */
231 j_coord_offsetA = DIM*jnrA;
233 /* load j atom coordinates */
234 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
237 /* Calculate displacement vector */
238 dx00 = _mm_sub_pd(ix0,jx0);
239 dy00 = _mm_sub_pd(iy0,jy0);
240 dz00 = _mm_sub_pd(iz0,jz0);
242 /* Calculate squared distance and things based on it */
243 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
245 rinv00 = gmx_mm_invsqrt_pd(rsq00);
247 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
249 /* Load parameters for j particles */
250 jq0 = _mm_load_sd(charge+jnrA+0);
251 vdwjidx0A = 2*vdwtype[jnrA+0];
253 /**************************
254 * CALCULATE INTERACTIONS *
255 **************************/
257 /* Compute parameters for interactions between i and j atoms */
258 qq00 = _mm_mul_pd(iq0,jq0);
259 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
261 /* REACTION-FIELD ELECTROSTATICS */
262 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
263 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
265 /* LENNARD-JONES DISPERSION/REPULSION */
267 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
268 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
269 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
270 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
271 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
273 /* Update potential sum for this i atom from the interaction with this j atom. */
274 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
275 velecsum = _mm_add_pd(velecsum,velec);
276 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
277 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
279 fscal = _mm_add_pd(felec,fvdw);
281 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
283 /* Update vectorial force */
284 fix0 = _mm_macc_pd(dx00,fscal,fix0);
285 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
286 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
288 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
289 _mm_mul_pd(dx00,fscal),
290 _mm_mul_pd(dy00,fscal),
291 _mm_mul_pd(dz00,fscal));
293 /* Inner loop uses 47 flops */
296 /* End of innermost loop */
298 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
299 f+i_coord_offset,fshift+i_shift_offset);
302 /* Update potential energies */
303 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
304 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
306 /* Increment number of inner iterations */
307 inneriter += j_index_end - j_index_start;
309 /* Outer loop uses 9 flops */
312 /* Increment number of outer iterations */
315 /* Update outer/inner flops */
317 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*47);
320 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_double
321 * Electrostatics interaction: ReactionField
322 * VdW interaction: LennardJones
323 * Geometry: Particle-Particle
324 * Calculate force/pot: Force
327 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_double
328 (t_nblist * gmx_restrict nlist,
329 rvec * gmx_restrict xx,
330 rvec * gmx_restrict ff,
331 t_forcerec * gmx_restrict fr,
332 t_mdatoms * gmx_restrict mdatoms,
333 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
334 t_nrnb * gmx_restrict nrnb)
336 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
337 * just 0 for non-waters.
338 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
339 * jnr indices corresponding to data put in the four positions in the SIMD register.
341 int i_shift_offset,i_coord_offset,outeriter,inneriter;
342 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
344 int j_coord_offsetA,j_coord_offsetB;
345 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
347 real *shiftvec,*fshift,*x,*f;
348 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
350 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
351 int vdwjidx0A,vdwjidx0B;
352 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
353 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
354 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
357 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
360 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
361 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
362 __m128d dummy_mask,cutoff_mask;
363 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
364 __m128d one = _mm_set1_pd(1.0);
365 __m128d two = _mm_set1_pd(2.0);
371 jindex = nlist->jindex;
373 shiftidx = nlist->shift;
375 shiftvec = fr->shift_vec[0];
376 fshift = fr->fshift[0];
377 facel = _mm_set1_pd(fr->epsfac);
378 charge = mdatoms->chargeA;
379 krf = _mm_set1_pd(fr->ic->k_rf);
380 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
381 crf = _mm_set1_pd(fr->ic->c_rf);
382 nvdwtype = fr->ntype;
384 vdwtype = mdatoms->typeA;
386 /* Avoid stupid compiler warnings */
394 /* Start outer loop over neighborlists */
395 for(iidx=0; iidx<nri; iidx++)
397 /* Load shift vector for this list */
398 i_shift_offset = DIM*shiftidx[iidx];
400 /* Load limits for loop over neighbors */
401 j_index_start = jindex[iidx];
402 j_index_end = jindex[iidx+1];
404 /* Get outer coordinate index */
406 i_coord_offset = DIM*inr;
408 /* Load i particle coords and add shift vector */
409 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
411 fix0 = _mm_setzero_pd();
412 fiy0 = _mm_setzero_pd();
413 fiz0 = _mm_setzero_pd();
415 /* Load parameters for i particles */
416 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
417 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
419 /* Start inner kernel loop */
420 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
423 /* Get j neighbor index, and coordinate index */
426 j_coord_offsetA = DIM*jnrA;
427 j_coord_offsetB = DIM*jnrB;
429 /* load j atom coordinates */
430 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
433 /* Calculate displacement vector */
434 dx00 = _mm_sub_pd(ix0,jx0);
435 dy00 = _mm_sub_pd(iy0,jy0);
436 dz00 = _mm_sub_pd(iz0,jz0);
438 /* Calculate squared distance and things based on it */
439 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
441 rinv00 = gmx_mm_invsqrt_pd(rsq00);
443 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
445 /* Load parameters for j particles */
446 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
447 vdwjidx0A = 2*vdwtype[jnrA+0];
448 vdwjidx0B = 2*vdwtype[jnrB+0];
450 /**************************
451 * CALCULATE INTERACTIONS *
452 **************************/
454 /* Compute parameters for interactions between i and j atoms */
455 qq00 = _mm_mul_pd(iq0,jq0);
456 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
457 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
459 /* REACTION-FIELD ELECTROSTATICS */
460 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
462 /* LENNARD-JONES DISPERSION/REPULSION */
464 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
465 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
467 fscal = _mm_add_pd(felec,fvdw);
469 /* Update vectorial force */
470 fix0 = _mm_macc_pd(dx00,fscal,fix0);
471 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
472 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
474 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
475 _mm_mul_pd(dx00,fscal),
476 _mm_mul_pd(dy00,fscal),
477 _mm_mul_pd(dz00,fscal));
479 /* Inner loop uses 37 flops */
486 j_coord_offsetA = DIM*jnrA;
488 /* load j atom coordinates */
489 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
492 /* Calculate displacement vector */
493 dx00 = _mm_sub_pd(ix0,jx0);
494 dy00 = _mm_sub_pd(iy0,jy0);
495 dz00 = _mm_sub_pd(iz0,jz0);
497 /* Calculate squared distance and things based on it */
498 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
500 rinv00 = gmx_mm_invsqrt_pd(rsq00);
502 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
504 /* Load parameters for j particles */
505 jq0 = _mm_load_sd(charge+jnrA+0);
506 vdwjidx0A = 2*vdwtype[jnrA+0];
508 /**************************
509 * CALCULATE INTERACTIONS *
510 **************************/
512 /* Compute parameters for interactions between i and j atoms */
513 qq00 = _mm_mul_pd(iq0,jq0);
514 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
516 /* REACTION-FIELD ELECTROSTATICS */
517 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
519 /* LENNARD-JONES DISPERSION/REPULSION */
521 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
522 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
524 fscal = _mm_add_pd(felec,fvdw);
526 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
528 /* Update vectorial force */
529 fix0 = _mm_macc_pd(dx00,fscal,fix0);
530 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
531 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
533 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
534 _mm_mul_pd(dx00,fscal),
535 _mm_mul_pd(dy00,fscal),
536 _mm_mul_pd(dz00,fscal));
538 /* Inner loop uses 37 flops */
541 /* End of innermost loop */
543 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
544 f+i_coord_offset,fshift+i_shift_offset);
546 /* Increment number of inner iterations */
547 inneriter += j_index_end - j_index_start;
549 /* Outer loop uses 7 flops */
552 /* Increment number of outer iterations */
555 /* Update outer/inner flops */
557 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);