2 * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_avx_128_fma_double.h"
34 #include "kernelutil_x86_avx_128_fma_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_128_fma_double
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: LennardJones
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_128_fma_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
68 int vdwjidx0A,vdwjidx0B;
69 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
70 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
71 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
74 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
77 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
78 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
79 __m128d dummy_mask,cutoff_mask;
80 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
81 __m128d one = _mm_set1_pd(1.0);
82 __m128d two = _mm_set1_pd(2.0);
88 jindex = nlist->jindex;
90 shiftidx = nlist->shift;
92 shiftvec = fr->shift_vec[0];
93 fshift = fr->fshift[0];
94 facel = _mm_set1_pd(fr->epsfac);
95 charge = mdatoms->chargeA;
96 krf = _mm_set1_pd(fr->ic->k_rf);
97 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
98 crf = _mm_set1_pd(fr->ic->c_rf);
101 vdwtype = mdatoms->typeA;
103 /* Avoid stupid compiler warnings */
111 /* Start outer loop over neighborlists */
112 for(iidx=0; iidx<nri; iidx++)
114 /* Load shift vector for this list */
115 i_shift_offset = DIM*shiftidx[iidx];
117 /* Load limits for loop over neighbors */
118 j_index_start = jindex[iidx];
119 j_index_end = jindex[iidx+1];
121 /* Get outer coordinate index */
123 i_coord_offset = DIM*inr;
125 /* Load i particle coords and add shift vector */
126 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
128 fix0 = _mm_setzero_pd();
129 fiy0 = _mm_setzero_pd();
130 fiz0 = _mm_setzero_pd();
132 /* Load parameters for i particles */
133 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
134 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
136 /* Reset potential sums */
137 velecsum = _mm_setzero_pd();
138 vvdwsum = _mm_setzero_pd();
140 /* Start inner kernel loop */
141 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
144 /* Get j neighbor index, and coordinate index */
147 j_coord_offsetA = DIM*jnrA;
148 j_coord_offsetB = DIM*jnrB;
150 /* load j atom coordinates */
151 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
154 /* Calculate displacement vector */
155 dx00 = _mm_sub_pd(ix0,jx0);
156 dy00 = _mm_sub_pd(iy0,jy0);
157 dz00 = _mm_sub_pd(iz0,jz0);
159 /* Calculate squared distance and things based on it */
160 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
162 rinv00 = gmx_mm_invsqrt_pd(rsq00);
164 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
166 /* Load parameters for j particles */
167 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
168 vdwjidx0A = 2*vdwtype[jnrA+0];
169 vdwjidx0B = 2*vdwtype[jnrB+0];
171 /**************************
172 * CALCULATE INTERACTIONS *
173 **************************/
175 /* Compute parameters for interactions between i and j atoms */
176 qq00 = _mm_mul_pd(iq0,jq0);
177 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
178 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
180 /* REACTION-FIELD ELECTROSTATICS */
181 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
182 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
184 /* LENNARD-JONES DISPERSION/REPULSION */
186 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
187 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
188 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
189 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
190 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
192 /* Update potential sum for this i atom from the interaction with this j atom. */
193 velecsum = _mm_add_pd(velecsum,velec);
194 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
196 fscal = _mm_add_pd(felec,fvdw);
198 /* Update vectorial force */
199 fix0 = _mm_macc_pd(dx00,fscal,fix0);
200 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
201 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
203 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
204 _mm_mul_pd(dx00,fscal),
205 _mm_mul_pd(dy00,fscal),
206 _mm_mul_pd(dz00,fscal));
208 /* Inner loop uses 47 flops */
215 j_coord_offsetA = DIM*jnrA;
217 /* load j atom coordinates */
218 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
221 /* Calculate displacement vector */
222 dx00 = _mm_sub_pd(ix0,jx0);
223 dy00 = _mm_sub_pd(iy0,jy0);
224 dz00 = _mm_sub_pd(iz0,jz0);
226 /* Calculate squared distance and things based on it */
227 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
229 rinv00 = gmx_mm_invsqrt_pd(rsq00);
231 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
233 /* Load parameters for j particles */
234 jq0 = _mm_load_sd(charge+jnrA+0);
235 vdwjidx0A = 2*vdwtype[jnrA+0];
237 /**************************
238 * CALCULATE INTERACTIONS *
239 **************************/
241 /* Compute parameters for interactions between i and j atoms */
242 qq00 = _mm_mul_pd(iq0,jq0);
243 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
245 /* REACTION-FIELD ELECTROSTATICS */
246 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
247 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
249 /* LENNARD-JONES DISPERSION/REPULSION */
251 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
252 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
253 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
254 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
255 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
257 /* Update potential sum for this i atom from the interaction with this j atom. */
258 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
259 velecsum = _mm_add_pd(velecsum,velec);
260 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
261 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
263 fscal = _mm_add_pd(felec,fvdw);
265 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
267 /* Update vectorial force */
268 fix0 = _mm_macc_pd(dx00,fscal,fix0);
269 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
270 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
272 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
273 _mm_mul_pd(dx00,fscal),
274 _mm_mul_pd(dy00,fscal),
275 _mm_mul_pd(dz00,fscal));
277 /* Inner loop uses 47 flops */
280 /* End of innermost loop */
282 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
283 f+i_coord_offset,fshift+i_shift_offset);
286 /* Update potential energies */
287 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
288 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
290 /* Increment number of inner iterations */
291 inneriter += j_index_end - j_index_start;
293 /* Outer loop uses 9 flops */
296 /* Increment number of outer iterations */
299 /* Update outer/inner flops */
301 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*47);
304 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_double
305 * Electrostatics interaction: ReactionField
306 * VdW interaction: LennardJones
307 * Geometry: Particle-Particle
308 * Calculate force/pot: Force
311 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_double
312 (t_nblist * gmx_restrict nlist,
313 rvec * gmx_restrict xx,
314 rvec * gmx_restrict ff,
315 t_forcerec * gmx_restrict fr,
316 t_mdatoms * gmx_restrict mdatoms,
317 nb_kernel_data_t * gmx_restrict kernel_data,
318 t_nrnb * gmx_restrict nrnb)
320 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
321 * just 0 for non-waters.
322 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
323 * jnr indices corresponding to data put in the four positions in the SIMD register.
325 int i_shift_offset,i_coord_offset,outeriter,inneriter;
326 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
328 int j_coord_offsetA,j_coord_offsetB;
329 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
331 real *shiftvec,*fshift,*x,*f;
332 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
334 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
335 int vdwjidx0A,vdwjidx0B;
336 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
337 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
338 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
341 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
344 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
345 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
346 __m128d dummy_mask,cutoff_mask;
347 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
348 __m128d one = _mm_set1_pd(1.0);
349 __m128d two = _mm_set1_pd(2.0);
355 jindex = nlist->jindex;
357 shiftidx = nlist->shift;
359 shiftvec = fr->shift_vec[0];
360 fshift = fr->fshift[0];
361 facel = _mm_set1_pd(fr->epsfac);
362 charge = mdatoms->chargeA;
363 krf = _mm_set1_pd(fr->ic->k_rf);
364 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
365 crf = _mm_set1_pd(fr->ic->c_rf);
366 nvdwtype = fr->ntype;
368 vdwtype = mdatoms->typeA;
370 /* Avoid stupid compiler warnings */
378 /* Start outer loop over neighborlists */
379 for(iidx=0; iidx<nri; iidx++)
381 /* Load shift vector for this list */
382 i_shift_offset = DIM*shiftidx[iidx];
384 /* Load limits for loop over neighbors */
385 j_index_start = jindex[iidx];
386 j_index_end = jindex[iidx+1];
388 /* Get outer coordinate index */
390 i_coord_offset = DIM*inr;
392 /* Load i particle coords and add shift vector */
393 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
395 fix0 = _mm_setzero_pd();
396 fiy0 = _mm_setzero_pd();
397 fiz0 = _mm_setzero_pd();
399 /* Load parameters for i particles */
400 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
401 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
403 /* Start inner kernel loop */
404 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
407 /* Get j neighbor index, and coordinate index */
410 j_coord_offsetA = DIM*jnrA;
411 j_coord_offsetB = DIM*jnrB;
413 /* load j atom coordinates */
414 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
417 /* Calculate displacement vector */
418 dx00 = _mm_sub_pd(ix0,jx0);
419 dy00 = _mm_sub_pd(iy0,jy0);
420 dz00 = _mm_sub_pd(iz0,jz0);
422 /* Calculate squared distance and things based on it */
423 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
425 rinv00 = gmx_mm_invsqrt_pd(rsq00);
427 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
429 /* Load parameters for j particles */
430 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
431 vdwjidx0A = 2*vdwtype[jnrA+0];
432 vdwjidx0B = 2*vdwtype[jnrB+0];
434 /**************************
435 * CALCULATE INTERACTIONS *
436 **************************/
438 /* Compute parameters for interactions between i and j atoms */
439 qq00 = _mm_mul_pd(iq0,jq0);
440 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
441 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
443 /* REACTION-FIELD ELECTROSTATICS */
444 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
446 /* LENNARD-JONES DISPERSION/REPULSION */
448 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
449 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
451 fscal = _mm_add_pd(felec,fvdw);
453 /* Update vectorial force */
454 fix0 = _mm_macc_pd(dx00,fscal,fix0);
455 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
456 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
458 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
459 _mm_mul_pd(dx00,fscal),
460 _mm_mul_pd(dy00,fscal),
461 _mm_mul_pd(dz00,fscal));
463 /* Inner loop uses 37 flops */
470 j_coord_offsetA = DIM*jnrA;
472 /* load j atom coordinates */
473 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
476 /* Calculate displacement vector */
477 dx00 = _mm_sub_pd(ix0,jx0);
478 dy00 = _mm_sub_pd(iy0,jy0);
479 dz00 = _mm_sub_pd(iz0,jz0);
481 /* Calculate squared distance and things based on it */
482 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
484 rinv00 = gmx_mm_invsqrt_pd(rsq00);
486 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
488 /* Load parameters for j particles */
489 jq0 = _mm_load_sd(charge+jnrA+0);
490 vdwjidx0A = 2*vdwtype[jnrA+0];
492 /**************************
493 * CALCULATE INTERACTIONS *
494 **************************/
496 /* Compute parameters for interactions between i and j atoms */
497 qq00 = _mm_mul_pd(iq0,jq0);
498 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
500 /* REACTION-FIELD ELECTROSTATICS */
501 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
503 /* LENNARD-JONES DISPERSION/REPULSION */
505 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
506 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
508 fscal = _mm_add_pd(felec,fvdw);
510 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
512 /* Update vectorial force */
513 fix0 = _mm_macc_pd(dx00,fscal,fix0);
514 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
515 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
517 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
518 _mm_mul_pd(dx00,fscal),
519 _mm_mul_pd(dy00,fscal),
520 _mm_mul_pd(dz00,fscal));
522 /* Inner loop uses 37 flops */
525 /* End of innermost loop */
527 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
528 f+i_coord_offset,fshift+i_shift_offset);
530 /* Increment number of inner iterations */
531 inneriter += j_index_end - j_index_start;
533 /* Outer loop uses 7 flops */
536 /* Increment number of outer iterations */
539 /* Update outer/inner flops */
541 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);