2 * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_double.h"
34 #include "kernelutil_x86_sse4_1_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse4_1_double
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse4_1_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 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
76 __m128d dummy_mask,cutoff_mask;
77 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
78 __m128d one = _mm_set1_pd(1.0);
79 __m128d two = _mm_set1_pd(2.0);
85 jindex = nlist->jindex;
87 shiftidx = nlist->shift;
89 shiftvec = fr->shift_vec[0];
90 fshift = fr->fshift[0];
91 facel = _mm_set1_pd(fr->epsfac);
92 charge = mdatoms->chargeA;
94 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
95 ewtab = fr->ic->tabq_coul_FDV0;
96 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
97 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
99 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
100 rcutoff_scalar = fr->rcoulomb;
101 rcutoff = _mm_set1_pd(rcutoff_scalar);
102 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
104 /* Avoid stupid compiler warnings */
112 /* Start outer loop over neighborlists */
113 for(iidx=0; iidx<nri; iidx++)
115 /* Load shift vector for this list */
116 i_shift_offset = DIM*shiftidx[iidx];
118 /* Load limits for loop over neighbors */
119 j_index_start = jindex[iidx];
120 j_index_end = jindex[iidx+1];
122 /* Get outer coordinate index */
124 i_coord_offset = DIM*inr;
126 /* Load i particle coords and add shift vector */
127 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
129 fix0 = _mm_setzero_pd();
130 fiy0 = _mm_setzero_pd();
131 fiz0 = _mm_setzero_pd();
133 /* Load parameters for i particles */
134 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
136 /* Reset potential sums */
137 velecsum = _mm_setzero_pd();
139 /* Start inner kernel loop */
140 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
143 /* Get j neighbor index, and coordinate index */
146 j_coord_offsetA = DIM*jnrA;
147 j_coord_offsetB = DIM*jnrB;
149 /* load j atom coordinates */
150 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
153 /* Calculate displacement vector */
154 dx00 = _mm_sub_pd(ix0,jx0);
155 dy00 = _mm_sub_pd(iy0,jy0);
156 dz00 = _mm_sub_pd(iz0,jz0);
158 /* Calculate squared distance and things based on it */
159 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
161 rinv00 = gmx_mm_invsqrt_pd(rsq00);
163 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
165 /* Load parameters for j particles */
166 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
168 /**************************
169 * CALCULATE INTERACTIONS *
170 **************************/
172 if (gmx_mm_any_lt(rsq00,rcutoff2))
175 r00 = _mm_mul_pd(rsq00,rinv00);
177 /* Compute parameters for interactions between i and j atoms */
178 qq00 = _mm_mul_pd(iq0,jq0);
180 /* EWALD ELECTROSTATICS */
182 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
183 ewrt = _mm_mul_pd(r00,ewtabscale);
184 ewitab = _mm_cvttpd_epi32(ewrt);
185 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
186 ewitab = _mm_slli_epi32(ewitab,2);
187 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
188 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
189 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
190 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
191 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
192 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
193 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
194 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
195 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
196 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
198 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
200 /* Update potential sum for this i atom from the interaction with this j atom. */
201 velec = _mm_and_pd(velec,cutoff_mask);
202 velecsum = _mm_add_pd(velecsum,velec);
206 fscal = _mm_and_pd(fscal,cutoff_mask);
208 /* Calculate temporary vectorial force */
209 tx = _mm_mul_pd(fscal,dx00);
210 ty = _mm_mul_pd(fscal,dy00);
211 tz = _mm_mul_pd(fscal,dz00);
213 /* Update vectorial force */
214 fix0 = _mm_add_pd(fix0,tx);
215 fiy0 = _mm_add_pd(fiy0,ty);
216 fiz0 = _mm_add_pd(fiz0,tz);
218 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
222 /* Inner loop uses 46 flops */
229 j_coord_offsetA = DIM*jnrA;
231 /* load j atom coordinates */
232 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
235 /* Calculate displacement vector */
236 dx00 = _mm_sub_pd(ix0,jx0);
237 dy00 = _mm_sub_pd(iy0,jy0);
238 dz00 = _mm_sub_pd(iz0,jz0);
240 /* Calculate squared distance and things based on it */
241 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
243 rinv00 = gmx_mm_invsqrt_pd(rsq00);
245 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
247 /* Load parameters for j particles */
248 jq0 = _mm_load_sd(charge+jnrA+0);
250 /**************************
251 * CALCULATE INTERACTIONS *
252 **************************/
254 if (gmx_mm_any_lt(rsq00,rcutoff2))
257 r00 = _mm_mul_pd(rsq00,rinv00);
259 /* Compute parameters for interactions between i and j atoms */
260 qq00 = _mm_mul_pd(iq0,jq0);
262 /* EWALD ELECTROSTATICS */
264 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
265 ewrt = _mm_mul_pd(r00,ewtabscale);
266 ewitab = _mm_cvttpd_epi32(ewrt);
267 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
268 ewitab = _mm_slli_epi32(ewitab,2);
269 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
270 ewtabD = _mm_setzero_pd();
271 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
272 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
273 ewtabFn = _mm_setzero_pd();
274 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
275 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
276 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
277 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
278 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
280 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
282 /* Update potential sum for this i atom from the interaction with this j atom. */
283 velec = _mm_and_pd(velec,cutoff_mask);
284 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
285 velecsum = _mm_add_pd(velecsum,velec);
289 fscal = _mm_and_pd(fscal,cutoff_mask);
291 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
293 /* Calculate temporary vectorial force */
294 tx = _mm_mul_pd(fscal,dx00);
295 ty = _mm_mul_pd(fscal,dy00);
296 tz = _mm_mul_pd(fscal,dz00);
298 /* Update vectorial force */
299 fix0 = _mm_add_pd(fix0,tx);
300 fiy0 = _mm_add_pd(fiy0,ty);
301 fiz0 = _mm_add_pd(fiz0,tz);
303 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
307 /* Inner loop uses 46 flops */
310 /* End of innermost loop */
312 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
313 f+i_coord_offset,fshift+i_shift_offset);
316 /* Update potential energies */
317 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
319 /* Increment number of inner iterations */
320 inneriter += j_index_end - j_index_start;
322 /* Outer loop uses 8 flops */
325 /* Increment number of outer iterations */
328 /* Update outer/inner flops */
330 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*46);
333 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse4_1_double
334 * Electrostatics interaction: Ewald
335 * VdW interaction: None
336 * Geometry: Particle-Particle
337 * Calculate force/pot: Force
340 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse4_1_double
341 (t_nblist * gmx_restrict nlist,
342 rvec * gmx_restrict xx,
343 rvec * gmx_restrict ff,
344 t_forcerec * gmx_restrict fr,
345 t_mdatoms * gmx_restrict mdatoms,
346 nb_kernel_data_t * gmx_restrict kernel_data,
347 t_nrnb * gmx_restrict nrnb)
349 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
350 * just 0 for non-waters.
351 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
352 * jnr indices corresponding to data put in the four positions in the SIMD register.
354 int i_shift_offset,i_coord_offset,outeriter,inneriter;
355 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
357 int j_coord_offsetA,j_coord_offsetB;
358 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
360 real *shiftvec,*fshift,*x,*f;
361 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
363 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
364 int vdwjidx0A,vdwjidx0B;
365 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
366 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
367 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
370 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
372 __m128d dummy_mask,cutoff_mask;
373 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
374 __m128d one = _mm_set1_pd(1.0);
375 __m128d two = _mm_set1_pd(2.0);
381 jindex = nlist->jindex;
383 shiftidx = nlist->shift;
385 shiftvec = fr->shift_vec[0];
386 fshift = fr->fshift[0];
387 facel = _mm_set1_pd(fr->epsfac);
388 charge = mdatoms->chargeA;
390 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
391 ewtab = fr->ic->tabq_coul_F;
392 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
393 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
395 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
396 rcutoff_scalar = fr->rcoulomb;
397 rcutoff = _mm_set1_pd(rcutoff_scalar);
398 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
400 /* Avoid stupid compiler warnings */
408 /* Start outer loop over neighborlists */
409 for(iidx=0; iidx<nri; iidx++)
411 /* Load shift vector for this list */
412 i_shift_offset = DIM*shiftidx[iidx];
414 /* Load limits for loop over neighbors */
415 j_index_start = jindex[iidx];
416 j_index_end = jindex[iidx+1];
418 /* Get outer coordinate index */
420 i_coord_offset = DIM*inr;
422 /* Load i particle coords and add shift vector */
423 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
425 fix0 = _mm_setzero_pd();
426 fiy0 = _mm_setzero_pd();
427 fiz0 = _mm_setzero_pd();
429 /* Load parameters for i particles */
430 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
432 /* Start inner kernel loop */
433 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
436 /* Get j neighbor index, and coordinate index */
439 j_coord_offsetA = DIM*jnrA;
440 j_coord_offsetB = DIM*jnrB;
442 /* load j atom coordinates */
443 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
446 /* Calculate displacement vector */
447 dx00 = _mm_sub_pd(ix0,jx0);
448 dy00 = _mm_sub_pd(iy0,jy0);
449 dz00 = _mm_sub_pd(iz0,jz0);
451 /* Calculate squared distance and things based on it */
452 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
454 rinv00 = gmx_mm_invsqrt_pd(rsq00);
456 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
458 /* Load parameters for j particles */
459 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
461 /**************************
462 * CALCULATE INTERACTIONS *
463 **************************/
465 if (gmx_mm_any_lt(rsq00,rcutoff2))
468 r00 = _mm_mul_pd(rsq00,rinv00);
470 /* Compute parameters for interactions between i and j atoms */
471 qq00 = _mm_mul_pd(iq0,jq0);
473 /* EWALD ELECTROSTATICS */
475 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
476 ewrt = _mm_mul_pd(r00,ewtabscale);
477 ewitab = _mm_cvttpd_epi32(ewrt);
478 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
479 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
481 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
482 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
484 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
488 fscal = _mm_and_pd(fscal,cutoff_mask);
490 /* Calculate temporary vectorial force */
491 tx = _mm_mul_pd(fscal,dx00);
492 ty = _mm_mul_pd(fscal,dy00);
493 tz = _mm_mul_pd(fscal,dz00);
495 /* Update vectorial force */
496 fix0 = _mm_add_pd(fix0,tx);
497 fiy0 = _mm_add_pd(fiy0,ty);
498 fiz0 = _mm_add_pd(fiz0,tz);
500 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
504 /* Inner loop uses 39 flops */
511 j_coord_offsetA = DIM*jnrA;
513 /* load j atom coordinates */
514 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
517 /* Calculate displacement vector */
518 dx00 = _mm_sub_pd(ix0,jx0);
519 dy00 = _mm_sub_pd(iy0,jy0);
520 dz00 = _mm_sub_pd(iz0,jz0);
522 /* Calculate squared distance and things based on it */
523 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
525 rinv00 = gmx_mm_invsqrt_pd(rsq00);
527 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
529 /* Load parameters for j particles */
530 jq0 = _mm_load_sd(charge+jnrA+0);
532 /**************************
533 * CALCULATE INTERACTIONS *
534 **************************/
536 if (gmx_mm_any_lt(rsq00,rcutoff2))
539 r00 = _mm_mul_pd(rsq00,rinv00);
541 /* Compute parameters for interactions between i and j atoms */
542 qq00 = _mm_mul_pd(iq0,jq0);
544 /* EWALD ELECTROSTATICS */
546 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
547 ewrt = _mm_mul_pd(r00,ewtabscale);
548 ewitab = _mm_cvttpd_epi32(ewrt);
549 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
550 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
551 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
552 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
554 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
558 fscal = _mm_and_pd(fscal,cutoff_mask);
560 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
562 /* Calculate temporary vectorial force */
563 tx = _mm_mul_pd(fscal,dx00);
564 ty = _mm_mul_pd(fscal,dy00);
565 tz = _mm_mul_pd(fscal,dz00);
567 /* Update vectorial force */
568 fix0 = _mm_add_pd(fix0,tx);
569 fiy0 = _mm_add_pd(fiy0,ty);
570 fiz0 = _mm_add_pd(fiz0,tz);
572 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
576 /* Inner loop uses 39 flops */
579 /* End of innermost loop */
581 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
582 f+i_coord_offset,fshift+i_shift_offset);
584 /* Increment number of inner iterations */
585 inneriter += j_index_end - j_index_start;
587 /* Outer loop uses 7 flops */
590 /* Increment number of outer iterations */
593 /* Update outer/inner flops */
595 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);