2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
48 #include "kernelutil_x86_avx_128_fma_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_128_fma_single
52 * Electrostatics interaction: Coulomb
53 * VdW interaction: LennardJones
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_128_fma_single
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m128 fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
89 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
90 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
92 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
97 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
100 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
101 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
102 __m128 dummy_mask,cutoff_mask;
103 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
104 __m128 one = _mm_set1_ps(1.0);
105 __m128 two = _mm_set1_ps(2.0);
111 jindex = nlist->jindex;
113 shiftidx = nlist->shift;
115 shiftvec = fr->shift_vec[0];
116 fshift = fr->fshift[0];
117 facel = _mm_set1_ps(fr->epsfac);
118 charge = mdatoms->chargeA;
119 nvdwtype = fr->ntype;
121 vdwtype = mdatoms->typeA;
123 /* Setup water-specific parameters */
124 inr = nlist->iinr[0];
125 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
126 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
127 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
128 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
130 /* Avoid stupid compiler warnings */
131 jnrA = jnrB = jnrC = jnrD = 0;
140 for(iidx=0;iidx<4*DIM;iidx++)
145 /* Start outer loop over neighborlists */
146 for(iidx=0; iidx<nri; iidx++)
148 /* Load shift vector for this list */
149 i_shift_offset = DIM*shiftidx[iidx];
151 /* Load limits for loop over neighbors */
152 j_index_start = jindex[iidx];
153 j_index_end = jindex[iidx+1];
155 /* Get outer coordinate index */
157 i_coord_offset = DIM*inr;
159 /* Load i particle coords and add shift vector */
160 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
161 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
163 fix0 = _mm_setzero_ps();
164 fiy0 = _mm_setzero_ps();
165 fiz0 = _mm_setzero_ps();
166 fix1 = _mm_setzero_ps();
167 fiy1 = _mm_setzero_ps();
168 fiz1 = _mm_setzero_ps();
169 fix2 = _mm_setzero_ps();
170 fiy2 = _mm_setzero_ps();
171 fiz2 = _mm_setzero_ps();
173 /* Reset potential sums */
174 velecsum = _mm_setzero_ps();
175 vvdwsum = _mm_setzero_ps();
177 /* Start inner kernel loop */
178 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
181 /* Get j neighbor index, and coordinate index */
186 j_coord_offsetA = DIM*jnrA;
187 j_coord_offsetB = DIM*jnrB;
188 j_coord_offsetC = DIM*jnrC;
189 j_coord_offsetD = DIM*jnrD;
191 /* load j atom coordinates */
192 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
193 x+j_coord_offsetC,x+j_coord_offsetD,
196 /* Calculate displacement vector */
197 dx00 = _mm_sub_ps(ix0,jx0);
198 dy00 = _mm_sub_ps(iy0,jy0);
199 dz00 = _mm_sub_ps(iz0,jz0);
200 dx10 = _mm_sub_ps(ix1,jx0);
201 dy10 = _mm_sub_ps(iy1,jy0);
202 dz10 = _mm_sub_ps(iz1,jz0);
203 dx20 = _mm_sub_ps(ix2,jx0);
204 dy20 = _mm_sub_ps(iy2,jy0);
205 dz20 = _mm_sub_ps(iz2,jz0);
207 /* Calculate squared distance and things based on it */
208 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
209 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
210 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
212 rinv00 = gmx_mm_invsqrt_ps(rsq00);
213 rinv10 = gmx_mm_invsqrt_ps(rsq10);
214 rinv20 = gmx_mm_invsqrt_ps(rsq20);
216 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
217 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
218 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
220 /* Load parameters for j particles */
221 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
222 charge+jnrC+0,charge+jnrD+0);
223 vdwjidx0A = 2*vdwtype[jnrA+0];
224 vdwjidx0B = 2*vdwtype[jnrB+0];
225 vdwjidx0C = 2*vdwtype[jnrC+0];
226 vdwjidx0D = 2*vdwtype[jnrD+0];
228 fjx0 = _mm_setzero_ps();
229 fjy0 = _mm_setzero_ps();
230 fjz0 = _mm_setzero_ps();
232 /**************************
233 * CALCULATE INTERACTIONS *
234 **************************/
236 /* Compute parameters for interactions between i and j atoms */
237 qq00 = _mm_mul_ps(iq0,jq0);
238 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
239 vdwparam+vdwioffset0+vdwjidx0B,
240 vdwparam+vdwioffset0+vdwjidx0C,
241 vdwparam+vdwioffset0+vdwjidx0D,
244 /* COULOMB ELECTROSTATICS */
245 velec = _mm_mul_ps(qq00,rinv00);
246 felec = _mm_mul_ps(velec,rinvsq00);
248 /* LENNARD-JONES DISPERSION/REPULSION */
250 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
251 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
252 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
253 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
254 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
256 /* Update potential sum for this i atom from the interaction with this j atom. */
257 velecsum = _mm_add_ps(velecsum,velec);
258 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
260 fscal = _mm_add_ps(felec,fvdw);
262 /* Update vectorial force */
263 fix0 = _mm_macc_ps(dx00,fscal,fix0);
264 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
265 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
267 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
268 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
269 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
271 /**************************
272 * CALCULATE INTERACTIONS *
273 **************************/
275 /* Compute parameters for interactions between i and j atoms */
276 qq10 = _mm_mul_ps(iq1,jq0);
278 /* COULOMB ELECTROSTATICS */
279 velec = _mm_mul_ps(qq10,rinv10);
280 felec = _mm_mul_ps(velec,rinvsq10);
282 /* Update potential sum for this i atom from the interaction with this j atom. */
283 velecsum = _mm_add_ps(velecsum,velec);
287 /* Update vectorial force */
288 fix1 = _mm_macc_ps(dx10,fscal,fix1);
289 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
290 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
292 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
293 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
294 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
296 /**************************
297 * CALCULATE INTERACTIONS *
298 **************************/
300 /* Compute parameters for interactions between i and j atoms */
301 qq20 = _mm_mul_ps(iq2,jq0);
303 /* COULOMB ELECTROSTATICS */
304 velec = _mm_mul_ps(qq20,rinv20);
305 felec = _mm_mul_ps(velec,rinvsq20);
307 /* Update potential sum for this i atom from the interaction with this j atom. */
308 velecsum = _mm_add_ps(velecsum,velec);
312 /* Update vectorial force */
313 fix2 = _mm_macc_ps(dx20,fscal,fix2);
314 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
315 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
317 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
318 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
319 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
321 fjptrA = f+j_coord_offsetA;
322 fjptrB = f+j_coord_offsetB;
323 fjptrC = f+j_coord_offsetC;
324 fjptrD = f+j_coord_offsetD;
326 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
328 /* Inner loop uses 105 flops */
334 /* Get j neighbor index, and coordinate index */
335 jnrlistA = jjnr[jidx];
336 jnrlistB = jjnr[jidx+1];
337 jnrlistC = jjnr[jidx+2];
338 jnrlistD = jjnr[jidx+3];
339 /* Sign of each element will be negative for non-real atoms.
340 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
341 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
343 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
344 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
345 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
346 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
347 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
348 j_coord_offsetA = DIM*jnrA;
349 j_coord_offsetB = DIM*jnrB;
350 j_coord_offsetC = DIM*jnrC;
351 j_coord_offsetD = DIM*jnrD;
353 /* load j atom coordinates */
354 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
355 x+j_coord_offsetC,x+j_coord_offsetD,
358 /* Calculate displacement vector */
359 dx00 = _mm_sub_ps(ix0,jx0);
360 dy00 = _mm_sub_ps(iy0,jy0);
361 dz00 = _mm_sub_ps(iz0,jz0);
362 dx10 = _mm_sub_ps(ix1,jx0);
363 dy10 = _mm_sub_ps(iy1,jy0);
364 dz10 = _mm_sub_ps(iz1,jz0);
365 dx20 = _mm_sub_ps(ix2,jx0);
366 dy20 = _mm_sub_ps(iy2,jy0);
367 dz20 = _mm_sub_ps(iz2,jz0);
369 /* Calculate squared distance and things based on it */
370 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
371 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
372 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
374 rinv00 = gmx_mm_invsqrt_ps(rsq00);
375 rinv10 = gmx_mm_invsqrt_ps(rsq10);
376 rinv20 = gmx_mm_invsqrt_ps(rsq20);
378 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
379 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
380 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
382 /* Load parameters for j particles */
383 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
384 charge+jnrC+0,charge+jnrD+0);
385 vdwjidx0A = 2*vdwtype[jnrA+0];
386 vdwjidx0B = 2*vdwtype[jnrB+0];
387 vdwjidx0C = 2*vdwtype[jnrC+0];
388 vdwjidx0D = 2*vdwtype[jnrD+0];
390 fjx0 = _mm_setzero_ps();
391 fjy0 = _mm_setzero_ps();
392 fjz0 = _mm_setzero_ps();
394 /**************************
395 * CALCULATE INTERACTIONS *
396 **************************/
398 /* Compute parameters for interactions between i and j atoms */
399 qq00 = _mm_mul_ps(iq0,jq0);
400 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
401 vdwparam+vdwioffset0+vdwjidx0B,
402 vdwparam+vdwioffset0+vdwjidx0C,
403 vdwparam+vdwioffset0+vdwjidx0D,
406 /* COULOMB ELECTROSTATICS */
407 velec = _mm_mul_ps(qq00,rinv00);
408 felec = _mm_mul_ps(velec,rinvsq00);
410 /* LENNARD-JONES DISPERSION/REPULSION */
412 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
413 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
414 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
415 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
416 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
418 /* Update potential sum for this i atom from the interaction with this j atom. */
419 velec = _mm_andnot_ps(dummy_mask,velec);
420 velecsum = _mm_add_ps(velecsum,velec);
421 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
422 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
424 fscal = _mm_add_ps(felec,fvdw);
426 fscal = _mm_andnot_ps(dummy_mask,fscal);
428 /* Update vectorial force */
429 fix0 = _mm_macc_ps(dx00,fscal,fix0);
430 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
431 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
433 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
434 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
435 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
437 /**************************
438 * CALCULATE INTERACTIONS *
439 **************************/
441 /* Compute parameters for interactions between i and j atoms */
442 qq10 = _mm_mul_ps(iq1,jq0);
444 /* COULOMB ELECTROSTATICS */
445 velec = _mm_mul_ps(qq10,rinv10);
446 felec = _mm_mul_ps(velec,rinvsq10);
448 /* Update potential sum for this i atom from the interaction with this j atom. */
449 velec = _mm_andnot_ps(dummy_mask,velec);
450 velecsum = _mm_add_ps(velecsum,velec);
454 fscal = _mm_andnot_ps(dummy_mask,fscal);
456 /* Update vectorial force */
457 fix1 = _mm_macc_ps(dx10,fscal,fix1);
458 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
459 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
461 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
462 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
463 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
465 /**************************
466 * CALCULATE INTERACTIONS *
467 **************************/
469 /* Compute parameters for interactions between i and j atoms */
470 qq20 = _mm_mul_ps(iq2,jq0);
472 /* COULOMB ELECTROSTATICS */
473 velec = _mm_mul_ps(qq20,rinv20);
474 felec = _mm_mul_ps(velec,rinvsq20);
476 /* Update potential sum for this i atom from the interaction with this j atom. */
477 velec = _mm_andnot_ps(dummy_mask,velec);
478 velecsum = _mm_add_ps(velecsum,velec);
482 fscal = _mm_andnot_ps(dummy_mask,fscal);
484 /* Update vectorial force */
485 fix2 = _mm_macc_ps(dx20,fscal,fix2);
486 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
487 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
489 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
490 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
491 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
493 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
494 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
495 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
496 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
498 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
500 /* Inner loop uses 105 flops */
503 /* End of innermost loop */
505 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
506 f+i_coord_offset,fshift+i_shift_offset);
509 /* Update potential energies */
510 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
511 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
513 /* Increment number of inner iterations */
514 inneriter += j_index_end - j_index_start;
516 /* Outer loop uses 20 flops */
519 /* Increment number of outer iterations */
522 /* Update outer/inner flops */
524 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*105);
527 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_single
528 * Electrostatics interaction: Coulomb
529 * VdW interaction: LennardJones
530 * Geometry: Water3-Particle
531 * Calculate force/pot: Force
534 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_single
535 (t_nblist * gmx_restrict nlist,
536 rvec * gmx_restrict xx,
537 rvec * gmx_restrict ff,
538 t_forcerec * gmx_restrict fr,
539 t_mdatoms * gmx_restrict mdatoms,
540 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
541 t_nrnb * gmx_restrict nrnb)
543 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
544 * just 0 for non-waters.
545 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
546 * jnr indices corresponding to data put in the four positions in the SIMD register.
548 int i_shift_offset,i_coord_offset,outeriter,inneriter;
549 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
550 int jnrA,jnrB,jnrC,jnrD;
551 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
552 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
553 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
555 real *shiftvec,*fshift,*x,*f;
556 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
558 __m128 fscal,rcutoff,rcutoff2,jidxall;
560 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
562 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
564 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
565 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
566 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
567 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
568 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
569 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
570 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
573 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
576 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
577 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
578 __m128 dummy_mask,cutoff_mask;
579 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
580 __m128 one = _mm_set1_ps(1.0);
581 __m128 two = _mm_set1_ps(2.0);
587 jindex = nlist->jindex;
589 shiftidx = nlist->shift;
591 shiftvec = fr->shift_vec[0];
592 fshift = fr->fshift[0];
593 facel = _mm_set1_ps(fr->epsfac);
594 charge = mdatoms->chargeA;
595 nvdwtype = fr->ntype;
597 vdwtype = mdatoms->typeA;
599 /* Setup water-specific parameters */
600 inr = nlist->iinr[0];
601 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
602 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
603 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
604 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
606 /* Avoid stupid compiler warnings */
607 jnrA = jnrB = jnrC = jnrD = 0;
616 for(iidx=0;iidx<4*DIM;iidx++)
621 /* Start outer loop over neighborlists */
622 for(iidx=0; iidx<nri; iidx++)
624 /* Load shift vector for this list */
625 i_shift_offset = DIM*shiftidx[iidx];
627 /* Load limits for loop over neighbors */
628 j_index_start = jindex[iidx];
629 j_index_end = jindex[iidx+1];
631 /* Get outer coordinate index */
633 i_coord_offset = DIM*inr;
635 /* Load i particle coords and add shift vector */
636 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
637 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
639 fix0 = _mm_setzero_ps();
640 fiy0 = _mm_setzero_ps();
641 fiz0 = _mm_setzero_ps();
642 fix1 = _mm_setzero_ps();
643 fiy1 = _mm_setzero_ps();
644 fiz1 = _mm_setzero_ps();
645 fix2 = _mm_setzero_ps();
646 fiy2 = _mm_setzero_ps();
647 fiz2 = _mm_setzero_ps();
649 /* Start inner kernel loop */
650 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
653 /* Get j neighbor index, and coordinate index */
658 j_coord_offsetA = DIM*jnrA;
659 j_coord_offsetB = DIM*jnrB;
660 j_coord_offsetC = DIM*jnrC;
661 j_coord_offsetD = DIM*jnrD;
663 /* load j atom coordinates */
664 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
665 x+j_coord_offsetC,x+j_coord_offsetD,
668 /* Calculate displacement vector */
669 dx00 = _mm_sub_ps(ix0,jx0);
670 dy00 = _mm_sub_ps(iy0,jy0);
671 dz00 = _mm_sub_ps(iz0,jz0);
672 dx10 = _mm_sub_ps(ix1,jx0);
673 dy10 = _mm_sub_ps(iy1,jy0);
674 dz10 = _mm_sub_ps(iz1,jz0);
675 dx20 = _mm_sub_ps(ix2,jx0);
676 dy20 = _mm_sub_ps(iy2,jy0);
677 dz20 = _mm_sub_ps(iz2,jz0);
679 /* Calculate squared distance and things based on it */
680 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
681 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
682 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
684 rinv00 = gmx_mm_invsqrt_ps(rsq00);
685 rinv10 = gmx_mm_invsqrt_ps(rsq10);
686 rinv20 = gmx_mm_invsqrt_ps(rsq20);
688 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
689 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
690 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
692 /* Load parameters for j particles */
693 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
694 charge+jnrC+0,charge+jnrD+0);
695 vdwjidx0A = 2*vdwtype[jnrA+0];
696 vdwjidx0B = 2*vdwtype[jnrB+0];
697 vdwjidx0C = 2*vdwtype[jnrC+0];
698 vdwjidx0D = 2*vdwtype[jnrD+0];
700 fjx0 = _mm_setzero_ps();
701 fjy0 = _mm_setzero_ps();
702 fjz0 = _mm_setzero_ps();
704 /**************************
705 * CALCULATE INTERACTIONS *
706 **************************/
708 /* Compute parameters for interactions between i and j atoms */
709 qq00 = _mm_mul_ps(iq0,jq0);
710 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
711 vdwparam+vdwioffset0+vdwjidx0B,
712 vdwparam+vdwioffset0+vdwjidx0C,
713 vdwparam+vdwioffset0+vdwjidx0D,
716 /* COULOMB ELECTROSTATICS */
717 velec = _mm_mul_ps(qq00,rinv00);
718 felec = _mm_mul_ps(velec,rinvsq00);
720 /* LENNARD-JONES DISPERSION/REPULSION */
722 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
723 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
725 fscal = _mm_add_ps(felec,fvdw);
727 /* Update vectorial force */
728 fix0 = _mm_macc_ps(dx00,fscal,fix0);
729 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
730 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
732 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
733 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
734 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
736 /**************************
737 * CALCULATE INTERACTIONS *
738 **************************/
740 /* Compute parameters for interactions between i and j atoms */
741 qq10 = _mm_mul_ps(iq1,jq0);
743 /* COULOMB ELECTROSTATICS */
744 velec = _mm_mul_ps(qq10,rinv10);
745 felec = _mm_mul_ps(velec,rinvsq10);
749 /* Update vectorial force */
750 fix1 = _mm_macc_ps(dx10,fscal,fix1);
751 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
752 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
754 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
755 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
756 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
758 /**************************
759 * CALCULATE INTERACTIONS *
760 **************************/
762 /* Compute parameters for interactions between i and j atoms */
763 qq20 = _mm_mul_ps(iq2,jq0);
765 /* COULOMB ELECTROSTATICS */
766 velec = _mm_mul_ps(qq20,rinv20);
767 felec = _mm_mul_ps(velec,rinvsq20);
771 /* Update vectorial force */
772 fix2 = _mm_macc_ps(dx20,fscal,fix2);
773 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
774 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
776 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
777 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
778 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
780 fjptrA = f+j_coord_offsetA;
781 fjptrB = f+j_coord_offsetB;
782 fjptrC = f+j_coord_offsetC;
783 fjptrD = f+j_coord_offsetD;
785 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
787 /* Inner loop uses 97 flops */
793 /* Get j neighbor index, and coordinate index */
794 jnrlistA = jjnr[jidx];
795 jnrlistB = jjnr[jidx+1];
796 jnrlistC = jjnr[jidx+2];
797 jnrlistD = jjnr[jidx+3];
798 /* Sign of each element will be negative for non-real atoms.
799 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
800 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
802 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
803 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
804 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
805 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
806 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
807 j_coord_offsetA = DIM*jnrA;
808 j_coord_offsetB = DIM*jnrB;
809 j_coord_offsetC = DIM*jnrC;
810 j_coord_offsetD = DIM*jnrD;
812 /* load j atom coordinates */
813 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
814 x+j_coord_offsetC,x+j_coord_offsetD,
817 /* Calculate displacement vector */
818 dx00 = _mm_sub_ps(ix0,jx0);
819 dy00 = _mm_sub_ps(iy0,jy0);
820 dz00 = _mm_sub_ps(iz0,jz0);
821 dx10 = _mm_sub_ps(ix1,jx0);
822 dy10 = _mm_sub_ps(iy1,jy0);
823 dz10 = _mm_sub_ps(iz1,jz0);
824 dx20 = _mm_sub_ps(ix2,jx0);
825 dy20 = _mm_sub_ps(iy2,jy0);
826 dz20 = _mm_sub_ps(iz2,jz0);
828 /* Calculate squared distance and things based on it */
829 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
830 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
831 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
833 rinv00 = gmx_mm_invsqrt_ps(rsq00);
834 rinv10 = gmx_mm_invsqrt_ps(rsq10);
835 rinv20 = gmx_mm_invsqrt_ps(rsq20);
837 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
838 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
839 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
841 /* Load parameters for j particles */
842 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
843 charge+jnrC+0,charge+jnrD+0);
844 vdwjidx0A = 2*vdwtype[jnrA+0];
845 vdwjidx0B = 2*vdwtype[jnrB+0];
846 vdwjidx0C = 2*vdwtype[jnrC+0];
847 vdwjidx0D = 2*vdwtype[jnrD+0];
849 fjx0 = _mm_setzero_ps();
850 fjy0 = _mm_setzero_ps();
851 fjz0 = _mm_setzero_ps();
853 /**************************
854 * CALCULATE INTERACTIONS *
855 **************************/
857 /* Compute parameters for interactions between i and j atoms */
858 qq00 = _mm_mul_ps(iq0,jq0);
859 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
860 vdwparam+vdwioffset0+vdwjidx0B,
861 vdwparam+vdwioffset0+vdwjidx0C,
862 vdwparam+vdwioffset0+vdwjidx0D,
865 /* COULOMB ELECTROSTATICS */
866 velec = _mm_mul_ps(qq00,rinv00);
867 felec = _mm_mul_ps(velec,rinvsq00);
869 /* LENNARD-JONES DISPERSION/REPULSION */
871 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
872 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
874 fscal = _mm_add_ps(felec,fvdw);
876 fscal = _mm_andnot_ps(dummy_mask,fscal);
878 /* Update vectorial force */
879 fix0 = _mm_macc_ps(dx00,fscal,fix0);
880 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
881 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
883 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
884 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
885 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
887 /**************************
888 * CALCULATE INTERACTIONS *
889 **************************/
891 /* Compute parameters for interactions between i and j atoms */
892 qq10 = _mm_mul_ps(iq1,jq0);
894 /* COULOMB ELECTROSTATICS */
895 velec = _mm_mul_ps(qq10,rinv10);
896 felec = _mm_mul_ps(velec,rinvsq10);
900 fscal = _mm_andnot_ps(dummy_mask,fscal);
902 /* Update vectorial force */
903 fix1 = _mm_macc_ps(dx10,fscal,fix1);
904 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
905 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
907 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
908 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
909 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
911 /**************************
912 * CALCULATE INTERACTIONS *
913 **************************/
915 /* Compute parameters for interactions between i and j atoms */
916 qq20 = _mm_mul_ps(iq2,jq0);
918 /* COULOMB ELECTROSTATICS */
919 velec = _mm_mul_ps(qq20,rinv20);
920 felec = _mm_mul_ps(velec,rinvsq20);
924 fscal = _mm_andnot_ps(dummy_mask,fscal);
926 /* Update vectorial force */
927 fix2 = _mm_macc_ps(dx20,fscal,fix2);
928 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
929 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
931 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
932 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
933 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
935 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
936 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
937 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
938 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
940 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
942 /* Inner loop uses 97 flops */
945 /* End of innermost loop */
947 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
948 f+i_coord_offset,fshift+i_shift_offset);
950 /* Increment number of inner iterations */
951 inneriter += j_index_end - j_index_start;
953 /* Outer loop uses 18 flops */
956 /* Increment number of outer iterations */
959 /* Update outer/inner flops */
961 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*97);