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_ElecCoul_VdwLJ_GeomW4P1_VF_sse2_single
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: LennardJones
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_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 j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
63 real shX,shY,shZ,rcutoff_scalar;
64 real *shiftvec,*fshift,*x,*f;
65 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
69 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
71 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
73 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
74 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
75 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
76 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
77 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
78 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
79 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
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 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
112 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
113 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
114 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
116 /* Avoid stupid compiler warnings */
117 jnrA = jnrB = jnrC = jnrD = 0;
126 /* Start outer loop over neighborlists */
127 for(iidx=0; iidx<nri; iidx++)
129 /* Load shift vector for this list */
130 i_shift_offset = DIM*shiftidx[iidx];
131 shX = shiftvec[i_shift_offset+XX];
132 shY = shiftvec[i_shift_offset+YY];
133 shZ = shiftvec[i_shift_offset+ZZ];
135 /* Load limits for loop over neighbors */
136 j_index_start = jindex[iidx];
137 j_index_end = jindex[iidx+1];
139 /* Get outer coordinate index */
141 i_coord_offset = DIM*inr;
143 /* Load i particle coords and add shift vector */
144 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
145 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
146 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
147 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
148 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
149 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
150 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
151 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
152 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
153 ix3 = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
154 iy3 = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
155 iz3 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
157 fix0 = _mm_setzero_ps();
158 fiy0 = _mm_setzero_ps();
159 fiz0 = _mm_setzero_ps();
160 fix1 = _mm_setzero_ps();
161 fiy1 = _mm_setzero_ps();
162 fiz1 = _mm_setzero_ps();
163 fix2 = _mm_setzero_ps();
164 fiy2 = _mm_setzero_ps();
165 fiz2 = _mm_setzero_ps();
166 fix3 = _mm_setzero_ps();
167 fiy3 = _mm_setzero_ps();
168 fiz3 = _mm_setzero_ps();
170 /* Reset potential sums */
171 velecsum = _mm_setzero_ps();
172 vvdwsum = _mm_setzero_ps();
174 /* Start inner kernel loop */
175 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
178 /* Get j neighbor index, and coordinate index */
184 j_coord_offsetA = DIM*jnrA;
185 j_coord_offsetB = DIM*jnrB;
186 j_coord_offsetC = DIM*jnrC;
187 j_coord_offsetD = DIM*jnrD;
189 /* load j atom coordinates */
190 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
191 x+j_coord_offsetC,x+j_coord_offsetD,
194 /* Calculate displacement vector */
195 dx00 = _mm_sub_ps(ix0,jx0);
196 dy00 = _mm_sub_ps(iy0,jy0);
197 dz00 = _mm_sub_ps(iz0,jz0);
198 dx10 = _mm_sub_ps(ix1,jx0);
199 dy10 = _mm_sub_ps(iy1,jy0);
200 dz10 = _mm_sub_ps(iz1,jz0);
201 dx20 = _mm_sub_ps(ix2,jx0);
202 dy20 = _mm_sub_ps(iy2,jy0);
203 dz20 = _mm_sub_ps(iz2,jz0);
204 dx30 = _mm_sub_ps(ix3,jx0);
205 dy30 = _mm_sub_ps(iy3,jy0);
206 dz30 = _mm_sub_ps(iz3,jz0);
208 /* Calculate squared distance and things based on it */
209 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
210 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
211 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
212 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
214 rinv10 = gmx_mm_invsqrt_ps(rsq10);
215 rinv20 = gmx_mm_invsqrt_ps(rsq20);
216 rinv30 = gmx_mm_invsqrt_ps(rsq30);
218 rinvsq00 = gmx_mm_inv_ps(rsq00);
219 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
220 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
221 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
223 /* Load parameters for j particles */
224 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
225 charge+jnrC+0,charge+jnrD+0);
226 vdwjidx0A = 2*vdwtype[jnrA+0];
227 vdwjidx0B = 2*vdwtype[jnrB+0];
228 vdwjidx0C = 2*vdwtype[jnrC+0];
229 vdwjidx0D = 2*vdwtype[jnrD+0];
231 /**************************
232 * CALCULATE INTERACTIONS *
233 **************************/
235 /* Compute parameters for interactions between i and j atoms */
236 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
237 vdwparam+vdwioffset0+vdwjidx0B,
238 vdwparam+vdwioffset0+vdwjidx0C,
239 vdwparam+vdwioffset0+vdwjidx0D,
242 /* LENNARD-JONES DISPERSION/REPULSION */
244 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
245 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
246 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
247 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
248 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
250 /* Update potential sum for this i atom from the interaction with this j atom. */
251 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
255 /* Calculate temporary vectorial force */
256 tx = _mm_mul_ps(fscal,dx00);
257 ty = _mm_mul_ps(fscal,dy00);
258 tz = _mm_mul_ps(fscal,dz00);
260 /* Update vectorial force */
261 fix0 = _mm_add_ps(fix0,tx);
262 fiy0 = _mm_add_ps(fiy0,ty);
263 fiz0 = _mm_add_ps(fiz0,tz);
265 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
266 f+j_coord_offsetC,f+j_coord_offsetD,
269 /**************************
270 * CALCULATE INTERACTIONS *
271 **************************/
273 /* Compute parameters for interactions between i and j atoms */
274 qq10 = _mm_mul_ps(iq1,jq0);
276 /* COULOMB ELECTROSTATICS */
277 velec = _mm_mul_ps(qq10,rinv10);
278 felec = _mm_mul_ps(velec,rinvsq10);
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 velecsum = _mm_add_ps(velecsum,velec);
285 /* Calculate temporary vectorial force */
286 tx = _mm_mul_ps(fscal,dx10);
287 ty = _mm_mul_ps(fscal,dy10);
288 tz = _mm_mul_ps(fscal,dz10);
290 /* Update vectorial force */
291 fix1 = _mm_add_ps(fix1,tx);
292 fiy1 = _mm_add_ps(fiy1,ty);
293 fiz1 = _mm_add_ps(fiz1,tz);
295 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
296 f+j_coord_offsetC,f+j_coord_offsetD,
299 /**************************
300 * CALCULATE INTERACTIONS *
301 **************************/
303 /* Compute parameters for interactions between i and j atoms */
304 qq20 = _mm_mul_ps(iq2,jq0);
306 /* COULOMB ELECTROSTATICS */
307 velec = _mm_mul_ps(qq20,rinv20);
308 felec = _mm_mul_ps(velec,rinvsq20);
310 /* Update potential sum for this i atom from the interaction with this j atom. */
311 velecsum = _mm_add_ps(velecsum,velec);
315 /* Calculate temporary vectorial force */
316 tx = _mm_mul_ps(fscal,dx20);
317 ty = _mm_mul_ps(fscal,dy20);
318 tz = _mm_mul_ps(fscal,dz20);
320 /* Update vectorial force */
321 fix2 = _mm_add_ps(fix2,tx);
322 fiy2 = _mm_add_ps(fiy2,ty);
323 fiz2 = _mm_add_ps(fiz2,tz);
325 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
326 f+j_coord_offsetC,f+j_coord_offsetD,
329 /**************************
330 * CALCULATE INTERACTIONS *
331 **************************/
333 /* Compute parameters for interactions between i and j atoms */
334 qq30 = _mm_mul_ps(iq3,jq0);
336 /* COULOMB ELECTROSTATICS */
337 velec = _mm_mul_ps(qq30,rinv30);
338 felec = _mm_mul_ps(velec,rinvsq30);
340 /* Update potential sum for this i atom from the interaction with this j atom. */
341 velecsum = _mm_add_ps(velecsum,velec);
345 /* Calculate temporary vectorial force */
346 tx = _mm_mul_ps(fscal,dx30);
347 ty = _mm_mul_ps(fscal,dy30);
348 tz = _mm_mul_ps(fscal,dz30);
350 /* Update vectorial force */
351 fix3 = _mm_add_ps(fix3,tx);
352 fiy3 = _mm_add_ps(fiy3,ty);
353 fiz3 = _mm_add_ps(fiz3,tz);
355 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
356 f+j_coord_offsetC,f+j_coord_offsetD,
359 /* Inner loop uses 116 flops */
365 /* Get j neighbor index, and coordinate index */
371 /* Sign of each element will be negative for non-real atoms.
372 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
373 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
375 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
376 jnrA = (jnrA>=0) ? jnrA : 0;
377 jnrB = (jnrB>=0) ? jnrB : 0;
378 jnrC = (jnrC>=0) ? jnrC : 0;
379 jnrD = (jnrD>=0) ? jnrD : 0;
381 j_coord_offsetA = DIM*jnrA;
382 j_coord_offsetB = DIM*jnrB;
383 j_coord_offsetC = DIM*jnrC;
384 j_coord_offsetD = DIM*jnrD;
386 /* load j atom coordinates */
387 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
388 x+j_coord_offsetC,x+j_coord_offsetD,
391 /* Calculate displacement vector */
392 dx00 = _mm_sub_ps(ix0,jx0);
393 dy00 = _mm_sub_ps(iy0,jy0);
394 dz00 = _mm_sub_ps(iz0,jz0);
395 dx10 = _mm_sub_ps(ix1,jx0);
396 dy10 = _mm_sub_ps(iy1,jy0);
397 dz10 = _mm_sub_ps(iz1,jz0);
398 dx20 = _mm_sub_ps(ix2,jx0);
399 dy20 = _mm_sub_ps(iy2,jy0);
400 dz20 = _mm_sub_ps(iz2,jz0);
401 dx30 = _mm_sub_ps(ix3,jx0);
402 dy30 = _mm_sub_ps(iy3,jy0);
403 dz30 = _mm_sub_ps(iz3,jz0);
405 /* Calculate squared distance and things based on it */
406 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
407 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
408 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
409 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
411 rinv10 = gmx_mm_invsqrt_ps(rsq10);
412 rinv20 = gmx_mm_invsqrt_ps(rsq20);
413 rinv30 = gmx_mm_invsqrt_ps(rsq30);
415 rinvsq00 = gmx_mm_inv_ps(rsq00);
416 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
417 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
418 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
420 /* Load parameters for j particles */
421 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
422 charge+jnrC+0,charge+jnrD+0);
423 vdwjidx0A = 2*vdwtype[jnrA+0];
424 vdwjidx0B = 2*vdwtype[jnrB+0];
425 vdwjidx0C = 2*vdwtype[jnrC+0];
426 vdwjidx0D = 2*vdwtype[jnrD+0];
428 /**************************
429 * CALCULATE INTERACTIONS *
430 **************************/
432 /* Compute parameters for interactions between i and j atoms */
433 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
434 vdwparam+vdwioffset0+vdwjidx0B,
435 vdwparam+vdwioffset0+vdwjidx0C,
436 vdwparam+vdwioffset0+vdwjidx0D,
439 /* LENNARD-JONES DISPERSION/REPULSION */
441 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
442 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
443 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
444 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
445 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
447 /* Update potential sum for this i atom from the interaction with this j atom. */
448 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
449 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
453 fscal = _mm_andnot_ps(dummy_mask,fscal);
455 /* Calculate temporary vectorial force */
456 tx = _mm_mul_ps(fscal,dx00);
457 ty = _mm_mul_ps(fscal,dy00);
458 tz = _mm_mul_ps(fscal,dz00);
460 /* Update vectorial force */
461 fix0 = _mm_add_ps(fix0,tx);
462 fiy0 = _mm_add_ps(fiy0,ty);
463 fiz0 = _mm_add_ps(fiz0,tz);
465 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
466 f+j_coord_offsetC,f+j_coord_offsetD,
469 /**************************
470 * CALCULATE INTERACTIONS *
471 **************************/
473 /* Compute parameters for interactions between i and j atoms */
474 qq10 = _mm_mul_ps(iq1,jq0);
476 /* COULOMB ELECTROSTATICS */
477 velec = _mm_mul_ps(qq10,rinv10);
478 felec = _mm_mul_ps(velec,rinvsq10);
480 /* Update potential sum for this i atom from the interaction with this j atom. */
481 velec = _mm_andnot_ps(dummy_mask,velec);
482 velecsum = _mm_add_ps(velecsum,velec);
486 fscal = _mm_andnot_ps(dummy_mask,fscal);
488 /* Calculate temporary vectorial force */
489 tx = _mm_mul_ps(fscal,dx10);
490 ty = _mm_mul_ps(fscal,dy10);
491 tz = _mm_mul_ps(fscal,dz10);
493 /* Update vectorial force */
494 fix1 = _mm_add_ps(fix1,tx);
495 fiy1 = _mm_add_ps(fiy1,ty);
496 fiz1 = _mm_add_ps(fiz1,tz);
498 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
499 f+j_coord_offsetC,f+j_coord_offsetD,
502 /**************************
503 * CALCULATE INTERACTIONS *
504 **************************/
506 /* Compute parameters for interactions between i and j atoms */
507 qq20 = _mm_mul_ps(iq2,jq0);
509 /* COULOMB ELECTROSTATICS */
510 velec = _mm_mul_ps(qq20,rinv20);
511 felec = _mm_mul_ps(velec,rinvsq20);
513 /* Update potential sum for this i atom from the interaction with this j atom. */
514 velec = _mm_andnot_ps(dummy_mask,velec);
515 velecsum = _mm_add_ps(velecsum,velec);
519 fscal = _mm_andnot_ps(dummy_mask,fscal);
521 /* Calculate temporary vectorial force */
522 tx = _mm_mul_ps(fscal,dx20);
523 ty = _mm_mul_ps(fscal,dy20);
524 tz = _mm_mul_ps(fscal,dz20);
526 /* Update vectorial force */
527 fix2 = _mm_add_ps(fix2,tx);
528 fiy2 = _mm_add_ps(fiy2,ty);
529 fiz2 = _mm_add_ps(fiz2,tz);
531 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
532 f+j_coord_offsetC,f+j_coord_offsetD,
535 /**************************
536 * CALCULATE INTERACTIONS *
537 **************************/
539 /* Compute parameters for interactions between i and j atoms */
540 qq30 = _mm_mul_ps(iq3,jq0);
542 /* COULOMB ELECTROSTATICS */
543 velec = _mm_mul_ps(qq30,rinv30);
544 felec = _mm_mul_ps(velec,rinvsq30);
546 /* Update potential sum for this i atom from the interaction with this j atom. */
547 velec = _mm_andnot_ps(dummy_mask,velec);
548 velecsum = _mm_add_ps(velecsum,velec);
552 fscal = _mm_andnot_ps(dummy_mask,fscal);
554 /* Calculate temporary vectorial force */
555 tx = _mm_mul_ps(fscal,dx30);
556 ty = _mm_mul_ps(fscal,dy30);
557 tz = _mm_mul_ps(fscal,dz30);
559 /* Update vectorial force */
560 fix3 = _mm_add_ps(fix3,tx);
561 fiy3 = _mm_add_ps(fiy3,ty);
562 fiz3 = _mm_add_ps(fiz3,tz);
564 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
565 f+j_coord_offsetC,f+j_coord_offsetD,
568 /* Inner loop uses 116 flops */
571 /* End of innermost loop */
573 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
574 f+i_coord_offset,fshift+i_shift_offset);
577 /* Update potential energies */
578 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
579 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
581 /* Increment number of inner iterations */
582 inneriter += j_index_end - j_index_start;
584 /* Outer loop uses 38 flops */
587 /* Increment number of outer iterations */
590 /* Update outer/inner flops */
592 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*38 + inneriter*116);
595 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_sse2_single
596 * Electrostatics interaction: Coulomb
597 * VdW interaction: LennardJones
598 * Geometry: Water4-Particle
599 * Calculate force/pot: Force
602 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_sse2_single
603 (t_nblist * gmx_restrict nlist,
604 rvec * gmx_restrict xx,
605 rvec * gmx_restrict ff,
606 t_forcerec * gmx_restrict fr,
607 t_mdatoms * gmx_restrict mdatoms,
608 nb_kernel_data_t * gmx_restrict kernel_data,
609 t_nrnb * gmx_restrict nrnb)
611 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
612 * just 0 for non-waters.
613 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
614 * jnr indices corresponding to data put in the four positions in the SIMD register.
616 int i_shift_offset,i_coord_offset,outeriter,inneriter;
617 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
618 int jnrA,jnrB,jnrC,jnrD;
619 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
620 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
621 real shX,shY,shZ,rcutoff_scalar;
622 real *shiftvec,*fshift,*x,*f;
623 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
625 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
627 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
629 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
631 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
632 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
633 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
634 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
635 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
636 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
637 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
638 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
641 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
644 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
645 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
646 __m128 dummy_mask,cutoff_mask;
647 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
648 __m128 one = _mm_set1_ps(1.0);
649 __m128 two = _mm_set1_ps(2.0);
655 jindex = nlist->jindex;
657 shiftidx = nlist->shift;
659 shiftvec = fr->shift_vec[0];
660 fshift = fr->fshift[0];
661 facel = _mm_set1_ps(fr->epsfac);
662 charge = mdatoms->chargeA;
663 nvdwtype = fr->ntype;
665 vdwtype = mdatoms->typeA;
667 /* Setup water-specific parameters */
668 inr = nlist->iinr[0];
669 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
670 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
671 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
672 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
674 /* Avoid stupid compiler warnings */
675 jnrA = jnrB = jnrC = jnrD = 0;
684 /* Start outer loop over neighborlists */
685 for(iidx=0; iidx<nri; iidx++)
687 /* Load shift vector for this list */
688 i_shift_offset = DIM*shiftidx[iidx];
689 shX = shiftvec[i_shift_offset+XX];
690 shY = shiftvec[i_shift_offset+YY];
691 shZ = shiftvec[i_shift_offset+ZZ];
693 /* Load limits for loop over neighbors */
694 j_index_start = jindex[iidx];
695 j_index_end = jindex[iidx+1];
697 /* Get outer coordinate index */
699 i_coord_offset = DIM*inr;
701 /* Load i particle coords and add shift vector */
702 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
703 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
704 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
705 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
706 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
707 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
708 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
709 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
710 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
711 ix3 = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
712 iy3 = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
713 iz3 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
715 fix0 = _mm_setzero_ps();
716 fiy0 = _mm_setzero_ps();
717 fiz0 = _mm_setzero_ps();
718 fix1 = _mm_setzero_ps();
719 fiy1 = _mm_setzero_ps();
720 fiz1 = _mm_setzero_ps();
721 fix2 = _mm_setzero_ps();
722 fiy2 = _mm_setzero_ps();
723 fiz2 = _mm_setzero_ps();
724 fix3 = _mm_setzero_ps();
725 fiy3 = _mm_setzero_ps();
726 fiz3 = _mm_setzero_ps();
728 /* Start inner kernel loop */
729 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
732 /* Get j neighbor index, and coordinate index */
738 j_coord_offsetA = DIM*jnrA;
739 j_coord_offsetB = DIM*jnrB;
740 j_coord_offsetC = DIM*jnrC;
741 j_coord_offsetD = DIM*jnrD;
743 /* load j atom coordinates */
744 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
745 x+j_coord_offsetC,x+j_coord_offsetD,
748 /* Calculate displacement vector */
749 dx00 = _mm_sub_ps(ix0,jx0);
750 dy00 = _mm_sub_ps(iy0,jy0);
751 dz00 = _mm_sub_ps(iz0,jz0);
752 dx10 = _mm_sub_ps(ix1,jx0);
753 dy10 = _mm_sub_ps(iy1,jy0);
754 dz10 = _mm_sub_ps(iz1,jz0);
755 dx20 = _mm_sub_ps(ix2,jx0);
756 dy20 = _mm_sub_ps(iy2,jy0);
757 dz20 = _mm_sub_ps(iz2,jz0);
758 dx30 = _mm_sub_ps(ix3,jx0);
759 dy30 = _mm_sub_ps(iy3,jy0);
760 dz30 = _mm_sub_ps(iz3,jz0);
762 /* Calculate squared distance and things based on it */
763 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
764 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
765 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
766 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
768 rinv10 = gmx_mm_invsqrt_ps(rsq10);
769 rinv20 = gmx_mm_invsqrt_ps(rsq20);
770 rinv30 = gmx_mm_invsqrt_ps(rsq30);
772 rinvsq00 = gmx_mm_inv_ps(rsq00);
773 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
774 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
775 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
777 /* Load parameters for j particles */
778 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
779 charge+jnrC+0,charge+jnrD+0);
780 vdwjidx0A = 2*vdwtype[jnrA+0];
781 vdwjidx0B = 2*vdwtype[jnrB+0];
782 vdwjidx0C = 2*vdwtype[jnrC+0];
783 vdwjidx0D = 2*vdwtype[jnrD+0];
785 /**************************
786 * CALCULATE INTERACTIONS *
787 **************************/
789 /* Compute parameters for interactions between i and j atoms */
790 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
791 vdwparam+vdwioffset0+vdwjidx0B,
792 vdwparam+vdwioffset0+vdwjidx0C,
793 vdwparam+vdwioffset0+vdwjidx0D,
796 /* LENNARD-JONES DISPERSION/REPULSION */
798 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
799 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
803 /* Calculate temporary vectorial force */
804 tx = _mm_mul_ps(fscal,dx00);
805 ty = _mm_mul_ps(fscal,dy00);
806 tz = _mm_mul_ps(fscal,dz00);
808 /* Update vectorial force */
809 fix0 = _mm_add_ps(fix0,tx);
810 fiy0 = _mm_add_ps(fiy0,ty);
811 fiz0 = _mm_add_ps(fiz0,tz);
813 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
814 f+j_coord_offsetC,f+j_coord_offsetD,
817 /**************************
818 * CALCULATE INTERACTIONS *
819 **************************/
821 /* Compute parameters for interactions between i and j atoms */
822 qq10 = _mm_mul_ps(iq1,jq0);
824 /* COULOMB ELECTROSTATICS */
825 velec = _mm_mul_ps(qq10,rinv10);
826 felec = _mm_mul_ps(velec,rinvsq10);
830 /* Calculate temporary vectorial force */
831 tx = _mm_mul_ps(fscal,dx10);
832 ty = _mm_mul_ps(fscal,dy10);
833 tz = _mm_mul_ps(fscal,dz10);
835 /* Update vectorial force */
836 fix1 = _mm_add_ps(fix1,tx);
837 fiy1 = _mm_add_ps(fiy1,ty);
838 fiz1 = _mm_add_ps(fiz1,tz);
840 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
841 f+j_coord_offsetC,f+j_coord_offsetD,
844 /**************************
845 * CALCULATE INTERACTIONS *
846 **************************/
848 /* Compute parameters for interactions between i and j atoms */
849 qq20 = _mm_mul_ps(iq2,jq0);
851 /* COULOMB ELECTROSTATICS */
852 velec = _mm_mul_ps(qq20,rinv20);
853 felec = _mm_mul_ps(velec,rinvsq20);
857 /* Calculate temporary vectorial force */
858 tx = _mm_mul_ps(fscal,dx20);
859 ty = _mm_mul_ps(fscal,dy20);
860 tz = _mm_mul_ps(fscal,dz20);
862 /* Update vectorial force */
863 fix2 = _mm_add_ps(fix2,tx);
864 fiy2 = _mm_add_ps(fiy2,ty);
865 fiz2 = _mm_add_ps(fiz2,tz);
867 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
868 f+j_coord_offsetC,f+j_coord_offsetD,
871 /**************************
872 * CALCULATE INTERACTIONS *
873 **************************/
875 /* Compute parameters for interactions between i and j atoms */
876 qq30 = _mm_mul_ps(iq3,jq0);
878 /* COULOMB ELECTROSTATICS */
879 velec = _mm_mul_ps(qq30,rinv30);
880 felec = _mm_mul_ps(velec,rinvsq30);
884 /* Calculate temporary vectorial force */
885 tx = _mm_mul_ps(fscal,dx30);
886 ty = _mm_mul_ps(fscal,dy30);
887 tz = _mm_mul_ps(fscal,dz30);
889 /* Update vectorial force */
890 fix3 = _mm_add_ps(fix3,tx);
891 fiy3 = _mm_add_ps(fiy3,ty);
892 fiz3 = _mm_add_ps(fiz3,tz);
894 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
895 f+j_coord_offsetC,f+j_coord_offsetD,
898 /* Inner loop uses 108 flops */
904 /* Get j neighbor index, and coordinate index */
910 /* Sign of each element will be negative for non-real atoms.
911 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
912 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
914 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
915 jnrA = (jnrA>=0) ? jnrA : 0;
916 jnrB = (jnrB>=0) ? jnrB : 0;
917 jnrC = (jnrC>=0) ? jnrC : 0;
918 jnrD = (jnrD>=0) ? jnrD : 0;
920 j_coord_offsetA = DIM*jnrA;
921 j_coord_offsetB = DIM*jnrB;
922 j_coord_offsetC = DIM*jnrC;
923 j_coord_offsetD = DIM*jnrD;
925 /* load j atom coordinates */
926 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
927 x+j_coord_offsetC,x+j_coord_offsetD,
930 /* Calculate displacement vector */
931 dx00 = _mm_sub_ps(ix0,jx0);
932 dy00 = _mm_sub_ps(iy0,jy0);
933 dz00 = _mm_sub_ps(iz0,jz0);
934 dx10 = _mm_sub_ps(ix1,jx0);
935 dy10 = _mm_sub_ps(iy1,jy0);
936 dz10 = _mm_sub_ps(iz1,jz0);
937 dx20 = _mm_sub_ps(ix2,jx0);
938 dy20 = _mm_sub_ps(iy2,jy0);
939 dz20 = _mm_sub_ps(iz2,jz0);
940 dx30 = _mm_sub_ps(ix3,jx0);
941 dy30 = _mm_sub_ps(iy3,jy0);
942 dz30 = _mm_sub_ps(iz3,jz0);
944 /* Calculate squared distance and things based on it */
945 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
946 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
947 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
948 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
950 rinv10 = gmx_mm_invsqrt_ps(rsq10);
951 rinv20 = gmx_mm_invsqrt_ps(rsq20);
952 rinv30 = gmx_mm_invsqrt_ps(rsq30);
954 rinvsq00 = gmx_mm_inv_ps(rsq00);
955 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
956 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
957 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
959 /* Load parameters for j particles */
960 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
961 charge+jnrC+0,charge+jnrD+0);
962 vdwjidx0A = 2*vdwtype[jnrA+0];
963 vdwjidx0B = 2*vdwtype[jnrB+0];
964 vdwjidx0C = 2*vdwtype[jnrC+0];
965 vdwjidx0D = 2*vdwtype[jnrD+0];
967 /**************************
968 * CALCULATE INTERACTIONS *
969 **************************/
971 /* Compute parameters for interactions between i and j atoms */
972 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
973 vdwparam+vdwioffset0+vdwjidx0B,
974 vdwparam+vdwioffset0+vdwjidx0C,
975 vdwparam+vdwioffset0+vdwjidx0D,
978 /* LENNARD-JONES DISPERSION/REPULSION */
980 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
981 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
985 fscal = _mm_andnot_ps(dummy_mask,fscal);
987 /* Calculate temporary vectorial force */
988 tx = _mm_mul_ps(fscal,dx00);
989 ty = _mm_mul_ps(fscal,dy00);
990 tz = _mm_mul_ps(fscal,dz00);
992 /* Update vectorial force */
993 fix0 = _mm_add_ps(fix0,tx);
994 fiy0 = _mm_add_ps(fiy0,ty);
995 fiz0 = _mm_add_ps(fiz0,tz);
997 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
998 f+j_coord_offsetC,f+j_coord_offsetD,
1001 /**************************
1002 * CALCULATE INTERACTIONS *
1003 **************************/
1005 /* Compute parameters for interactions between i and j atoms */
1006 qq10 = _mm_mul_ps(iq1,jq0);
1008 /* COULOMB ELECTROSTATICS */
1009 velec = _mm_mul_ps(qq10,rinv10);
1010 felec = _mm_mul_ps(velec,rinvsq10);
1014 fscal = _mm_andnot_ps(dummy_mask,fscal);
1016 /* Calculate temporary vectorial force */
1017 tx = _mm_mul_ps(fscal,dx10);
1018 ty = _mm_mul_ps(fscal,dy10);
1019 tz = _mm_mul_ps(fscal,dz10);
1021 /* Update vectorial force */
1022 fix1 = _mm_add_ps(fix1,tx);
1023 fiy1 = _mm_add_ps(fiy1,ty);
1024 fiz1 = _mm_add_ps(fiz1,tz);
1026 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1027 f+j_coord_offsetC,f+j_coord_offsetD,
1030 /**************************
1031 * CALCULATE INTERACTIONS *
1032 **************************/
1034 /* Compute parameters for interactions between i and j atoms */
1035 qq20 = _mm_mul_ps(iq2,jq0);
1037 /* COULOMB ELECTROSTATICS */
1038 velec = _mm_mul_ps(qq20,rinv20);
1039 felec = _mm_mul_ps(velec,rinvsq20);
1043 fscal = _mm_andnot_ps(dummy_mask,fscal);
1045 /* Calculate temporary vectorial force */
1046 tx = _mm_mul_ps(fscal,dx20);
1047 ty = _mm_mul_ps(fscal,dy20);
1048 tz = _mm_mul_ps(fscal,dz20);
1050 /* Update vectorial force */
1051 fix2 = _mm_add_ps(fix2,tx);
1052 fiy2 = _mm_add_ps(fiy2,ty);
1053 fiz2 = _mm_add_ps(fiz2,tz);
1055 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1056 f+j_coord_offsetC,f+j_coord_offsetD,
1059 /**************************
1060 * CALCULATE INTERACTIONS *
1061 **************************/
1063 /* Compute parameters for interactions between i and j atoms */
1064 qq30 = _mm_mul_ps(iq3,jq0);
1066 /* COULOMB ELECTROSTATICS */
1067 velec = _mm_mul_ps(qq30,rinv30);
1068 felec = _mm_mul_ps(velec,rinvsq30);
1072 fscal = _mm_andnot_ps(dummy_mask,fscal);
1074 /* Calculate temporary vectorial force */
1075 tx = _mm_mul_ps(fscal,dx30);
1076 ty = _mm_mul_ps(fscal,dy30);
1077 tz = _mm_mul_ps(fscal,dz30);
1079 /* Update vectorial force */
1080 fix3 = _mm_add_ps(fix3,tx);
1081 fiy3 = _mm_add_ps(fiy3,ty);
1082 fiz3 = _mm_add_ps(fiz3,tz);
1084 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1085 f+j_coord_offsetC,f+j_coord_offsetD,
1088 /* Inner loop uses 108 flops */
1091 /* End of innermost loop */
1093 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1094 f+i_coord_offset,fshift+i_shift_offset);
1096 /* Increment number of inner iterations */
1097 inneriter += j_index_end - j_index_start;
1099 /* Outer loop uses 36 flops */
1102 /* Increment number of outer iterations */
1105 /* Update outer/inner flops */
1107 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*36 + inneriter*108);