2 * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
34 #include "kernelutil_x86_sse2_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_sse2_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: LennardJones
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_sse2_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 krf = _mm_set1_ps(fr->ic->k_rf);
106 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
107 crf = _mm_set1_ps(fr->ic->c_rf);
108 nvdwtype = fr->ntype;
110 vdwtype = mdatoms->typeA;
112 /* Setup water-specific parameters */
113 inr = nlist->iinr[0];
114 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
115 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
116 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
117 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
119 /* Avoid stupid compiler warnings */
120 jnrA = jnrB = jnrC = jnrD = 0;
129 for(iidx=0;iidx<4*DIM;iidx++)
134 /* Start outer loop over neighborlists */
135 for(iidx=0; iidx<nri; iidx++)
137 /* Load shift vector for this list */
138 i_shift_offset = DIM*shiftidx[iidx];
140 /* Load limits for loop over neighbors */
141 j_index_start = jindex[iidx];
142 j_index_end = jindex[iidx+1];
144 /* Get outer coordinate index */
146 i_coord_offset = DIM*inr;
148 /* Load i particle coords and add shift vector */
149 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
150 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
152 fix0 = _mm_setzero_ps();
153 fiy0 = _mm_setzero_ps();
154 fiz0 = _mm_setzero_ps();
155 fix1 = _mm_setzero_ps();
156 fiy1 = _mm_setzero_ps();
157 fiz1 = _mm_setzero_ps();
158 fix2 = _mm_setzero_ps();
159 fiy2 = _mm_setzero_ps();
160 fiz2 = _mm_setzero_ps();
162 /* Reset potential sums */
163 velecsum = _mm_setzero_ps();
164 vvdwsum = _mm_setzero_ps();
166 /* Start inner kernel loop */
167 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
170 /* Get j neighbor index, and coordinate index */
175 j_coord_offsetA = DIM*jnrA;
176 j_coord_offsetB = DIM*jnrB;
177 j_coord_offsetC = DIM*jnrC;
178 j_coord_offsetD = DIM*jnrD;
180 /* load j atom coordinates */
181 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
182 x+j_coord_offsetC,x+j_coord_offsetD,
185 /* Calculate displacement vector */
186 dx00 = _mm_sub_ps(ix0,jx0);
187 dy00 = _mm_sub_ps(iy0,jy0);
188 dz00 = _mm_sub_ps(iz0,jz0);
189 dx10 = _mm_sub_ps(ix1,jx0);
190 dy10 = _mm_sub_ps(iy1,jy0);
191 dz10 = _mm_sub_ps(iz1,jz0);
192 dx20 = _mm_sub_ps(ix2,jx0);
193 dy20 = _mm_sub_ps(iy2,jy0);
194 dz20 = _mm_sub_ps(iz2,jz0);
196 /* Calculate squared distance and things based on it */
197 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
198 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
199 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
201 rinv00 = gmx_mm_invsqrt_ps(rsq00);
202 rinv10 = gmx_mm_invsqrt_ps(rsq10);
203 rinv20 = gmx_mm_invsqrt_ps(rsq20);
205 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
206 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
207 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
209 /* Load parameters for j particles */
210 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
211 charge+jnrC+0,charge+jnrD+0);
212 vdwjidx0A = 2*vdwtype[jnrA+0];
213 vdwjidx0B = 2*vdwtype[jnrB+0];
214 vdwjidx0C = 2*vdwtype[jnrC+0];
215 vdwjidx0D = 2*vdwtype[jnrD+0];
217 /**************************
218 * CALCULATE INTERACTIONS *
219 **************************/
221 /* Compute parameters for interactions between i and j atoms */
222 qq00 = _mm_mul_ps(iq0,jq0);
223 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
224 vdwparam+vdwioffset0+vdwjidx0B,
225 vdwparam+vdwioffset0+vdwjidx0C,
226 vdwparam+vdwioffset0+vdwjidx0D,
229 /* REACTION-FIELD ELECTROSTATICS */
230 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
231 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
233 /* LENNARD-JONES DISPERSION/REPULSION */
235 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
236 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
237 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
238 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
239 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
241 /* Update potential sum for this i atom from the interaction with this j atom. */
242 velecsum = _mm_add_ps(velecsum,velec);
243 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
245 fscal = _mm_add_ps(felec,fvdw);
247 /* Calculate temporary vectorial force */
248 tx = _mm_mul_ps(fscal,dx00);
249 ty = _mm_mul_ps(fscal,dy00);
250 tz = _mm_mul_ps(fscal,dz00);
252 /* Update vectorial force */
253 fix0 = _mm_add_ps(fix0,tx);
254 fiy0 = _mm_add_ps(fiy0,ty);
255 fiz0 = _mm_add_ps(fiz0,tz);
257 fjptrA = f+j_coord_offsetA;
258 fjptrB = f+j_coord_offsetB;
259 fjptrC = f+j_coord_offsetC;
260 fjptrD = f+j_coord_offsetD;
261 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
263 /**************************
264 * CALCULATE INTERACTIONS *
265 **************************/
267 /* Compute parameters for interactions between i and j atoms */
268 qq10 = _mm_mul_ps(iq1,jq0);
270 /* REACTION-FIELD ELECTROSTATICS */
271 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
272 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
274 /* Update potential sum for this i atom from the interaction with this j atom. */
275 velecsum = _mm_add_ps(velecsum,velec);
279 /* Calculate temporary vectorial force */
280 tx = _mm_mul_ps(fscal,dx10);
281 ty = _mm_mul_ps(fscal,dy10);
282 tz = _mm_mul_ps(fscal,dz10);
284 /* Update vectorial force */
285 fix1 = _mm_add_ps(fix1,tx);
286 fiy1 = _mm_add_ps(fiy1,ty);
287 fiz1 = _mm_add_ps(fiz1,tz);
289 fjptrA = f+j_coord_offsetA;
290 fjptrB = f+j_coord_offsetB;
291 fjptrC = f+j_coord_offsetC;
292 fjptrD = f+j_coord_offsetD;
293 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
295 /**************************
296 * CALCULATE INTERACTIONS *
297 **************************/
299 /* Compute parameters for interactions between i and j atoms */
300 qq20 = _mm_mul_ps(iq2,jq0);
302 /* REACTION-FIELD ELECTROSTATICS */
303 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
304 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
306 /* Update potential sum for this i atom from the interaction with this j atom. */
307 velecsum = _mm_add_ps(velecsum,velec);
311 /* Calculate temporary vectorial force */
312 tx = _mm_mul_ps(fscal,dx20);
313 ty = _mm_mul_ps(fscal,dy20);
314 tz = _mm_mul_ps(fscal,dz20);
316 /* Update vectorial force */
317 fix2 = _mm_add_ps(fix2,tx);
318 fiy2 = _mm_add_ps(fiy2,ty);
319 fiz2 = _mm_add_ps(fiz2,tz);
321 fjptrA = f+j_coord_offsetA;
322 fjptrB = f+j_coord_offsetB;
323 fjptrC = f+j_coord_offsetC;
324 fjptrD = f+j_coord_offsetD;
325 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
327 /* Inner loop uses 108 flops */
333 /* Get j neighbor index, and coordinate index */
334 jnrlistA = jjnr[jidx];
335 jnrlistB = jjnr[jidx+1];
336 jnrlistC = jjnr[jidx+2];
337 jnrlistD = jjnr[jidx+3];
338 /* Sign of each element will be negative for non-real atoms.
339 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
340 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
342 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
343 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
344 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
345 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
346 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
347 j_coord_offsetA = DIM*jnrA;
348 j_coord_offsetB = DIM*jnrB;
349 j_coord_offsetC = DIM*jnrC;
350 j_coord_offsetD = DIM*jnrD;
352 /* load j atom coordinates */
353 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
354 x+j_coord_offsetC,x+j_coord_offsetD,
357 /* Calculate displacement vector */
358 dx00 = _mm_sub_ps(ix0,jx0);
359 dy00 = _mm_sub_ps(iy0,jy0);
360 dz00 = _mm_sub_ps(iz0,jz0);
361 dx10 = _mm_sub_ps(ix1,jx0);
362 dy10 = _mm_sub_ps(iy1,jy0);
363 dz10 = _mm_sub_ps(iz1,jz0);
364 dx20 = _mm_sub_ps(ix2,jx0);
365 dy20 = _mm_sub_ps(iy2,jy0);
366 dz20 = _mm_sub_ps(iz2,jz0);
368 /* Calculate squared distance and things based on it */
369 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
370 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
371 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
373 rinv00 = gmx_mm_invsqrt_ps(rsq00);
374 rinv10 = gmx_mm_invsqrt_ps(rsq10);
375 rinv20 = gmx_mm_invsqrt_ps(rsq20);
377 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
378 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
379 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
381 /* Load parameters for j particles */
382 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
383 charge+jnrC+0,charge+jnrD+0);
384 vdwjidx0A = 2*vdwtype[jnrA+0];
385 vdwjidx0B = 2*vdwtype[jnrB+0];
386 vdwjidx0C = 2*vdwtype[jnrC+0];
387 vdwjidx0D = 2*vdwtype[jnrD+0];
389 /**************************
390 * CALCULATE INTERACTIONS *
391 **************************/
393 /* Compute parameters for interactions between i and j atoms */
394 qq00 = _mm_mul_ps(iq0,jq0);
395 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
396 vdwparam+vdwioffset0+vdwjidx0B,
397 vdwparam+vdwioffset0+vdwjidx0C,
398 vdwparam+vdwioffset0+vdwjidx0D,
401 /* REACTION-FIELD ELECTROSTATICS */
402 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
403 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
405 /* LENNARD-JONES DISPERSION/REPULSION */
407 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
408 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
409 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
410 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
411 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
413 /* Update potential sum for this i atom from the interaction with this j atom. */
414 velec = _mm_andnot_ps(dummy_mask,velec);
415 velecsum = _mm_add_ps(velecsum,velec);
416 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
417 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
419 fscal = _mm_add_ps(felec,fvdw);
421 fscal = _mm_andnot_ps(dummy_mask,fscal);
423 /* Calculate temporary vectorial force */
424 tx = _mm_mul_ps(fscal,dx00);
425 ty = _mm_mul_ps(fscal,dy00);
426 tz = _mm_mul_ps(fscal,dz00);
428 /* Update vectorial force */
429 fix0 = _mm_add_ps(fix0,tx);
430 fiy0 = _mm_add_ps(fiy0,ty);
431 fiz0 = _mm_add_ps(fiz0,tz);
433 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
434 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
435 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
436 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
437 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
439 /**************************
440 * CALCULATE INTERACTIONS *
441 **************************/
443 /* Compute parameters for interactions between i and j atoms */
444 qq10 = _mm_mul_ps(iq1,jq0);
446 /* REACTION-FIELD ELECTROSTATICS */
447 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
448 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
450 /* Update potential sum for this i atom from the interaction with this j atom. */
451 velec = _mm_andnot_ps(dummy_mask,velec);
452 velecsum = _mm_add_ps(velecsum,velec);
456 fscal = _mm_andnot_ps(dummy_mask,fscal);
458 /* Calculate temporary vectorial force */
459 tx = _mm_mul_ps(fscal,dx10);
460 ty = _mm_mul_ps(fscal,dy10);
461 tz = _mm_mul_ps(fscal,dz10);
463 /* Update vectorial force */
464 fix1 = _mm_add_ps(fix1,tx);
465 fiy1 = _mm_add_ps(fiy1,ty);
466 fiz1 = _mm_add_ps(fiz1,tz);
468 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
469 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
470 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
471 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
472 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
474 /**************************
475 * CALCULATE INTERACTIONS *
476 **************************/
478 /* Compute parameters for interactions between i and j atoms */
479 qq20 = _mm_mul_ps(iq2,jq0);
481 /* REACTION-FIELD ELECTROSTATICS */
482 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
483 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
485 /* Update potential sum for this i atom from the interaction with this j atom. */
486 velec = _mm_andnot_ps(dummy_mask,velec);
487 velecsum = _mm_add_ps(velecsum,velec);
491 fscal = _mm_andnot_ps(dummy_mask,fscal);
493 /* Calculate temporary vectorial force */
494 tx = _mm_mul_ps(fscal,dx20);
495 ty = _mm_mul_ps(fscal,dy20);
496 tz = _mm_mul_ps(fscal,dz20);
498 /* Update vectorial force */
499 fix2 = _mm_add_ps(fix2,tx);
500 fiy2 = _mm_add_ps(fiy2,ty);
501 fiz2 = _mm_add_ps(fiz2,tz);
503 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
504 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
505 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
506 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
507 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
509 /* Inner loop uses 108 flops */
512 /* End of innermost loop */
514 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
515 f+i_coord_offset,fshift+i_shift_offset);
518 /* Update potential energies */
519 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
520 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
522 /* Increment number of inner iterations */
523 inneriter += j_index_end - j_index_start;
525 /* Outer loop uses 20 flops */
528 /* Increment number of outer iterations */
531 /* Update outer/inner flops */
533 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*108);
536 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_sse2_single
537 * Electrostatics interaction: ReactionField
538 * VdW interaction: LennardJones
539 * Geometry: Water3-Particle
540 * Calculate force/pot: Force
543 nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_sse2_single
544 (t_nblist * gmx_restrict nlist,
545 rvec * gmx_restrict xx,
546 rvec * gmx_restrict ff,
547 t_forcerec * gmx_restrict fr,
548 t_mdatoms * gmx_restrict mdatoms,
549 nb_kernel_data_t * gmx_restrict kernel_data,
550 t_nrnb * gmx_restrict nrnb)
552 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
553 * just 0 for non-waters.
554 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
555 * jnr indices corresponding to data put in the four positions in the SIMD register.
557 int i_shift_offset,i_coord_offset,outeriter,inneriter;
558 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
559 int jnrA,jnrB,jnrC,jnrD;
560 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
561 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
562 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
564 real *shiftvec,*fshift,*x,*f;
565 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
567 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
569 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
571 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
573 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
574 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
575 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
576 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
577 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
578 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
579 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
582 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
585 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
586 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
587 __m128 dummy_mask,cutoff_mask;
588 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
589 __m128 one = _mm_set1_ps(1.0);
590 __m128 two = _mm_set1_ps(2.0);
596 jindex = nlist->jindex;
598 shiftidx = nlist->shift;
600 shiftvec = fr->shift_vec[0];
601 fshift = fr->fshift[0];
602 facel = _mm_set1_ps(fr->epsfac);
603 charge = mdatoms->chargeA;
604 krf = _mm_set1_ps(fr->ic->k_rf);
605 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
606 crf = _mm_set1_ps(fr->ic->c_rf);
607 nvdwtype = fr->ntype;
609 vdwtype = mdatoms->typeA;
611 /* Setup water-specific parameters */
612 inr = nlist->iinr[0];
613 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
614 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
615 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
616 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
618 /* Avoid stupid compiler warnings */
619 jnrA = jnrB = jnrC = jnrD = 0;
628 for(iidx=0;iidx<4*DIM;iidx++)
633 /* Start outer loop over neighborlists */
634 for(iidx=0; iidx<nri; iidx++)
636 /* Load shift vector for this list */
637 i_shift_offset = DIM*shiftidx[iidx];
639 /* Load limits for loop over neighbors */
640 j_index_start = jindex[iidx];
641 j_index_end = jindex[iidx+1];
643 /* Get outer coordinate index */
645 i_coord_offset = DIM*inr;
647 /* Load i particle coords and add shift vector */
648 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
649 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
651 fix0 = _mm_setzero_ps();
652 fiy0 = _mm_setzero_ps();
653 fiz0 = _mm_setzero_ps();
654 fix1 = _mm_setzero_ps();
655 fiy1 = _mm_setzero_ps();
656 fiz1 = _mm_setzero_ps();
657 fix2 = _mm_setzero_ps();
658 fiy2 = _mm_setzero_ps();
659 fiz2 = _mm_setzero_ps();
661 /* Start inner kernel loop */
662 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
665 /* Get j neighbor index, and coordinate index */
670 j_coord_offsetA = DIM*jnrA;
671 j_coord_offsetB = DIM*jnrB;
672 j_coord_offsetC = DIM*jnrC;
673 j_coord_offsetD = DIM*jnrD;
675 /* load j atom coordinates */
676 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
677 x+j_coord_offsetC,x+j_coord_offsetD,
680 /* Calculate displacement vector */
681 dx00 = _mm_sub_ps(ix0,jx0);
682 dy00 = _mm_sub_ps(iy0,jy0);
683 dz00 = _mm_sub_ps(iz0,jz0);
684 dx10 = _mm_sub_ps(ix1,jx0);
685 dy10 = _mm_sub_ps(iy1,jy0);
686 dz10 = _mm_sub_ps(iz1,jz0);
687 dx20 = _mm_sub_ps(ix2,jx0);
688 dy20 = _mm_sub_ps(iy2,jy0);
689 dz20 = _mm_sub_ps(iz2,jz0);
691 /* Calculate squared distance and things based on it */
692 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
693 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
694 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
696 rinv00 = gmx_mm_invsqrt_ps(rsq00);
697 rinv10 = gmx_mm_invsqrt_ps(rsq10);
698 rinv20 = gmx_mm_invsqrt_ps(rsq20);
700 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
701 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
702 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
704 /* Load parameters for j particles */
705 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
706 charge+jnrC+0,charge+jnrD+0);
707 vdwjidx0A = 2*vdwtype[jnrA+0];
708 vdwjidx0B = 2*vdwtype[jnrB+0];
709 vdwjidx0C = 2*vdwtype[jnrC+0];
710 vdwjidx0D = 2*vdwtype[jnrD+0];
712 /**************************
713 * CALCULATE INTERACTIONS *
714 **************************/
716 /* Compute parameters for interactions between i and j atoms */
717 qq00 = _mm_mul_ps(iq0,jq0);
718 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
719 vdwparam+vdwioffset0+vdwjidx0B,
720 vdwparam+vdwioffset0+vdwjidx0C,
721 vdwparam+vdwioffset0+vdwjidx0D,
724 /* REACTION-FIELD ELECTROSTATICS */
725 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
727 /* LENNARD-JONES DISPERSION/REPULSION */
729 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
730 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
732 fscal = _mm_add_ps(felec,fvdw);
734 /* Calculate temporary vectorial force */
735 tx = _mm_mul_ps(fscal,dx00);
736 ty = _mm_mul_ps(fscal,dy00);
737 tz = _mm_mul_ps(fscal,dz00);
739 /* Update vectorial force */
740 fix0 = _mm_add_ps(fix0,tx);
741 fiy0 = _mm_add_ps(fiy0,ty);
742 fiz0 = _mm_add_ps(fiz0,tz);
744 fjptrA = f+j_coord_offsetA;
745 fjptrB = f+j_coord_offsetB;
746 fjptrC = f+j_coord_offsetC;
747 fjptrD = f+j_coord_offsetD;
748 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
750 /**************************
751 * CALCULATE INTERACTIONS *
752 **************************/
754 /* Compute parameters for interactions between i and j atoms */
755 qq10 = _mm_mul_ps(iq1,jq0);
757 /* REACTION-FIELD ELECTROSTATICS */
758 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
762 /* Calculate temporary vectorial force */
763 tx = _mm_mul_ps(fscal,dx10);
764 ty = _mm_mul_ps(fscal,dy10);
765 tz = _mm_mul_ps(fscal,dz10);
767 /* Update vectorial force */
768 fix1 = _mm_add_ps(fix1,tx);
769 fiy1 = _mm_add_ps(fiy1,ty);
770 fiz1 = _mm_add_ps(fiz1,tz);
772 fjptrA = f+j_coord_offsetA;
773 fjptrB = f+j_coord_offsetB;
774 fjptrC = f+j_coord_offsetC;
775 fjptrD = f+j_coord_offsetD;
776 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
778 /**************************
779 * CALCULATE INTERACTIONS *
780 **************************/
782 /* Compute parameters for interactions between i and j atoms */
783 qq20 = _mm_mul_ps(iq2,jq0);
785 /* REACTION-FIELD ELECTROSTATICS */
786 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
790 /* Calculate temporary vectorial force */
791 tx = _mm_mul_ps(fscal,dx20);
792 ty = _mm_mul_ps(fscal,dy20);
793 tz = _mm_mul_ps(fscal,dz20);
795 /* Update vectorial force */
796 fix2 = _mm_add_ps(fix2,tx);
797 fiy2 = _mm_add_ps(fiy2,ty);
798 fiz2 = _mm_add_ps(fiz2,tz);
800 fjptrA = f+j_coord_offsetA;
801 fjptrB = f+j_coord_offsetB;
802 fjptrC = f+j_coord_offsetC;
803 fjptrD = f+j_coord_offsetD;
804 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
806 /* Inner loop uses 88 flops */
812 /* Get j neighbor index, and coordinate index */
813 jnrlistA = jjnr[jidx];
814 jnrlistB = jjnr[jidx+1];
815 jnrlistC = jjnr[jidx+2];
816 jnrlistD = jjnr[jidx+3];
817 /* Sign of each element will be negative for non-real atoms.
818 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
819 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
821 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
822 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
823 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
824 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
825 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
826 j_coord_offsetA = DIM*jnrA;
827 j_coord_offsetB = DIM*jnrB;
828 j_coord_offsetC = DIM*jnrC;
829 j_coord_offsetD = DIM*jnrD;
831 /* load j atom coordinates */
832 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
833 x+j_coord_offsetC,x+j_coord_offsetD,
836 /* Calculate displacement vector */
837 dx00 = _mm_sub_ps(ix0,jx0);
838 dy00 = _mm_sub_ps(iy0,jy0);
839 dz00 = _mm_sub_ps(iz0,jz0);
840 dx10 = _mm_sub_ps(ix1,jx0);
841 dy10 = _mm_sub_ps(iy1,jy0);
842 dz10 = _mm_sub_ps(iz1,jz0);
843 dx20 = _mm_sub_ps(ix2,jx0);
844 dy20 = _mm_sub_ps(iy2,jy0);
845 dz20 = _mm_sub_ps(iz2,jz0);
847 /* Calculate squared distance and things based on it */
848 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
849 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
850 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
852 rinv00 = gmx_mm_invsqrt_ps(rsq00);
853 rinv10 = gmx_mm_invsqrt_ps(rsq10);
854 rinv20 = gmx_mm_invsqrt_ps(rsq20);
856 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
857 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
858 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
860 /* Load parameters for j particles */
861 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
862 charge+jnrC+0,charge+jnrD+0);
863 vdwjidx0A = 2*vdwtype[jnrA+0];
864 vdwjidx0B = 2*vdwtype[jnrB+0];
865 vdwjidx0C = 2*vdwtype[jnrC+0];
866 vdwjidx0D = 2*vdwtype[jnrD+0];
868 /**************************
869 * CALCULATE INTERACTIONS *
870 **************************/
872 /* Compute parameters for interactions between i and j atoms */
873 qq00 = _mm_mul_ps(iq0,jq0);
874 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
875 vdwparam+vdwioffset0+vdwjidx0B,
876 vdwparam+vdwioffset0+vdwjidx0C,
877 vdwparam+vdwioffset0+vdwjidx0D,
880 /* REACTION-FIELD ELECTROSTATICS */
881 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
883 /* LENNARD-JONES DISPERSION/REPULSION */
885 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
886 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
888 fscal = _mm_add_ps(felec,fvdw);
890 fscal = _mm_andnot_ps(dummy_mask,fscal);
892 /* Calculate temporary vectorial force */
893 tx = _mm_mul_ps(fscal,dx00);
894 ty = _mm_mul_ps(fscal,dy00);
895 tz = _mm_mul_ps(fscal,dz00);
897 /* Update vectorial force */
898 fix0 = _mm_add_ps(fix0,tx);
899 fiy0 = _mm_add_ps(fiy0,ty);
900 fiz0 = _mm_add_ps(fiz0,tz);
902 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
903 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
904 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
905 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
906 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
908 /**************************
909 * CALCULATE INTERACTIONS *
910 **************************/
912 /* Compute parameters for interactions between i and j atoms */
913 qq10 = _mm_mul_ps(iq1,jq0);
915 /* REACTION-FIELD ELECTROSTATICS */
916 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
920 fscal = _mm_andnot_ps(dummy_mask,fscal);
922 /* Calculate temporary vectorial force */
923 tx = _mm_mul_ps(fscal,dx10);
924 ty = _mm_mul_ps(fscal,dy10);
925 tz = _mm_mul_ps(fscal,dz10);
927 /* Update vectorial force */
928 fix1 = _mm_add_ps(fix1,tx);
929 fiy1 = _mm_add_ps(fiy1,ty);
930 fiz1 = _mm_add_ps(fiz1,tz);
932 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
933 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
934 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
935 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
936 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
938 /**************************
939 * CALCULATE INTERACTIONS *
940 **************************/
942 /* Compute parameters for interactions between i and j atoms */
943 qq20 = _mm_mul_ps(iq2,jq0);
945 /* REACTION-FIELD ELECTROSTATICS */
946 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
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);