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_ElecRFCut_VdwNone_GeomP1P1_VF_sse4_1_double
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: None
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRFCut_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;
73 __m128d dummy_mask,cutoff_mask;
74 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
75 __m128d one = _mm_set1_pd(1.0);
76 __m128d two = _mm_set1_pd(2.0);
82 jindex = nlist->jindex;
84 shiftidx = nlist->shift;
86 shiftvec = fr->shift_vec[0];
87 fshift = fr->fshift[0];
88 facel = _mm_set1_pd(fr->epsfac);
89 charge = mdatoms->chargeA;
90 krf = _mm_set1_pd(fr->ic->k_rf);
91 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
92 crf = _mm_set1_pd(fr->ic->c_rf);
94 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
95 rcutoff_scalar = fr->rcoulomb;
96 rcutoff = _mm_set1_pd(rcutoff_scalar);
97 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
99 /* Avoid stupid compiler warnings */
107 /* Start outer loop over neighborlists */
108 for(iidx=0; iidx<nri; iidx++)
110 /* Load shift vector for this list */
111 i_shift_offset = DIM*shiftidx[iidx];
113 /* Load limits for loop over neighbors */
114 j_index_start = jindex[iidx];
115 j_index_end = jindex[iidx+1];
117 /* Get outer coordinate index */
119 i_coord_offset = DIM*inr;
121 /* Load i particle coords and add shift vector */
122 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
124 fix0 = _mm_setzero_pd();
125 fiy0 = _mm_setzero_pd();
126 fiz0 = _mm_setzero_pd();
128 /* Load parameters for i particles */
129 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
131 /* Reset potential sums */
132 velecsum = _mm_setzero_pd();
134 /* Start inner kernel loop */
135 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
138 /* Get j neighbor index, and coordinate index */
141 j_coord_offsetA = DIM*jnrA;
142 j_coord_offsetB = DIM*jnrB;
144 /* load j atom coordinates */
145 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
148 /* Calculate displacement vector */
149 dx00 = _mm_sub_pd(ix0,jx0);
150 dy00 = _mm_sub_pd(iy0,jy0);
151 dz00 = _mm_sub_pd(iz0,jz0);
153 /* Calculate squared distance and things based on it */
154 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
156 rinv00 = gmx_mm_invsqrt_pd(rsq00);
158 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
160 /* Load parameters for j particles */
161 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
163 /**************************
164 * CALCULATE INTERACTIONS *
165 **************************/
167 if (gmx_mm_any_lt(rsq00,rcutoff2))
170 /* Compute parameters for interactions between i and j atoms */
171 qq00 = _mm_mul_pd(iq0,jq0);
173 /* REACTION-FIELD ELECTROSTATICS */
174 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
175 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
177 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
179 /* Update potential sum for this i atom from the interaction with this j atom. */
180 velec = _mm_and_pd(velec,cutoff_mask);
181 velecsum = _mm_add_pd(velecsum,velec);
185 fscal = _mm_and_pd(fscal,cutoff_mask);
187 /* Calculate temporary vectorial force */
188 tx = _mm_mul_pd(fscal,dx00);
189 ty = _mm_mul_pd(fscal,dy00);
190 tz = _mm_mul_pd(fscal,dz00);
192 /* Update vectorial force */
193 fix0 = _mm_add_pd(fix0,tx);
194 fiy0 = _mm_add_pd(fiy0,ty);
195 fiz0 = _mm_add_pd(fiz0,tz);
197 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
201 /* Inner loop uses 36 flops */
208 j_coord_offsetA = DIM*jnrA;
210 /* load j atom coordinates */
211 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
214 /* Calculate displacement vector */
215 dx00 = _mm_sub_pd(ix0,jx0);
216 dy00 = _mm_sub_pd(iy0,jy0);
217 dz00 = _mm_sub_pd(iz0,jz0);
219 /* Calculate squared distance and things based on it */
220 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
222 rinv00 = gmx_mm_invsqrt_pd(rsq00);
224 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
226 /* Load parameters for j particles */
227 jq0 = _mm_load_sd(charge+jnrA+0);
229 /**************************
230 * CALCULATE INTERACTIONS *
231 **************************/
233 if (gmx_mm_any_lt(rsq00,rcutoff2))
236 /* Compute parameters for interactions between i and j atoms */
237 qq00 = _mm_mul_pd(iq0,jq0);
239 /* REACTION-FIELD ELECTROSTATICS */
240 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
241 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
243 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
245 /* Update potential sum for this i atom from the interaction with this j atom. */
246 velec = _mm_and_pd(velec,cutoff_mask);
247 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
248 velecsum = _mm_add_pd(velecsum,velec);
252 fscal = _mm_and_pd(fscal,cutoff_mask);
254 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
256 /* Calculate temporary vectorial force */
257 tx = _mm_mul_pd(fscal,dx00);
258 ty = _mm_mul_pd(fscal,dy00);
259 tz = _mm_mul_pd(fscal,dz00);
261 /* Update vectorial force */
262 fix0 = _mm_add_pd(fix0,tx);
263 fiy0 = _mm_add_pd(fiy0,ty);
264 fiz0 = _mm_add_pd(fiz0,tz);
266 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
270 /* Inner loop uses 36 flops */
273 /* End of innermost loop */
275 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
276 f+i_coord_offset,fshift+i_shift_offset);
279 /* Update potential energies */
280 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
282 /* Increment number of inner iterations */
283 inneriter += j_index_end - j_index_start;
285 /* Outer loop uses 8 flops */
288 /* Increment number of outer iterations */
291 /* Update outer/inner flops */
293 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
296 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse4_1_double
297 * Electrostatics interaction: ReactionField
298 * VdW interaction: None
299 * Geometry: Particle-Particle
300 * Calculate force/pot: Force
303 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse4_1_double
304 (t_nblist * gmx_restrict nlist,
305 rvec * gmx_restrict xx,
306 rvec * gmx_restrict ff,
307 t_forcerec * gmx_restrict fr,
308 t_mdatoms * gmx_restrict mdatoms,
309 nb_kernel_data_t * gmx_restrict kernel_data,
310 t_nrnb * gmx_restrict nrnb)
312 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
313 * just 0 for non-waters.
314 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
315 * jnr indices corresponding to data put in the four positions in the SIMD register.
317 int i_shift_offset,i_coord_offset,outeriter,inneriter;
318 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
320 int j_coord_offsetA,j_coord_offsetB;
321 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
323 real *shiftvec,*fshift,*x,*f;
324 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
326 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
327 int vdwjidx0A,vdwjidx0B;
328 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
329 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
330 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
332 __m128d dummy_mask,cutoff_mask;
333 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
334 __m128d one = _mm_set1_pd(1.0);
335 __m128d two = _mm_set1_pd(2.0);
341 jindex = nlist->jindex;
343 shiftidx = nlist->shift;
345 shiftvec = fr->shift_vec[0];
346 fshift = fr->fshift[0];
347 facel = _mm_set1_pd(fr->epsfac);
348 charge = mdatoms->chargeA;
349 krf = _mm_set1_pd(fr->ic->k_rf);
350 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
351 crf = _mm_set1_pd(fr->ic->c_rf);
353 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
354 rcutoff_scalar = fr->rcoulomb;
355 rcutoff = _mm_set1_pd(rcutoff_scalar);
356 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
358 /* Avoid stupid compiler warnings */
366 /* Start outer loop over neighborlists */
367 for(iidx=0; iidx<nri; iidx++)
369 /* Load shift vector for this list */
370 i_shift_offset = DIM*shiftidx[iidx];
372 /* Load limits for loop over neighbors */
373 j_index_start = jindex[iidx];
374 j_index_end = jindex[iidx+1];
376 /* Get outer coordinate index */
378 i_coord_offset = DIM*inr;
380 /* Load i particle coords and add shift vector */
381 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
383 fix0 = _mm_setzero_pd();
384 fiy0 = _mm_setzero_pd();
385 fiz0 = _mm_setzero_pd();
387 /* Load parameters for i particles */
388 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
390 /* Start inner kernel loop */
391 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
394 /* Get j neighbor index, and coordinate index */
397 j_coord_offsetA = DIM*jnrA;
398 j_coord_offsetB = DIM*jnrB;
400 /* load j atom coordinates */
401 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
404 /* Calculate displacement vector */
405 dx00 = _mm_sub_pd(ix0,jx0);
406 dy00 = _mm_sub_pd(iy0,jy0);
407 dz00 = _mm_sub_pd(iz0,jz0);
409 /* Calculate squared distance and things based on it */
410 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
412 rinv00 = gmx_mm_invsqrt_pd(rsq00);
414 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
416 /* Load parameters for j particles */
417 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
419 /**************************
420 * CALCULATE INTERACTIONS *
421 **************************/
423 if (gmx_mm_any_lt(rsq00,rcutoff2))
426 /* Compute parameters for interactions between i and j atoms */
427 qq00 = _mm_mul_pd(iq0,jq0);
429 /* REACTION-FIELD ELECTROSTATICS */
430 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
432 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
436 fscal = _mm_and_pd(fscal,cutoff_mask);
438 /* Calculate temporary vectorial force */
439 tx = _mm_mul_pd(fscal,dx00);
440 ty = _mm_mul_pd(fscal,dy00);
441 tz = _mm_mul_pd(fscal,dz00);
443 /* Update vectorial force */
444 fix0 = _mm_add_pd(fix0,tx);
445 fiy0 = _mm_add_pd(fiy0,ty);
446 fiz0 = _mm_add_pd(fiz0,tz);
448 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
452 /* Inner loop uses 30 flops */
459 j_coord_offsetA = DIM*jnrA;
461 /* load j atom coordinates */
462 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
465 /* Calculate displacement vector */
466 dx00 = _mm_sub_pd(ix0,jx0);
467 dy00 = _mm_sub_pd(iy0,jy0);
468 dz00 = _mm_sub_pd(iz0,jz0);
470 /* Calculate squared distance and things based on it */
471 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
473 rinv00 = gmx_mm_invsqrt_pd(rsq00);
475 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
477 /* Load parameters for j particles */
478 jq0 = _mm_load_sd(charge+jnrA+0);
480 /**************************
481 * CALCULATE INTERACTIONS *
482 **************************/
484 if (gmx_mm_any_lt(rsq00,rcutoff2))
487 /* Compute parameters for interactions between i and j atoms */
488 qq00 = _mm_mul_pd(iq0,jq0);
490 /* REACTION-FIELD ELECTROSTATICS */
491 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
493 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
497 fscal = _mm_and_pd(fscal,cutoff_mask);
499 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
501 /* Calculate temporary vectorial force */
502 tx = _mm_mul_pd(fscal,dx00);
503 ty = _mm_mul_pd(fscal,dy00);
504 tz = _mm_mul_pd(fscal,dz00);
506 /* Update vectorial force */
507 fix0 = _mm_add_pd(fix0,tx);
508 fiy0 = _mm_add_pd(fiy0,ty);
509 fiz0 = _mm_add_pd(fiz0,tz);
511 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
515 /* Inner loop uses 30 flops */
518 /* End of innermost loop */
520 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
521 f+i_coord_offset,fshift+i_shift_offset);
523 /* Increment number of inner iterations */
524 inneriter += j_index_end - j_index_start;
526 /* Outer loop uses 7 flops */
529 /* Increment number of outer iterations */
532 /* Update outer/inner flops */
534 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);