2 * Note: this file was generated by the Gromacs sse2_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_sse2_double.h"
34 #include "kernelutil_x86_sse2_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse2_double
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: None
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse2_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 /* Avoid stupid compiler warnings */
102 /* Start outer loop over neighborlists */
103 for(iidx=0; iidx<nri; iidx++)
105 /* Load shift vector for this list */
106 i_shift_offset = DIM*shiftidx[iidx];
108 /* Load limits for loop over neighbors */
109 j_index_start = jindex[iidx];
110 j_index_end = jindex[iidx+1];
112 /* Get outer coordinate index */
114 i_coord_offset = DIM*inr;
116 /* Load i particle coords and add shift vector */
117 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
119 fix0 = _mm_setzero_pd();
120 fiy0 = _mm_setzero_pd();
121 fiz0 = _mm_setzero_pd();
123 /* Load parameters for i particles */
124 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
126 /* Reset potential sums */
127 velecsum = _mm_setzero_pd();
129 /* Start inner kernel loop */
130 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
133 /* Get j neighbor index, and coordinate index */
136 j_coord_offsetA = DIM*jnrA;
137 j_coord_offsetB = DIM*jnrB;
139 /* load j atom coordinates */
140 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
143 /* Calculate displacement vector */
144 dx00 = _mm_sub_pd(ix0,jx0);
145 dy00 = _mm_sub_pd(iy0,jy0);
146 dz00 = _mm_sub_pd(iz0,jz0);
148 /* Calculate squared distance and things based on it */
149 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
151 rinv00 = gmx_mm_invsqrt_pd(rsq00);
153 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
155 /* Load parameters for j particles */
156 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
158 /**************************
159 * CALCULATE INTERACTIONS *
160 **************************/
162 /* Compute parameters for interactions between i and j atoms */
163 qq00 = _mm_mul_pd(iq0,jq0);
165 /* REACTION-FIELD ELECTROSTATICS */
166 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
167 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
169 /* Update potential sum for this i atom from the interaction with this j atom. */
170 velecsum = _mm_add_pd(velecsum,velec);
174 /* Calculate temporary vectorial force */
175 tx = _mm_mul_pd(fscal,dx00);
176 ty = _mm_mul_pd(fscal,dy00);
177 tz = _mm_mul_pd(fscal,dz00);
179 /* Update vectorial force */
180 fix0 = _mm_add_pd(fix0,tx);
181 fiy0 = _mm_add_pd(fiy0,ty);
182 fiz0 = _mm_add_pd(fiz0,tz);
184 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
186 /* Inner loop uses 32 flops */
193 j_coord_offsetA = DIM*jnrA;
195 /* load j atom coordinates */
196 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
199 /* Calculate displacement vector */
200 dx00 = _mm_sub_pd(ix0,jx0);
201 dy00 = _mm_sub_pd(iy0,jy0);
202 dz00 = _mm_sub_pd(iz0,jz0);
204 /* Calculate squared distance and things based on it */
205 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
207 rinv00 = gmx_mm_invsqrt_pd(rsq00);
209 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
211 /* Load parameters for j particles */
212 jq0 = _mm_load_sd(charge+jnrA+0);
214 /**************************
215 * CALCULATE INTERACTIONS *
216 **************************/
218 /* Compute parameters for interactions between i and j atoms */
219 qq00 = _mm_mul_pd(iq0,jq0);
221 /* REACTION-FIELD ELECTROSTATICS */
222 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
223 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
225 /* Update potential sum for this i atom from the interaction with this j atom. */
226 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
227 velecsum = _mm_add_pd(velecsum,velec);
231 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
233 /* Calculate temporary vectorial force */
234 tx = _mm_mul_pd(fscal,dx00);
235 ty = _mm_mul_pd(fscal,dy00);
236 tz = _mm_mul_pd(fscal,dz00);
238 /* Update vectorial force */
239 fix0 = _mm_add_pd(fix0,tx);
240 fiy0 = _mm_add_pd(fiy0,ty);
241 fiz0 = _mm_add_pd(fiz0,tz);
243 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
245 /* Inner loop uses 32 flops */
248 /* End of innermost loop */
250 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
251 f+i_coord_offset,fshift+i_shift_offset);
254 /* Update potential energies */
255 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
257 /* Increment number of inner iterations */
258 inneriter += j_index_end - j_index_start;
260 /* Outer loop uses 8 flops */
263 /* Increment number of outer iterations */
266 /* Update outer/inner flops */
268 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
271 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_double
272 * Electrostatics interaction: ReactionField
273 * VdW interaction: None
274 * Geometry: Particle-Particle
275 * Calculate force/pot: Force
278 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_double
279 (t_nblist * gmx_restrict nlist,
280 rvec * gmx_restrict xx,
281 rvec * gmx_restrict ff,
282 t_forcerec * gmx_restrict fr,
283 t_mdatoms * gmx_restrict mdatoms,
284 nb_kernel_data_t * gmx_restrict kernel_data,
285 t_nrnb * gmx_restrict nrnb)
287 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
288 * just 0 for non-waters.
289 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
290 * jnr indices corresponding to data put in the four positions in the SIMD register.
292 int i_shift_offset,i_coord_offset,outeriter,inneriter;
293 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
295 int j_coord_offsetA,j_coord_offsetB;
296 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
298 real *shiftvec,*fshift,*x,*f;
299 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
301 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
302 int vdwjidx0A,vdwjidx0B;
303 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
304 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
305 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
307 __m128d dummy_mask,cutoff_mask;
308 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
309 __m128d one = _mm_set1_pd(1.0);
310 __m128d two = _mm_set1_pd(2.0);
316 jindex = nlist->jindex;
318 shiftidx = nlist->shift;
320 shiftvec = fr->shift_vec[0];
321 fshift = fr->fshift[0];
322 facel = _mm_set1_pd(fr->epsfac);
323 charge = mdatoms->chargeA;
324 krf = _mm_set1_pd(fr->ic->k_rf);
325 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
326 crf = _mm_set1_pd(fr->ic->c_rf);
328 /* Avoid stupid compiler warnings */
336 /* Start outer loop over neighborlists */
337 for(iidx=0; iidx<nri; iidx++)
339 /* Load shift vector for this list */
340 i_shift_offset = DIM*shiftidx[iidx];
342 /* Load limits for loop over neighbors */
343 j_index_start = jindex[iidx];
344 j_index_end = jindex[iidx+1];
346 /* Get outer coordinate index */
348 i_coord_offset = DIM*inr;
350 /* Load i particle coords and add shift vector */
351 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
353 fix0 = _mm_setzero_pd();
354 fiy0 = _mm_setzero_pd();
355 fiz0 = _mm_setzero_pd();
357 /* Load parameters for i particles */
358 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
360 /* Start inner kernel loop */
361 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
364 /* Get j neighbor index, and coordinate index */
367 j_coord_offsetA = DIM*jnrA;
368 j_coord_offsetB = DIM*jnrB;
370 /* load j atom coordinates */
371 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
374 /* Calculate displacement vector */
375 dx00 = _mm_sub_pd(ix0,jx0);
376 dy00 = _mm_sub_pd(iy0,jy0);
377 dz00 = _mm_sub_pd(iz0,jz0);
379 /* Calculate squared distance and things based on it */
380 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
382 rinv00 = gmx_mm_invsqrt_pd(rsq00);
384 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
386 /* Load parameters for j particles */
387 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
389 /**************************
390 * CALCULATE INTERACTIONS *
391 **************************/
393 /* Compute parameters for interactions between i and j atoms */
394 qq00 = _mm_mul_pd(iq0,jq0);
396 /* REACTION-FIELD ELECTROSTATICS */
397 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
401 /* Calculate temporary vectorial force */
402 tx = _mm_mul_pd(fscal,dx00);
403 ty = _mm_mul_pd(fscal,dy00);
404 tz = _mm_mul_pd(fscal,dz00);
406 /* Update vectorial force */
407 fix0 = _mm_add_pd(fix0,tx);
408 fiy0 = _mm_add_pd(fiy0,ty);
409 fiz0 = _mm_add_pd(fiz0,tz);
411 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
413 /* Inner loop uses 27 flops */
420 j_coord_offsetA = DIM*jnrA;
422 /* load j atom coordinates */
423 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
426 /* Calculate displacement vector */
427 dx00 = _mm_sub_pd(ix0,jx0);
428 dy00 = _mm_sub_pd(iy0,jy0);
429 dz00 = _mm_sub_pd(iz0,jz0);
431 /* Calculate squared distance and things based on it */
432 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
434 rinv00 = gmx_mm_invsqrt_pd(rsq00);
436 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
438 /* Load parameters for j particles */
439 jq0 = _mm_load_sd(charge+jnrA+0);
441 /**************************
442 * CALCULATE INTERACTIONS *
443 **************************/
445 /* Compute parameters for interactions between i and j atoms */
446 qq00 = _mm_mul_pd(iq0,jq0);
448 /* REACTION-FIELD ELECTROSTATICS */
449 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
453 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
455 /* Calculate temporary vectorial force */
456 tx = _mm_mul_pd(fscal,dx00);
457 ty = _mm_mul_pd(fscal,dy00);
458 tz = _mm_mul_pd(fscal,dz00);
460 /* Update vectorial force */
461 fix0 = _mm_add_pd(fix0,tx);
462 fiy0 = _mm_add_pd(fiy0,ty);
463 fiz0 = _mm_add_pd(fiz0,tz);
465 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
467 /* Inner loop uses 27 flops */
470 /* End of innermost loop */
472 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
473 f+i_coord_offset,fshift+i_shift_offset);
475 /* Increment number of inner iterations */
476 inneriter += j_index_end - j_index_start;
478 /* Outer loop uses 7 flops */
481 /* Increment number of outer iterations */
484 /* Update outer/inner flops */
486 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);