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_ElecCoul_VdwLJ_GeomP1P1_VF_avx_128_fma_double
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: LennardJones
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_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;
98 vdwtype = mdatoms->typeA;
100 /* Avoid stupid compiler warnings */
108 /* Start outer loop over neighborlists */
109 for(iidx=0; iidx<nri; iidx++)
111 /* Load shift vector for this list */
112 i_shift_offset = DIM*shiftidx[iidx];
114 /* Load limits for loop over neighbors */
115 j_index_start = jindex[iidx];
116 j_index_end = jindex[iidx+1];
118 /* Get outer coordinate index */
120 i_coord_offset = DIM*inr;
122 /* Load i particle coords and add shift vector */
123 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
125 fix0 = _mm_setzero_pd();
126 fiy0 = _mm_setzero_pd();
127 fiz0 = _mm_setzero_pd();
129 /* Load parameters for i particles */
130 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
131 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
133 /* Reset potential sums */
134 velecsum = _mm_setzero_pd();
135 vvdwsum = _mm_setzero_pd();
137 /* Start inner kernel loop */
138 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
141 /* Get j neighbor index, and coordinate index */
144 j_coord_offsetA = DIM*jnrA;
145 j_coord_offsetB = DIM*jnrB;
147 /* load j atom coordinates */
148 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
151 /* Calculate displacement vector */
152 dx00 = _mm_sub_pd(ix0,jx0);
153 dy00 = _mm_sub_pd(iy0,jy0);
154 dz00 = _mm_sub_pd(iz0,jz0);
156 /* Calculate squared distance and things based on it */
157 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
159 rinv00 = gmx_mm_invsqrt_pd(rsq00);
161 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
163 /* Load parameters for j particles */
164 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
165 vdwjidx0A = 2*vdwtype[jnrA+0];
166 vdwjidx0B = 2*vdwtype[jnrB+0];
168 /**************************
169 * CALCULATE INTERACTIONS *
170 **************************/
172 /* Compute parameters for interactions between i and j atoms */
173 qq00 = _mm_mul_pd(iq0,jq0);
174 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
175 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
177 /* COULOMB ELECTROSTATICS */
178 velec = _mm_mul_pd(qq00,rinv00);
179 felec = _mm_mul_pd(velec,rinvsq00);
181 /* LENNARD-JONES DISPERSION/REPULSION */
183 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
184 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
185 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
186 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
187 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
189 /* Update potential sum for this i atom from the interaction with this j atom. */
190 velecsum = _mm_add_pd(velecsum,velec);
191 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
193 fscal = _mm_add_pd(felec,fvdw);
195 /* Update vectorial force */
196 fix0 = _mm_macc_pd(dx00,fscal,fix0);
197 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
198 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
200 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
201 _mm_mul_pd(dx00,fscal),
202 _mm_mul_pd(dy00,fscal),
203 _mm_mul_pd(dz00,fscal));
205 /* Inner loop uses 43 flops */
212 j_coord_offsetA = DIM*jnrA;
214 /* load j atom coordinates */
215 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
218 /* Calculate displacement vector */
219 dx00 = _mm_sub_pd(ix0,jx0);
220 dy00 = _mm_sub_pd(iy0,jy0);
221 dz00 = _mm_sub_pd(iz0,jz0);
223 /* Calculate squared distance and things based on it */
224 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
226 rinv00 = gmx_mm_invsqrt_pd(rsq00);
228 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
230 /* Load parameters for j particles */
231 jq0 = _mm_load_sd(charge+jnrA+0);
232 vdwjidx0A = 2*vdwtype[jnrA+0];
234 /**************************
235 * CALCULATE INTERACTIONS *
236 **************************/
238 /* Compute parameters for interactions between i and j atoms */
239 qq00 = _mm_mul_pd(iq0,jq0);
240 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
242 /* COULOMB ELECTROSTATICS */
243 velec = _mm_mul_pd(qq00,rinv00);
244 felec = _mm_mul_pd(velec,rinvsq00);
246 /* LENNARD-JONES DISPERSION/REPULSION */
248 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
249 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
250 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
251 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
252 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
254 /* Update potential sum for this i atom from the interaction with this j atom. */
255 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
256 velecsum = _mm_add_pd(velecsum,velec);
257 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
258 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
260 fscal = _mm_add_pd(felec,fvdw);
262 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
264 /* Update vectorial force */
265 fix0 = _mm_macc_pd(dx00,fscal,fix0);
266 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
267 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
269 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
270 _mm_mul_pd(dx00,fscal),
271 _mm_mul_pd(dy00,fscal),
272 _mm_mul_pd(dz00,fscal));
274 /* Inner loop uses 43 flops */
277 /* End of innermost loop */
279 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
280 f+i_coord_offset,fshift+i_shift_offset);
283 /* Update potential energies */
284 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
285 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
287 /* Increment number of inner iterations */
288 inneriter += j_index_end - j_index_start;
290 /* Outer loop uses 9 flops */
293 /* Increment number of outer iterations */
296 /* Update outer/inner flops */
298 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*43);
301 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_128_fma_double
302 * Electrostatics interaction: Coulomb
303 * VdW interaction: LennardJones
304 * Geometry: Particle-Particle
305 * Calculate force/pot: Force
308 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_128_fma_double
309 (t_nblist * gmx_restrict nlist,
310 rvec * gmx_restrict xx,
311 rvec * gmx_restrict ff,
312 t_forcerec * gmx_restrict fr,
313 t_mdatoms * gmx_restrict mdatoms,
314 nb_kernel_data_t * gmx_restrict kernel_data,
315 t_nrnb * gmx_restrict nrnb)
317 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
318 * just 0 for non-waters.
319 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
320 * jnr indices corresponding to data put in the four positions in the SIMD register.
322 int i_shift_offset,i_coord_offset,outeriter,inneriter;
323 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
325 int j_coord_offsetA,j_coord_offsetB;
326 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
328 real *shiftvec,*fshift,*x,*f;
329 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
331 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
332 int vdwjidx0A,vdwjidx0B;
333 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
334 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
335 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
338 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
341 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
342 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
343 __m128d dummy_mask,cutoff_mask;
344 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
345 __m128d one = _mm_set1_pd(1.0);
346 __m128d two = _mm_set1_pd(2.0);
352 jindex = nlist->jindex;
354 shiftidx = nlist->shift;
356 shiftvec = fr->shift_vec[0];
357 fshift = fr->fshift[0];
358 facel = _mm_set1_pd(fr->epsfac);
359 charge = mdatoms->chargeA;
360 nvdwtype = fr->ntype;
362 vdwtype = mdatoms->typeA;
364 /* Avoid stupid compiler warnings */
372 /* Start outer loop over neighborlists */
373 for(iidx=0; iidx<nri; iidx++)
375 /* Load shift vector for this list */
376 i_shift_offset = DIM*shiftidx[iidx];
378 /* Load limits for loop over neighbors */
379 j_index_start = jindex[iidx];
380 j_index_end = jindex[iidx+1];
382 /* Get outer coordinate index */
384 i_coord_offset = DIM*inr;
386 /* Load i particle coords and add shift vector */
387 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
389 fix0 = _mm_setzero_pd();
390 fiy0 = _mm_setzero_pd();
391 fiz0 = _mm_setzero_pd();
393 /* Load parameters for i particles */
394 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
395 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
397 /* Start inner kernel loop */
398 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
401 /* Get j neighbor index, and coordinate index */
404 j_coord_offsetA = DIM*jnrA;
405 j_coord_offsetB = DIM*jnrB;
407 /* load j atom coordinates */
408 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
411 /* Calculate displacement vector */
412 dx00 = _mm_sub_pd(ix0,jx0);
413 dy00 = _mm_sub_pd(iy0,jy0);
414 dz00 = _mm_sub_pd(iz0,jz0);
416 /* Calculate squared distance and things based on it */
417 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
419 rinv00 = gmx_mm_invsqrt_pd(rsq00);
421 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
423 /* Load parameters for j particles */
424 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
425 vdwjidx0A = 2*vdwtype[jnrA+0];
426 vdwjidx0B = 2*vdwtype[jnrB+0];
428 /**************************
429 * CALCULATE INTERACTIONS *
430 **************************/
432 /* Compute parameters for interactions between i and j atoms */
433 qq00 = _mm_mul_pd(iq0,jq0);
434 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
435 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
437 /* COULOMB ELECTROSTATICS */
438 velec = _mm_mul_pd(qq00,rinv00);
439 felec = _mm_mul_pd(velec,rinvsq00);
441 /* LENNARD-JONES DISPERSION/REPULSION */
443 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
444 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
446 fscal = _mm_add_pd(felec,fvdw);
448 /* Update vectorial force */
449 fix0 = _mm_macc_pd(dx00,fscal,fix0);
450 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
451 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
453 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
454 _mm_mul_pd(dx00,fscal),
455 _mm_mul_pd(dy00,fscal),
456 _mm_mul_pd(dz00,fscal));
458 /* Inner loop uses 37 flops */
465 j_coord_offsetA = DIM*jnrA;
467 /* load j atom coordinates */
468 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
471 /* Calculate displacement vector */
472 dx00 = _mm_sub_pd(ix0,jx0);
473 dy00 = _mm_sub_pd(iy0,jy0);
474 dz00 = _mm_sub_pd(iz0,jz0);
476 /* Calculate squared distance and things based on it */
477 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
479 rinv00 = gmx_mm_invsqrt_pd(rsq00);
481 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
483 /* Load parameters for j particles */
484 jq0 = _mm_load_sd(charge+jnrA+0);
485 vdwjidx0A = 2*vdwtype[jnrA+0];
487 /**************************
488 * CALCULATE INTERACTIONS *
489 **************************/
491 /* Compute parameters for interactions between i and j atoms */
492 qq00 = _mm_mul_pd(iq0,jq0);
493 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
495 /* COULOMB ELECTROSTATICS */
496 velec = _mm_mul_pd(qq00,rinv00);
497 felec = _mm_mul_pd(velec,rinvsq00);
499 /* LENNARD-JONES DISPERSION/REPULSION */
501 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
502 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
504 fscal = _mm_add_pd(felec,fvdw);
506 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
508 /* Update vectorial force */
509 fix0 = _mm_macc_pd(dx00,fscal,fix0);
510 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
511 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
513 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
514 _mm_mul_pd(dx00,fscal),
515 _mm_mul_pd(dy00,fscal),
516 _mm_mul_pd(dz00,fscal));
518 /* Inner loop uses 37 flops */
521 /* End of innermost loop */
523 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
524 f+i_coord_offset,fshift+i_shift_offset);
526 /* Increment number of inner iterations */
527 inneriter += j_index_end - j_index_start;
529 /* Outer loop uses 7 flops */
532 /* Increment number of outer iterations */
535 /* Update outer/inner flops */
537 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);