2 * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_single.h"
34 #include "kernelutil_x86_avx_128_fma_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_128_fma_single
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_128_fma_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 AVX_128, 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 fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
72 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
74 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
75 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
76 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
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 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
84 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
86 __m128 dummy_mask,cutoff_mask;
87 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
88 __m128 one = _mm_set1_ps(1.0);
89 __m128 two = _mm_set1_ps(2.0);
95 jindex = nlist->jindex;
97 shiftidx = nlist->shift;
99 shiftvec = fr->shift_vec[0];
100 fshift = fr->fshift[0];
101 facel = _mm_set1_ps(fr->epsfac);
102 charge = mdatoms->chargeA;
104 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
105 beta = _mm_set1_ps(fr->ic->ewaldcoeff);
106 beta2 = _mm_mul_ps(beta,beta);
107 beta3 = _mm_mul_ps(beta,beta2);
108 ewtab = fr->ic->tabq_coul_FDV0;
109 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
110 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
112 /* Setup water-specific parameters */
113 inr = nlist->iinr[0];
114 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
115 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
116 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
118 /* Avoid stupid compiler warnings */
119 jnrA = jnrB = jnrC = jnrD = 0;
128 for(iidx=0;iidx<4*DIM;iidx++)
133 /* Start outer loop over neighborlists */
134 for(iidx=0; iidx<nri; iidx++)
136 /* Load shift vector for this list */
137 i_shift_offset = DIM*shiftidx[iidx];
139 /* Load limits for loop over neighbors */
140 j_index_start = jindex[iidx];
141 j_index_end = jindex[iidx+1];
143 /* Get outer coordinate index */
145 i_coord_offset = DIM*inr;
147 /* Load i particle coords and add shift vector */
148 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
149 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
151 fix1 = _mm_setzero_ps();
152 fiy1 = _mm_setzero_ps();
153 fiz1 = _mm_setzero_ps();
154 fix2 = _mm_setzero_ps();
155 fiy2 = _mm_setzero_ps();
156 fiz2 = _mm_setzero_ps();
157 fix3 = _mm_setzero_ps();
158 fiy3 = _mm_setzero_ps();
159 fiz3 = _mm_setzero_ps();
161 /* Reset potential sums */
162 velecsum = _mm_setzero_ps();
164 /* Start inner kernel loop */
165 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
168 /* Get j neighbor index, and coordinate index */
173 j_coord_offsetA = DIM*jnrA;
174 j_coord_offsetB = DIM*jnrB;
175 j_coord_offsetC = DIM*jnrC;
176 j_coord_offsetD = DIM*jnrD;
178 /* load j atom coordinates */
179 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
180 x+j_coord_offsetC,x+j_coord_offsetD,
183 /* Calculate displacement vector */
184 dx10 = _mm_sub_ps(ix1,jx0);
185 dy10 = _mm_sub_ps(iy1,jy0);
186 dz10 = _mm_sub_ps(iz1,jz0);
187 dx20 = _mm_sub_ps(ix2,jx0);
188 dy20 = _mm_sub_ps(iy2,jy0);
189 dz20 = _mm_sub_ps(iz2,jz0);
190 dx30 = _mm_sub_ps(ix3,jx0);
191 dy30 = _mm_sub_ps(iy3,jy0);
192 dz30 = _mm_sub_ps(iz3,jz0);
194 /* Calculate squared distance and things based on it */
195 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
196 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
197 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
199 rinv10 = gmx_mm_invsqrt_ps(rsq10);
200 rinv20 = gmx_mm_invsqrt_ps(rsq20);
201 rinv30 = gmx_mm_invsqrt_ps(rsq30);
203 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
204 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
205 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
207 /* Load parameters for j particles */
208 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
209 charge+jnrC+0,charge+jnrD+0);
211 fjx0 = _mm_setzero_ps();
212 fjy0 = _mm_setzero_ps();
213 fjz0 = _mm_setzero_ps();
215 /**************************
216 * CALCULATE INTERACTIONS *
217 **************************/
219 r10 = _mm_mul_ps(rsq10,rinv10);
221 /* Compute parameters for interactions between i and j atoms */
222 qq10 = _mm_mul_ps(iq1,jq0);
224 /* EWALD ELECTROSTATICS */
226 /* Analytical PME correction */
227 zeta2 = _mm_mul_ps(beta2,rsq10);
228 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
229 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
230 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
231 felec = _mm_mul_ps(qq10,felec);
232 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
233 velec = _mm_nmacc_ps(pmecorrV,beta,rinv10);
234 velec = _mm_mul_ps(qq10,velec);
236 /* Update potential sum for this i atom from the interaction with this j atom. */
237 velecsum = _mm_add_ps(velecsum,velec);
241 /* Update vectorial force */
242 fix1 = _mm_macc_ps(dx10,fscal,fix1);
243 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
244 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
246 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
247 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
248 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
250 /**************************
251 * CALCULATE INTERACTIONS *
252 **************************/
254 r20 = _mm_mul_ps(rsq20,rinv20);
256 /* Compute parameters for interactions between i and j atoms */
257 qq20 = _mm_mul_ps(iq2,jq0);
259 /* EWALD ELECTROSTATICS */
261 /* Analytical PME correction */
262 zeta2 = _mm_mul_ps(beta2,rsq20);
263 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
264 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
265 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
266 felec = _mm_mul_ps(qq20,felec);
267 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
268 velec = _mm_nmacc_ps(pmecorrV,beta,rinv20);
269 velec = _mm_mul_ps(qq20,velec);
271 /* Update potential sum for this i atom from the interaction with this j atom. */
272 velecsum = _mm_add_ps(velecsum,velec);
276 /* Update vectorial force */
277 fix2 = _mm_macc_ps(dx20,fscal,fix2);
278 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
279 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
281 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
282 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
283 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
285 /**************************
286 * CALCULATE INTERACTIONS *
287 **************************/
289 r30 = _mm_mul_ps(rsq30,rinv30);
291 /* Compute parameters for interactions between i and j atoms */
292 qq30 = _mm_mul_ps(iq3,jq0);
294 /* EWALD ELECTROSTATICS */
296 /* Analytical PME correction */
297 zeta2 = _mm_mul_ps(beta2,rsq30);
298 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
299 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
300 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
301 felec = _mm_mul_ps(qq30,felec);
302 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
303 velec = _mm_nmacc_ps(pmecorrV,beta,rinv30);
304 velec = _mm_mul_ps(qq30,velec);
306 /* Update potential sum for this i atom from the interaction with this j atom. */
307 velecsum = _mm_add_ps(velecsum,velec);
311 /* Update vectorial force */
312 fix3 = _mm_macc_ps(dx30,fscal,fix3);
313 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
314 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
316 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
317 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
318 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
320 fjptrA = f+j_coord_offsetA;
321 fjptrB = f+j_coord_offsetB;
322 fjptrC = f+j_coord_offsetC;
323 fjptrD = f+j_coord_offsetD;
325 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
327 /* Inner loop uses 87 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 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);
364 dx30 = _mm_sub_ps(ix3,jx0);
365 dy30 = _mm_sub_ps(iy3,jy0);
366 dz30 = _mm_sub_ps(iz3,jz0);
368 /* Calculate squared distance and things based on it */
369 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
370 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
371 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
373 rinv10 = gmx_mm_invsqrt_ps(rsq10);
374 rinv20 = gmx_mm_invsqrt_ps(rsq20);
375 rinv30 = gmx_mm_invsqrt_ps(rsq30);
377 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
378 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
379 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
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);
385 fjx0 = _mm_setzero_ps();
386 fjy0 = _mm_setzero_ps();
387 fjz0 = _mm_setzero_ps();
389 /**************************
390 * CALCULATE INTERACTIONS *
391 **************************/
393 r10 = _mm_mul_ps(rsq10,rinv10);
394 r10 = _mm_andnot_ps(dummy_mask,r10);
396 /* Compute parameters for interactions between i and j atoms */
397 qq10 = _mm_mul_ps(iq1,jq0);
399 /* EWALD ELECTROSTATICS */
401 /* Analytical PME correction */
402 zeta2 = _mm_mul_ps(beta2,rsq10);
403 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
404 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
405 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
406 felec = _mm_mul_ps(qq10,felec);
407 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
408 velec = _mm_nmacc_ps(pmecorrV,beta,rinv10);
409 velec = _mm_mul_ps(qq10,velec);
411 /* Update potential sum for this i atom from the interaction with this j atom. */
412 velec = _mm_andnot_ps(dummy_mask,velec);
413 velecsum = _mm_add_ps(velecsum,velec);
417 fscal = _mm_andnot_ps(dummy_mask,fscal);
419 /* Update vectorial force */
420 fix1 = _mm_macc_ps(dx10,fscal,fix1);
421 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
422 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
424 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
425 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
426 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
428 /**************************
429 * CALCULATE INTERACTIONS *
430 **************************/
432 r20 = _mm_mul_ps(rsq20,rinv20);
433 r20 = _mm_andnot_ps(dummy_mask,r20);
435 /* Compute parameters for interactions between i and j atoms */
436 qq20 = _mm_mul_ps(iq2,jq0);
438 /* EWALD ELECTROSTATICS */
440 /* Analytical PME correction */
441 zeta2 = _mm_mul_ps(beta2,rsq20);
442 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
443 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
444 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
445 felec = _mm_mul_ps(qq20,felec);
446 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
447 velec = _mm_nmacc_ps(pmecorrV,beta,rinv20);
448 velec = _mm_mul_ps(qq20,velec);
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 /* Update vectorial force */
459 fix2 = _mm_macc_ps(dx20,fscal,fix2);
460 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
461 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
463 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
464 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
465 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
467 /**************************
468 * CALCULATE INTERACTIONS *
469 **************************/
471 r30 = _mm_mul_ps(rsq30,rinv30);
472 r30 = _mm_andnot_ps(dummy_mask,r30);
474 /* Compute parameters for interactions between i and j atoms */
475 qq30 = _mm_mul_ps(iq3,jq0);
477 /* EWALD ELECTROSTATICS */
479 /* Analytical PME correction */
480 zeta2 = _mm_mul_ps(beta2,rsq30);
481 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
482 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
483 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
484 felec = _mm_mul_ps(qq30,felec);
485 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
486 velec = _mm_nmacc_ps(pmecorrV,beta,rinv30);
487 velec = _mm_mul_ps(qq30,velec);
489 /* Update potential sum for this i atom from the interaction with this j atom. */
490 velec = _mm_andnot_ps(dummy_mask,velec);
491 velecsum = _mm_add_ps(velecsum,velec);
495 fscal = _mm_andnot_ps(dummy_mask,fscal);
497 /* Update vectorial force */
498 fix3 = _mm_macc_ps(dx30,fscal,fix3);
499 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
500 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
502 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
503 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
504 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
506 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
507 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
508 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
509 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
511 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
513 /* Inner loop uses 90 flops */
516 /* End of innermost loop */
518 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
519 f+i_coord_offset+DIM,fshift+i_shift_offset);
522 /* Update potential energies */
523 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
525 /* Increment number of inner iterations */
526 inneriter += j_index_end - j_index_start;
528 /* Outer loop uses 19 flops */
531 /* Increment number of outer iterations */
534 /* Update outer/inner flops */
536 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*90);
539 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_128_fma_single
540 * Electrostatics interaction: Ewald
541 * VdW interaction: None
542 * Geometry: Water4-Particle
543 * Calculate force/pot: Force
546 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_128_fma_single
547 (t_nblist * gmx_restrict nlist,
548 rvec * gmx_restrict xx,
549 rvec * gmx_restrict ff,
550 t_forcerec * gmx_restrict fr,
551 t_mdatoms * gmx_restrict mdatoms,
552 nb_kernel_data_t * gmx_restrict kernel_data,
553 t_nrnb * gmx_restrict nrnb)
555 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
556 * just 0 for non-waters.
557 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
558 * jnr indices corresponding to data put in the four positions in the SIMD register.
560 int i_shift_offset,i_coord_offset,outeriter,inneriter;
561 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
562 int jnrA,jnrB,jnrC,jnrD;
563 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
564 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
565 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
567 real *shiftvec,*fshift,*x,*f;
568 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
570 __m128 fscal,rcutoff,rcutoff2,jidxall;
572 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
574 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
576 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
577 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
578 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
579 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
580 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
581 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
582 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
585 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
586 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
588 __m128 dummy_mask,cutoff_mask;
589 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
590 __m128 one = _mm_set1_ps(1.0);
591 __m128 two = _mm_set1_ps(2.0);
597 jindex = nlist->jindex;
599 shiftidx = nlist->shift;
601 shiftvec = fr->shift_vec[0];
602 fshift = fr->fshift[0];
603 facel = _mm_set1_ps(fr->epsfac);
604 charge = mdatoms->chargeA;
606 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
607 beta = _mm_set1_ps(fr->ic->ewaldcoeff);
608 beta2 = _mm_mul_ps(beta,beta);
609 beta3 = _mm_mul_ps(beta,beta2);
610 ewtab = fr->ic->tabq_coul_F;
611 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
612 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
614 /* Setup water-specific parameters */
615 inr = nlist->iinr[0];
616 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
617 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
618 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
620 /* Avoid stupid compiler warnings */
621 jnrA = jnrB = jnrC = jnrD = 0;
630 for(iidx=0;iidx<4*DIM;iidx++)
635 /* Start outer loop over neighborlists */
636 for(iidx=0; iidx<nri; iidx++)
638 /* Load shift vector for this list */
639 i_shift_offset = DIM*shiftidx[iidx];
641 /* Load limits for loop over neighbors */
642 j_index_start = jindex[iidx];
643 j_index_end = jindex[iidx+1];
645 /* Get outer coordinate index */
647 i_coord_offset = DIM*inr;
649 /* Load i particle coords and add shift vector */
650 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
651 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
653 fix1 = _mm_setzero_ps();
654 fiy1 = _mm_setzero_ps();
655 fiz1 = _mm_setzero_ps();
656 fix2 = _mm_setzero_ps();
657 fiy2 = _mm_setzero_ps();
658 fiz2 = _mm_setzero_ps();
659 fix3 = _mm_setzero_ps();
660 fiy3 = _mm_setzero_ps();
661 fiz3 = _mm_setzero_ps();
663 /* Start inner kernel loop */
664 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
667 /* Get j neighbor index, and coordinate index */
672 j_coord_offsetA = DIM*jnrA;
673 j_coord_offsetB = DIM*jnrB;
674 j_coord_offsetC = DIM*jnrC;
675 j_coord_offsetD = DIM*jnrD;
677 /* load j atom coordinates */
678 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
679 x+j_coord_offsetC,x+j_coord_offsetD,
682 /* Calculate displacement vector */
683 dx10 = _mm_sub_ps(ix1,jx0);
684 dy10 = _mm_sub_ps(iy1,jy0);
685 dz10 = _mm_sub_ps(iz1,jz0);
686 dx20 = _mm_sub_ps(ix2,jx0);
687 dy20 = _mm_sub_ps(iy2,jy0);
688 dz20 = _mm_sub_ps(iz2,jz0);
689 dx30 = _mm_sub_ps(ix3,jx0);
690 dy30 = _mm_sub_ps(iy3,jy0);
691 dz30 = _mm_sub_ps(iz3,jz0);
693 /* Calculate squared distance and things based on it */
694 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
695 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
696 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
698 rinv10 = gmx_mm_invsqrt_ps(rsq10);
699 rinv20 = gmx_mm_invsqrt_ps(rsq20);
700 rinv30 = gmx_mm_invsqrt_ps(rsq30);
702 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
703 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
704 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
706 /* Load parameters for j particles */
707 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
708 charge+jnrC+0,charge+jnrD+0);
710 fjx0 = _mm_setzero_ps();
711 fjy0 = _mm_setzero_ps();
712 fjz0 = _mm_setzero_ps();
714 /**************************
715 * CALCULATE INTERACTIONS *
716 **************************/
718 r10 = _mm_mul_ps(rsq10,rinv10);
720 /* Compute parameters for interactions between i and j atoms */
721 qq10 = _mm_mul_ps(iq1,jq0);
723 /* EWALD ELECTROSTATICS */
725 /* Analytical PME correction */
726 zeta2 = _mm_mul_ps(beta2,rsq10);
727 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
728 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
729 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
730 felec = _mm_mul_ps(qq10,felec);
734 /* Update vectorial force */
735 fix1 = _mm_macc_ps(dx10,fscal,fix1);
736 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
737 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
739 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
740 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
741 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
743 /**************************
744 * CALCULATE INTERACTIONS *
745 **************************/
747 r20 = _mm_mul_ps(rsq20,rinv20);
749 /* Compute parameters for interactions between i and j atoms */
750 qq20 = _mm_mul_ps(iq2,jq0);
752 /* EWALD ELECTROSTATICS */
754 /* Analytical PME correction */
755 zeta2 = _mm_mul_ps(beta2,rsq20);
756 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
757 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
758 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
759 felec = _mm_mul_ps(qq20,felec);
763 /* Update vectorial force */
764 fix2 = _mm_macc_ps(dx20,fscal,fix2);
765 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
766 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
768 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
769 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
770 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
772 /**************************
773 * CALCULATE INTERACTIONS *
774 **************************/
776 r30 = _mm_mul_ps(rsq30,rinv30);
778 /* Compute parameters for interactions between i and j atoms */
779 qq30 = _mm_mul_ps(iq3,jq0);
781 /* EWALD ELECTROSTATICS */
783 /* Analytical PME correction */
784 zeta2 = _mm_mul_ps(beta2,rsq30);
785 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
786 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
787 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
788 felec = _mm_mul_ps(qq30,felec);
792 /* Update vectorial force */
793 fix3 = _mm_macc_ps(dx30,fscal,fix3);
794 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
795 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
797 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
798 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
799 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
801 fjptrA = f+j_coord_offsetA;
802 fjptrB = f+j_coord_offsetB;
803 fjptrC = f+j_coord_offsetC;
804 fjptrD = f+j_coord_offsetD;
806 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
808 /* Inner loop uses 84 flops */
814 /* Get j neighbor index, and coordinate index */
815 jnrlistA = jjnr[jidx];
816 jnrlistB = jjnr[jidx+1];
817 jnrlistC = jjnr[jidx+2];
818 jnrlistD = jjnr[jidx+3];
819 /* Sign of each element will be negative for non-real atoms.
820 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
821 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
823 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
824 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
825 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
826 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
827 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
828 j_coord_offsetA = DIM*jnrA;
829 j_coord_offsetB = DIM*jnrB;
830 j_coord_offsetC = DIM*jnrC;
831 j_coord_offsetD = DIM*jnrD;
833 /* load j atom coordinates */
834 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
835 x+j_coord_offsetC,x+j_coord_offsetD,
838 /* Calculate displacement vector */
839 dx10 = _mm_sub_ps(ix1,jx0);
840 dy10 = _mm_sub_ps(iy1,jy0);
841 dz10 = _mm_sub_ps(iz1,jz0);
842 dx20 = _mm_sub_ps(ix2,jx0);
843 dy20 = _mm_sub_ps(iy2,jy0);
844 dz20 = _mm_sub_ps(iz2,jz0);
845 dx30 = _mm_sub_ps(ix3,jx0);
846 dy30 = _mm_sub_ps(iy3,jy0);
847 dz30 = _mm_sub_ps(iz3,jz0);
849 /* Calculate squared distance and things based on it */
850 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
851 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
852 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
854 rinv10 = gmx_mm_invsqrt_ps(rsq10);
855 rinv20 = gmx_mm_invsqrt_ps(rsq20);
856 rinv30 = gmx_mm_invsqrt_ps(rsq30);
858 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
859 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
860 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
862 /* Load parameters for j particles */
863 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
864 charge+jnrC+0,charge+jnrD+0);
866 fjx0 = _mm_setzero_ps();
867 fjy0 = _mm_setzero_ps();
868 fjz0 = _mm_setzero_ps();
870 /**************************
871 * CALCULATE INTERACTIONS *
872 **************************/
874 r10 = _mm_mul_ps(rsq10,rinv10);
875 r10 = _mm_andnot_ps(dummy_mask,r10);
877 /* Compute parameters for interactions between i and j atoms */
878 qq10 = _mm_mul_ps(iq1,jq0);
880 /* EWALD ELECTROSTATICS */
882 /* Analytical PME correction */
883 zeta2 = _mm_mul_ps(beta2,rsq10);
884 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
885 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
886 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
887 felec = _mm_mul_ps(qq10,felec);
891 fscal = _mm_andnot_ps(dummy_mask,fscal);
893 /* Update vectorial force */
894 fix1 = _mm_macc_ps(dx10,fscal,fix1);
895 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
896 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
898 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
899 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
900 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
902 /**************************
903 * CALCULATE INTERACTIONS *
904 **************************/
906 r20 = _mm_mul_ps(rsq20,rinv20);
907 r20 = _mm_andnot_ps(dummy_mask,r20);
909 /* Compute parameters for interactions between i and j atoms */
910 qq20 = _mm_mul_ps(iq2,jq0);
912 /* EWALD ELECTROSTATICS */
914 /* Analytical PME correction */
915 zeta2 = _mm_mul_ps(beta2,rsq20);
916 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
917 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
918 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
919 felec = _mm_mul_ps(qq20,felec);
923 fscal = _mm_andnot_ps(dummy_mask,fscal);
925 /* Update vectorial force */
926 fix2 = _mm_macc_ps(dx20,fscal,fix2);
927 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
928 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
930 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
931 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
932 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
934 /**************************
935 * CALCULATE INTERACTIONS *
936 **************************/
938 r30 = _mm_mul_ps(rsq30,rinv30);
939 r30 = _mm_andnot_ps(dummy_mask,r30);
941 /* Compute parameters for interactions between i and j atoms */
942 qq30 = _mm_mul_ps(iq3,jq0);
944 /* EWALD ELECTROSTATICS */
946 /* Analytical PME correction */
947 zeta2 = _mm_mul_ps(beta2,rsq30);
948 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
949 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
950 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
951 felec = _mm_mul_ps(qq30,felec);
955 fscal = _mm_andnot_ps(dummy_mask,fscal);
957 /* Update vectorial force */
958 fix3 = _mm_macc_ps(dx30,fscal,fix3);
959 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
960 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
962 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
963 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
964 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
966 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
967 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
968 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
969 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
971 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
973 /* Inner loop uses 87 flops */
976 /* End of innermost loop */
978 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
979 f+i_coord_offset+DIM,fshift+i_shift_offset);
981 /* Increment number of inner iterations */
982 inneriter += j_index_end - j_index_start;
984 /* Outer loop uses 18 flops */
987 /* Increment number of outer iterations */
990 /* Update outer/inner flops */
992 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*87);