2 * Note: this file was generated by the Gromacs sse4_1_single 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_single.h"
34 #include "kernelutil_x86_sse4_1_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_single
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: LennardJones
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_single
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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
63 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
65 real *shiftvec,*fshift,*x,*f;
66 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
68 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
72 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
75 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
76 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
77 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
78 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
79 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
80 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
83 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
86 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
87 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
88 __m128 dummy_mask,cutoff_mask;
89 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
90 __m128 one = _mm_set1_ps(1.0);
91 __m128 two = _mm_set1_ps(2.0);
97 jindex = nlist->jindex;
99 shiftidx = nlist->shift;
101 shiftvec = fr->shift_vec[0];
102 fshift = fr->fshift[0];
103 facel = _mm_set1_ps(fr->epsfac);
104 charge = mdatoms->chargeA;
105 nvdwtype = fr->ntype;
107 vdwtype = mdatoms->typeA;
109 /* Setup water-specific parameters */
110 inr = nlist->iinr[0];
111 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
112 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
113 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
114 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
116 /* Avoid stupid compiler warnings */
117 jnrA = jnrB = jnrC = jnrD = 0;
126 for(iidx=0;iidx<4*DIM;iidx++)
131 /* Start outer loop over neighborlists */
132 for(iidx=0; iidx<nri; iidx++)
134 /* Load shift vector for this list */
135 i_shift_offset = DIM*shiftidx[iidx];
137 /* Load limits for loop over neighbors */
138 j_index_start = jindex[iidx];
139 j_index_end = jindex[iidx+1];
141 /* Get outer coordinate index */
143 i_coord_offset = DIM*inr;
145 /* Load i particle coords and add shift vector */
146 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
147 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
149 fix0 = _mm_setzero_ps();
150 fiy0 = _mm_setzero_ps();
151 fiz0 = _mm_setzero_ps();
152 fix1 = _mm_setzero_ps();
153 fiy1 = _mm_setzero_ps();
154 fiz1 = _mm_setzero_ps();
155 fix2 = _mm_setzero_ps();
156 fiy2 = _mm_setzero_ps();
157 fiz2 = _mm_setzero_ps();
159 /* Reset potential sums */
160 velecsum = _mm_setzero_ps();
161 vvdwsum = _mm_setzero_ps();
163 /* Start inner kernel loop */
164 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
167 /* Get j neighbor index, and coordinate index */
172 j_coord_offsetA = DIM*jnrA;
173 j_coord_offsetB = DIM*jnrB;
174 j_coord_offsetC = DIM*jnrC;
175 j_coord_offsetD = DIM*jnrD;
177 /* load j atom coordinates */
178 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
179 x+j_coord_offsetC,x+j_coord_offsetD,
182 /* Calculate displacement vector */
183 dx00 = _mm_sub_ps(ix0,jx0);
184 dy00 = _mm_sub_ps(iy0,jy0);
185 dz00 = _mm_sub_ps(iz0,jz0);
186 dx10 = _mm_sub_ps(ix1,jx0);
187 dy10 = _mm_sub_ps(iy1,jy0);
188 dz10 = _mm_sub_ps(iz1,jz0);
189 dx20 = _mm_sub_ps(ix2,jx0);
190 dy20 = _mm_sub_ps(iy2,jy0);
191 dz20 = _mm_sub_ps(iz2,jz0);
193 /* Calculate squared distance and things based on it */
194 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
195 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
196 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
198 rinv00 = gmx_mm_invsqrt_ps(rsq00);
199 rinv10 = gmx_mm_invsqrt_ps(rsq10);
200 rinv20 = gmx_mm_invsqrt_ps(rsq20);
202 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
203 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
204 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
206 /* Load parameters for j particles */
207 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
208 charge+jnrC+0,charge+jnrD+0);
209 vdwjidx0A = 2*vdwtype[jnrA+0];
210 vdwjidx0B = 2*vdwtype[jnrB+0];
211 vdwjidx0C = 2*vdwtype[jnrC+0];
212 vdwjidx0D = 2*vdwtype[jnrD+0];
214 /**************************
215 * CALCULATE INTERACTIONS *
216 **************************/
218 /* Compute parameters for interactions between i and j atoms */
219 qq00 = _mm_mul_ps(iq0,jq0);
220 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
221 vdwparam+vdwioffset0+vdwjidx0B,
222 vdwparam+vdwioffset0+vdwjidx0C,
223 vdwparam+vdwioffset0+vdwjidx0D,
226 /* COULOMB ELECTROSTATICS */
227 velec = _mm_mul_ps(qq00,rinv00);
228 felec = _mm_mul_ps(velec,rinvsq00);
230 /* LENNARD-JONES DISPERSION/REPULSION */
232 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
233 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
234 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
235 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
236 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
238 /* Update potential sum for this i atom from the interaction with this j atom. */
239 velecsum = _mm_add_ps(velecsum,velec);
240 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
242 fscal = _mm_add_ps(felec,fvdw);
244 /* Calculate temporary vectorial force */
245 tx = _mm_mul_ps(fscal,dx00);
246 ty = _mm_mul_ps(fscal,dy00);
247 tz = _mm_mul_ps(fscal,dz00);
249 /* Update vectorial force */
250 fix0 = _mm_add_ps(fix0,tx);
251 fiy0 = _mm_add_ps(fiy0,ty);
252 fiz0 = _mm_add_ps(fiz0,tz);
254 fjptrA = f+j_coord_offsetA;
255 fjptrB = f+j_coord_offsetB;
256 fjptrC = f+j_coord_offsetC;
257 fjptrD = f+j_coord_offsetD;
258 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
260 /**************************
261 * CALCULATE INTERACTIONS *
262 **************************/
264 /* Compute parameters for interactions between i and j atoms */
265 qq10 = _mm_mul_ps(iq1,jq0);
267 /* COULOMB ELECTROSTATICS */
268 velec = _mm_mul_ps(qq10,rinv10);
269 felec = _mm_mul_ps(velec,rinvsq10);
271 /* Update potential sum for this i atom from the interaction with this j atom. */
272 velecsum = _mm_add_ps(velecsum,velec);
276 /* Calculate temporary vectorial force */
277 tx = _mm_mul_ps(fscal,dx10);
278 ty = _mm_mul_ps(fscal,dy10);
279 tz = _mm_mul_ps(fscal,dz10);
281 /* Update vectorial force */
282 fix1 = _mm_add_ps(fix1,tx);
283 fiy1 = _mm_add_ps(fiy1,ty);
284 fiz1 = _mm_add_ps(fiz1,tz);
286 fjptrA = f+j_coord_offsetA;
287 fjptrB = f+j_coord_offsetB;
288 fjptrC = f+j_coord_offsetC;
289 fjptrD = f+j_coord_offsetD;
290 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
292 /**************************
293 * CALCULATE INTERACTIONS *
294 **************************/
296 /* Compute parameters for interactions between i and j atoms */
297 qq20 = _mm_mul_ps(iq2,jq0);
299 /* COULOMB ELECTROSTATICS */
300 velec = _mm_mul_ps(qq20,rinv20);
301 felec = _mm_mul_ps(velec,rinvsq20);
303 /* Update potential sum for this i atom from the interaction with this j atom. */
304 velecsum = _mm_add_ps(velecsum,velec);
308 /* Calculate temporary vectorial force */
309 tx = _mm_mul_ps(fscal,dx20);
310 ty = _mm_mul_ps(fscal,dy20);
311 tz = _mm_mul_ps(fscal,dz20);
313 /* Update vectorial force */
314 fix2 = _mm_add_ps(fix2,tx);
315 fiy2 = _mm_add_ps(fiy2,ty);
316 fiz2 = _mm_add_ps(fiz2,tz);
318 fjptrA = f+j_coord_offsetA;
319 fjptrB = f+j_coord_offsetB;
320 fjptrC = f+j_coord_offsetC;
321 fjptrD = f+j_coord_offsetD;
322 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
324 /* Inner loop uses 96 flops */
330 /* Get j neighbor index, and coordinate index */
331 jnrlistA = jjnr[jidx];
332 jnrlistB = jjnr[jidx+1];
333 jnrlistC = jjnr[jidx+2];
334 jnrlistD = jjnr[jidx+3];
335 /* Sign of each element will be negative for non-real atoms.
336 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
337 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
339 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
340 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
341 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
342 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
343 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
344 j_coord_offsetA = DIM*jnrA;
345 j_coord_offsetB = DIM*jnrB;
346 j_coord_offsetC = DIM*jnrC;
347 j_coord_offsetD = DIM*jnrD;
349 /* load j atom coordinates */
350 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
351 x+j_coord_offsetC,x+j_coord_offsetD,
354 /* Calculate displacement vector */
355 dx00 = _mm_sub_ps(ix0,jx0);
356 dy00 = _mm_sub_ps(iy0,jy0);
357 dz00 = _mm_sub_ps(iz0,jz0);
358 dx10 = _mm_sub_ps(ix1,jx0);
359 dy10 = _mm_sub_ps(iy1,jy0);
360 dz10 = _mm_sub_ps(iz1,jz0);
361 dx20 = _mm_sub_ps(ix2,jx0);
362 dy20 = _mm_sub_ps(iy2,jy0);
363 dz20 = _mm_sub_ps(iz2,jz0);
365 /* Calculate squared distance and things based on it */
366 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
367 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
368 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
370 rinv00 = gmx_mm_invsqrt_ps(rsq00);
371 rinv10 = gmx_mm_invsqrt_ps(rsq10);
372 rinv20 = gmx_mm_invsqrt_ps(rsq20);
374 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
375 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
376 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
378 /* Load parameters for j particles */
379 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
380 charge+jnrC+0,charge+jnrD+0);
381 vdwjidx0A = 2*vdwtype[jnrA+0];
382 vdwjidx0B = 2*vdwtype[jnrB+0];
383 vdwjidx0C = 2*vdwtype[jnrC+0];
384 vdwjidx0D = 2*vdwtype[jnrD+0];
386 /**************************
387 * CALCULATE INTERACTIONS *
388 **************************/
390 /* Compute parameters for interactions between i and j atoms */
391 qq00 = _mm_mul_ps(iq0,jq0);
392 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
393 vdwparam+vdwioffset0+vdwjidx0B,
394 vdwparam+vdwioffset0+vdwjidx0C,
395 vdwparam+vdwioffset0+vdwjidx0D,
398 /* COULOMB ELECTROSTATICS */
399 velec = _mm_mul_ps(qq00,rinv00);
400 felec = _mm_mul_ps(velec,rinvsq00);
402 /* LENNARD-JONES DISPERSION/REPULSION */
404 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
405 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
406 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
407 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
408 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
410 /* Update potential sum for this i atom from the interaction with this j atom. */
411 velec = _mm_andnot_ps(dummy_mask,velec);
412 velecsum = _mm_add_ps(velecsum,velec);
413 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
414 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
416 fscal = _mm_add_ps(felec,fvdw);
418 fscal = _mm_andnot_ps(dummy_mask,fscal);
420 /* Calculate temporary vectorial force */
421 tx = _mm_mul_ps(fscal,dx00);
422 ty = _mm_mul_ps(fscal,dy00);
423 tz = _mm_mul_ps(fscal,dz00);
425 /* Update vectorial force */
426 fix0 = _mm_add_ps(fix0,tx);
427 fiy0 = _mm_add_ps(fiy0,ty);
428 fiz0 = _mm_add_ps(fiz0,tz);
430 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
431 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
432 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
433 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
434 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
436 /**************************
437 * CALCULATE INTERACTIONS *
438 **************************/
440 /* Compute parameters for interactions between i and j atoms */
441 qq10 = _mm_mul_ps(iq1,jq0);
443 /* COULOMB ELECTROSTATICS */
444 velec = _mm_mul_ps(qq10,rinv10);
445 felec = _mm_mul_ps(velec,rinvsq10);
447 /* Update potential sum for this i atom from the interaction with this j atom. */
448 velec = _mm_andnot_ps(dummy_mask,velec);
449 velecsum = _mm_add_ps(velecsum,velec);
453 fscal = _mm_andnot_ps(dummy_mask,fscal);
455 /* Calculate temporary vectorial force */
456 tx = _mm_mul_ps(fscal,dx10);
457 ty = _mm_mul_ps(fscal,dy10);
458 tz = _mm_mul_ps(fscal,dz10);
460 /* Update vectorial force */
461 fix1 = _mm_add_ps(fix1,tx);
462 fiy1 = _mm_add_ps(fiy1,ty);
463 fiz1 = _mm_add_ps(fiz1,tz);
465 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
466 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
467 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
468 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
469 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
471 /**************************
472 * CALCULATE INTERACTIONS *
473 **************************/
475 /* Compute parameters for interactions between i and j atoms */
476 qq20 = _mm_mul_ps(iq2,jq0);
478 /* COULOMB ELECTROSTATICS */
479 velec = _mm_mul_ps(qq20,rinv20);
480 felec = _mm_mul_ps(velec,rinvsq20);
482 /* Update potential sum for this i atom from the interaction with this j atom. */
483 velec = _mm_andnot_ps(dummy_mask,velec);
484 velecsum = _mm_add_ps(velecsum,velec);
488 fscal = _mm_andnot_ps(dummy_mask,fscal);
490 /* Calculate temporary vectorial force */
491 tx = _mm_mul_ps(fscal,dx20);
492 ty = _mm_mul_ps(fscal,dy20);
493 tz = _mm_mul_ps(fscal,dz20);
495 /* Update vectorial force */
496 fix2 = _mm_add_ps(fix2,tx);
497 fiy2 = _mm_add_ps(fiy2,ty);
498 fiz2 = _mm_add_ps(fiz2,tz);
500 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
501 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
502 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
503 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
504 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
506 /* Inner loop uses 96 flops */
509 /* End of innermost loop */
511 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
512 f+i_coord_offset,fshift+i_shift_offset);
515 /* Update potential energies */
516 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
517 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
519 /* Increment number of inner iterations */
520 inneriter += j_index_end - j_index_start;
522 /* Outer loop uses 20 flops */
525 /* Increment number of outer iterations */
528 /* Update outer/inner flops */
530 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
533 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_single
534 * Electrostatics interaction: Coulomb
535 * VdW interaction: LennardJones
536 * Geometry: Water3-Particle
537 * Calculate force/pot: Force
540 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_single
541 (t_nblist * gmx_restrict nlist,
542 rvec * gmx_restrict xx,
543 rvec * gmx_restrict ff,
544 t_forcerec * gmx_restrict fr,
545 t_mdatoms * gmx_restrict mdatoms,
546 nb_kernel_data_t * gmx_restrict kernel_data,
547 t_nrnb * gmx_restrict nrnb)
549 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
550 * just 0 for non-waters.
551 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
552 * jnr indices corresponding to data put in the four positions in the SIMD register.
554 int i_shift_offset,i_coord_offset,outeriter,inneriter;
555 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
556 int jnrA,jnrB,jnrC,jnrD;
557 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
558 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
559 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
561 real *shiftvec,*fshift,*x,*f;
562 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
564 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
566 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
568 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
570 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
571 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
572 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
573 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
574 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
575 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
576 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
579 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
582 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
583 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
584 __m128 dummy_mask,cutoff_mask;
585 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
586 __m128 one = _mm_set1_ps(1.0);
587 __m128 two = _mm_set1_ps(2.0);
593 jindex = nlist->jindex;
595 shiftidx = nlist->shift;
597 shiftvec = fr->shift_vec[0];
598 fshift = fr->fshift[0];
599 facel = _mm_set1_ps(fr->epsfac);
600 charge = mdatoms->chargeA;
601 nvdwtype = fr->ntype;
603 vdwtype = mdatoms->typeA;
605 /* Setup water-specific parameters */
606 inr = nlist->iinr[0];
607 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
608 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
609 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
610 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
612 /* Avoid stupid compiler warnings */
613 jnrA = jnrB = jnrC = jnrD = 0;
622 for(iidx=0;iidx<4*DIM;iidx++)
627 /* Start outer loop over neighborlists */
628 for(iidx=0; iidx<nri; iidx++)
630 /* Load shift vector for this list */
631 i_shift_offset = DIM*shiftidx[iidx];
633 /* Load limits for loop over neighbors */
634 j_index_start = jindex[iidx];
635 j_index_end = jindex[iidx+1];
637 /* Get outer coordinate index */
639 i_coord_offset = DIM*inr;
641 /* Load i particle coords and add shift vector */
642 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
643 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
645 fix0 = _mm_setzero_ps();
646 fiy0 = _mm_setzero_ps();
647 fiz0 = _mm_setzero_ps();
648 fix1 = _mm_setzero_ps();
649 fiy1 = _mm_setzero_ps();
650 fiz1 = _mm_setzero_ps();
651 fix2 = _mm_setzero_ps();
652 fiy2 = _mm_setzero_ps();
653 fiz2 = _mm_setzero_ps();
655 /* Start inner kernel loop */
656 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
659 /* Get j neighbor index, and coordinate index */
664 j_coord_offsetA = DIM*jnrA;
665 j_coord_offsetB = DIM*jnrB;
666 j_coord_offsetC = DIM*jnrC;
667 j_coord_offsetD = DIM*jnrD;
669 /* load j atom coordinates */
670 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
671 x+j_coord_offsetC,x+j_coord_offsetD,
674 /* Calculate displacement vector */
675 dx00 = _mm_sub_ps(ix0,jx0);
676 dy00 = _mm_sub_ps(iy0,jy0);
677 dz00 = _mm_sub_ps(iz0,jz0);
678 dx10 = _mm_sub_ps(ix1,jx0);
679 dy10 = _mm_sub_ps(iy1,jy0);
680 dz10 = _mm_sub_ps(iz1,jz0);
681 dx20 = _mm_sub_ps(ix2,jx0);
682 dy20 = _mm_sub_ps(iy2,jy0);
683 dz20 = _mm_sub_ps(iz2,jz0);
685 /* Calculate squared distance and things based on it */
686 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
687 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
688 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
690 rinv00 = gmx_mm_invsqrt_ps(rsq00);
691 rinv10 = gmx_mm_invsqrt_ps(rsq10);
692 rinv20 = gmx_mm_invsqrt_ps(rsq20);
694 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
695 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
696 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
698 /* Load parameters for j particles */
699 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
700 charge+jnrC+0,charge+jnrD+0);
701 vdwjidx0A = 2*vdwtype[jnrA+0];
702 vdwjidx0B = 2*vdwtype[jnrB+0];
703 vdwjidx0C = 2*vdwtype[jnrC+0];
704 vdwjidx0D = 2*vdwtype[jnrD+0];
706 /**************************
707 * CALCULATE INTERACTIONS *
708 **************************/
710 /* Compute parameters for interactions between i and j atoms */
711 qq00 = _mm_mul_ps(iq0,jq0);
712 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
713 vdwparam+vdwioffset0+vdwjidx0B,
714 vdwparam+vdwioffset0+vdwjidx0C,
715 vdwparam+vdwioffset0+vdwjidx0D,
718 /* COULOMB ELECTROSTATICS */
719 velec = _mm_mul_ps(qq00,rinv00);
720 felec = _mm_mul_ps(velec,rinvsq00);
722 /* LENNARD-JONES DISPERSION/REPULSION */
724 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
725 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
727 fscal = _mm_add_ps(felec,fvdw);
729 /* Calculate temporary vectorial force */
730 tx = _mm_mul_ps(fscal,dx00);
731 ty = _mm_mul_ps(fscal,dy00);
732 tz = _mm_mul_ps(fscal,dz00);
734 /* Update vectorial force */
735 fix0 = _mm_add_ps(fix0,tx);
736 fiy0 = _mm_add_ps(fiy0,ty);
737 fiz0 = _mm_add_ps(fiz0,tz);
739 fjptrA = f+j_coord_offsetA;
740 fjptrB = f+j_coord_offsetB;
741 fjptrC = f+j_coord_offsetC;
742 fjptrD = f+j_coord_offsetD;
743 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
745 /**************************
746 * CALCULATE INTERACTIONS *
747 **************************/
749 /* Compute parameters for interactions between i and j atoms */
750 qq10 = _mm_mul_ps(iq1,jq0);
752 /* COULOMB ELECTROSTATICS */
753 velec = _mm_mul_ps(qq10,rinv10);
754 felec = _mm_mul_ps(velec,rinvsq10);
758 /* Calculate temporary vectorial force */
759 tx = _mm_mul_ps(fscal,dx10);
760 ty = _mm_mul_ps(fscal,dy10);
761 tz = _mm_mul_ps(fscal,dz10);
763 /* Update vectorial force */
764 fix1 = _mm_add_ps(fix1,tx);
765 fiy1 = _mm_add_ps(fiy1,ty);
766 fiz1 = _mm_add_ps(fiz1,tz);
768 fjptrA = f+j_coord_offsetA;
769 fjptrB = f+j_coord_offsetB;
770 fjptrC = f+j_coord_offsetC;
771 fjptrD = f+j_coord_offsetD;
772 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
774 /**************************
775 * CALCULATE INTERACTIONS *
776 **************************/
778 /* Compute parameters for interactions between i and j atoms */
779 qq20 = _mm_mul_ps(iq2,jq0);
781 /* COULOMB ELECTROSTATICS */
782 velec = _mm_mul_ps(qq20,rinv20);
783 felec = _mm_mul_ps(velec,rinvsq20);
787 /* Calculate temporary vectorial force */
788 tx = _mm_mul_ps(fscal,dx20);
789 ty = _mm_mul_ps(fscal,dy20);
790 tz = _mm_mul_ps(fscal,dz20);
792 /* Update vectorial force */
793 fix2 = _mm_add_ps(fix2,tx);
794 fiy2 = _mm_add_ps(fiy2,ty);
795 fiz2 = _mm_add_ps(fiz2,tz);
797 fjptrA = f+j_coord_offsetA;
798 fjptrB = f+j_coord_offsetB;
799 fjptrC = f+j_coord_offsetC;
800 fjptrD = f+j_coord_offsetD;
801 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
803 /* Inner loop uses 88 flops */
809 /* Get j neighbor index, and coordinate index */
810 jnrlistA = jjnr[jidx];
811 jnrlistB = jjnr[jidx+1];
812 jnrlistC = jjnr[jidx+2];
813 jnrlistD = jjnr[jidx+3];
814 /* Sign of each element will be negative for non-real atoms.
815 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
816 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
818 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
819 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
820 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
821 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
822 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
823 j_coord_offsetA = DIM*jnrA;
824 j_coord_offsetB = DIM*jnrB;
825 j_coord_offsetC = DIM*jnrC;
826 j_coord_offsetD = DIM*jnrD;
828 /* load j atom coordinates */
829 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
830 x+j_coord_offsetC,x+j_coord_offsetD,
833 /* Calculate displacement vector */
834 dx00 = _mm_sub_ps(ix0,jx0);
835 dy00 = _mm_sub_ps(iy0,jy0);
836 dz00 = _mm_sub_ps(iz0,jz0);
837 dx10 = _mm_sub_ps(ix1,jx0);
838 dy10 = _mm_sub_ps(iy1,jy0);
839 dz10 = _mm_sub_ps(iz1,jz0);
840 dx20 = _mm_sub_ps(ix2,jx0);
841 dy20 = _mm_sub_ps(iy2,jy0);
842 dz20 = _mm_sub_ps(iz2,jz0);
844 /* Calculate squared distance and things based on it */
845 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
846 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
847 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
849 rinv00 = gmx_mm_invsqrt_ps(rsq00);
850 rinv10 = gmx_mm_invsqrt_ps(rsq10);
851 rinv20 = gmx_mm_invsqrt_ps(rsq20);
853 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
854 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
855 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
857 /* Load parameters for j particles */
858 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
859 charge+jnrC+0,charge+jnrD+0);
860 vdwjidx0A = 2*vdwtype[jnrA+0];
861 vdwjidx0B = 2*vdwtype[jnrB+0];
862 vdwjidx0C = 2*vdwtype[jnrC+0];
863 vdwjidx0D = 2*vdwtype[jnrD+0];
865 /**************************
866 * CALCULATE INTERACTIONS *
867 **************************/
869 /* Compute parameters for interactions between i and j atoms */
870 qq00 = _mm_mul_ps(iq0,jq0);
871 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
872 vdwparam+vdwioffset0+vdwjidx0B,
873 vdwparam+vdwioffset0+vdwjidx0C,
874 vdwparam+vdwioffset0+vdwjidx0D,
877 /* COULOMB ELECTROSTATICS */
878 velec = _mm_mul_ps(qq00,rinv00);
879 felec = _mm_mul_ps(velec,rinvsq00);
881 /* LENNARD-JONES DISPERSION/REPULSION */
883 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
884 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
886 fscal = _mm_add_ps(felec,fvdw);
888 fscal = _mm_andnot_ps(dummy_mask,fscal);
890 /* Calculate temporary vectorial force */
891 tx = _mm_mul_ps(fscal,dx00);
892 ty = _mm_mul_ps(fscal,dy00);
893 tz = _mm_mul_ps(fscal,dz00);
895 /* Update vectorial force */
896 fix0 = _mm_add_ps(fix0,tx);
897 fiy0 = _mm_add_ps(fiy0,ty);
898 fiz0 = _mm_add_ps(fiz0,tz);
900 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
901 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
902 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
903 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
904 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
906 /**************************
907 * CALCULATE INTERACTIONS *
908 **************************/
910 /* Compute parameters for interactions between i and j atoms */
911 qq10 = _mm_mul_ps(iq1,jq0);
913 /* COULOMB ELECTROSTATICS */
914 velec = _mm_mul_ps(qq10,rinv10);
915 felec = _mm_mul_ps(velec,rinvsq10);
919 fscal = _mm_andnot_ps(dummy_mask,fscal);
921 /* Calculate temporary vectorial force */
922 tx = _mm_mul_ps(fscal,dx10);
923 ty = _mm_mul_ps(fscal,dy10);
924 tz = _mm_mul_ps(fscal,dz10);
926 /* Update vectorial force */
927 fix1 = _mm_add_ps(fix1,tx);
928 fiy1 = _mm_add_ps(fiy1,ty);
929 fiz1 = _mm_add_ps(fiz1,tz);
931 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
932 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
933 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
934 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
935 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
937 /**************************
938 * CALCULATE INTERACTIONS *
939 **************************/
941 /* Compute parameters for interactions between i and j atoms */
942 qq20 = _mm_mul_ps(iq2,jq0);
944 /* COULOMB ELECTROSTATICS */
945 velec = _mm_mul_ps(qq20,rinv20);
946 felec = _mm_mul_ps(velec,rinvsq20);
950 fscal = _mm_andnot_ps(dummy_mask,fscal);
952 /* Calculate temporary vectorial force */
953 tx = _mm_mul_ps(fscal,dx20);
954 ty = _mm_mul_ps(fscal,dy20);
955 tz = _mm_mul_ps(fscal,dz20);
957 /* Update vectorial force */
958 fix2 = _mm_add_ps(fix2,tx);
959 fiy2 = _mm_add_ps(fiy2,ty);
960 fiz2 = _mm_add_ps(fiz2,tz);
962 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
963 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
964 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
965 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
966 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
968 /* Inner loop uses 88 flops */
971 /* End of innermost loop */
973 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
974 f+i_coord_offset,fshift+i_shift_offset);
976 /* Increment number of inner iterations */
977 inneriter += j_index_end - j_index_start;
979 /* Outer loop uses 18 flops */
982 /* Increment number of outer iterations */
985 /* Update outer/inner flops */
987 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);