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_ElecRF_VdwNone_GeomW4P1_VF_avx_128_fma_single
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: None
40 * Geometry: Water4-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_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;
82 __m128 dummy_mask,cutoff_mask;
83 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
84 __m128 one = _mm_set1_ps(1.0);
85 __m128 two = _mm_set1_ps(2.0);
91 jindex = nlist->jindex;
93 shiftidx = nlist->shift;
95 shiftvec = fr->shift_vec[0];
96 fshift = fr->fshift[0];
97 facel = _mm_set1_ps(fr->epsfac);
98 charge = mdatoms->chargeA;
99 krf = _mm_set1_ps(fr->ic->k_rf);
100 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
101 crf = _mm_set1_ps(fr->ic->c_rf);
103 /* Setup water-specific parameters */
104 inr = nlist->iinr[0];
105 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
106 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
107 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
109 /* Avoid stupid compiler warnings */
110 jnrA = jnrB = jnrC = jnrD = 0;
119 for(iidx=0;iidx<4*DIM;iidx++)
124 /* Start outer loop over neighborlists */
125 for(iidx=0; iidx<nri; iidx++)
127 /* Load shift vector for this list */
128 i_shift_offset = DIM*shiftidx[iidx];
130 /* Load limits for loop over neighbors */
131 j_index_start = jindex[iidx];
132 j_index_end = jindex[iidx+1];
134 /* Get outer coordinate index */
136 i_coord_offset = DIM*inr;
138 /* Load i particle coords and add shift vector */
139 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
140 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
142 fix1 = _mm_setzero_ps();
143 fiy1 = _mm_setzero_ps();
144 fiz1 = _mm_setzero_ps();
145 fix2 = _mm_setzero_ps();
146 fiy2 = _mm_setzero_ps();
147 fiz2 = _mm_setzero_ps();
148 fix3 = _mm_setzero_ps();
149 fiy3 = _mm_setzero_ps();
150 fiz3 = _mm_setzero_ps();
152 /* Reset potential sums */
153 velecsum = _mm_setzero_ps();
155 /* Start inner kernel loop */
156 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
159 /* Get j neighbor index, and coordinate index */
164 j_coord_offsetA = DIM*jnrA;
165 j_coord_offsetB = DIM*jnrB;
166 j_coord_offsetC = DIM*jnrC;
167 j_coord_offsetD = DIM*jnrD;
169 /* load j atom coordinates */
170 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
171 x+j_coord_offsetC,x+j_coord_offsetD,
174 /* Calculate displacement vector */
175 dx10 = _mm_sub_ps(ix1,jx0);
176 dy10 = _mm_sub_ps(iy1,jy0);
177 dz10 = _mm_sub_ps(iz1,jz0);
178 dx20 = _mm_sub_ps(ix2,jx0);
179 dy20 = _mm_sub_ps(iy2,jy0);
180 dz20 = _mm_sub_ps(iz2,jz0);
181 dx30 = _mm_sub_ps(ix3,jx0);
182 dy30 = _mm_sub_ps(iy3,jy0);
183 dz30 = _mm_sub_ps(iz3,jz0);
185 /* Calculate squared distance and things based on it */
186 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
187 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
188 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
190 rinv10 = gmx_mm_invsqrt_ps(rsq10);
191 rinv20 = gmx_mm_invsqrt_ps(rsq20);
192 rinv30 = gmx_mm_invsqrt_ps(rsq30);
194 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
195 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
196 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
198 /* Load parameters for j particles */
199 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
200 charge+jnrC+0,charge+jnrD+0);
202 fjx0 = _mm_setzero_ps();
203 fjy0 = _mm_setzero_ps();
204 fjz0 = _mm_setzero_ps();
206 /**************************
207 * CALCULATE INTERACTIONS *
208 **************************/
210 /* Compute parameters for interactions between i and j atoms */
211 qq10 = _mm_mul_ps(iq1,jq0);
213 /* REACTION-FIELD ELECTROSTATICS */
214 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
215 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
217 /* Update potential sum for this i atom from the interaction with this j atom. */
218 velecsum = _mm_add_ps(velecsum,velec);
222 /* Update vectorial force */
223 fix1 = _mm_macc_ps(dx10,fscal,fix1);
224 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
225 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
227 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
228 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
229 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
231 /**************************
232 * CALCULATE INTERACTIONS *
233 **************************/
235 /* Compute parameters for interactions between i and j atoms */
236 qq20 = _mm_mul_ps(iq2,jq0);
238 /* REACTION-FIELD ELECTROSTATICS */
239 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
240 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
242 /* Update potential sum for this i atom from the interaction with this j atom. */
243 velecsum = _mm_add_ps(velecsum,velec);
247 /* Update vectorial force */
248 fix2 = _mm_macc_ps(dx20,fscal,fix2);
249 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
250 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
252 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
253 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
254 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
256 /**************************
257 * CALCULATE INTERACTIONS *
258 **************************/
260 /* Compute parameters for interactions between i and j atoms */
261 qq30 = _mm_mul_ps(iq3,jq0);
263 /* REACTION-FIELD ELECTROSTATICS */
264 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
265 felec = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
267 /* Update potential sum for this i atom from the interaction with this j atom. */
268 velecsum = _mm_add_ps(velecsum,velec);
272 /* Update vectorial force */
273 fix3 = _mm_macc_ps(dx30,fscal,fix3);
274 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
275 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
277 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
278 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
279 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
281 fjptrA = f+j_coord_offsetA;
282 fjptrB = f+j_coord_offsetB;
283 fjptrC = f+j_coord_offsetC;
284 fjptrD = f+j_coord_offsetD;
286 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
288 /* Inner loop uses 105 flops */
294 /* Get j neighbor index, and coordinate index */
295 jnrlistA = jjnr[jidx];
296 jnrlistB = jjnr[jidx+1];
297 jnrlistC = jjnr[jidx+2];
298 jnrlistD = jjnr[jidx+3];
299 /* Sign of each element will be negative for non-real atoms.
300 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
301 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
303 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
304 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
305 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
306 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
307 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
308 j_coord_offsetA = DIM*jnrA;
309 j_coord_offsetB = DIM*jnrB;
310 j_coord_offsetC = DIM*jnrC;
311 j_coord_offsetD = DIM*jnrD;
313 /* load j atom coordinates */
314 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
315 x+j_coord_offsetC,x+j_coord_offsetD,
318 /* Calculate displacement vector */
319 dx10 = _mm_sub_ps(ix1,jx0);
320 dy10 = _mm_sub_ps(iy1,jy0);
321 dz10 = _mm_sub_ps(iz1,jz0);
322 dx20 = _mm_sub_ps(ix2,jx0);
323 dy20 = _mm_sub_ps(iy2,jy0);
324 dz20 = _mm_sub_ps(iz2,jz0);
325 dx30 = _mm_sub_ps(ix3,jx0);
326 dy30 = _mm_sub_ps(iy3,jy0);
327 dz30 = _mm_sub_ps(iz3,jz0);
329 /* Calculate squared distance and things based on it */
330 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
331 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
332 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
334 rinv10 = gmx_mm_invsqrt_ps(rsq10);
335 rinv20 = gmx_mm_invsqrt_ps(rsq20);
336 rinv30 = gmx_mm_invsqrt_ps(rsq30);
338 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
339 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
340 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
342 /* Load parameters for j particles */
343 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
344 charge+jnrC+0,charge+jnrD+0);
346 fjx0 = _mm_setzero_ps();
347 fjy0 = _mm_setzero_ps();
348 fjz0 = _mm_setzero_ps();
350 /**************************
351 * CALCULATE INTERACTIONS *
352 **************************/
354 /* Compute parameters for interactions between i and j atoms */
355 qq10 = _mm_mul_ps(iq1,jq0);
357 /* REACTION-FIELD ELECTROSTATICS */
358 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
359 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
361 /* Update potential sum for this i atom from the interaction with this j atom. */
362 velec = _mm_andnot_ps(dummy_mask,velec);
363 velecsum = _mm_add_ps(velecsum,velec);
367 fscal = _mm_andnot_ps(dummy_mask,fscal);
369 /* Update vectorial force */
370 fix1 = _mm_macc_ps(dx10,fscal,fix1);
371 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
372 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
374 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
375 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
376 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
378 /**************************
379 * CALCULATE INTERACTIONS *
380 **************************/
382 /* Compute parameters for interactions between i and j atoms */
383 qq20 = _mm_mul_ps(iq2,jq0);
385 /* REACTION-FIELD ELECTROSTATICS */
386 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
387 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
389 /* Update potential sum for this i atom from the interaction with this j atom. */
390 velec = _mm_andnot_ps(dummy_mask,velec);
391 velecsum = _mm_add_ps(velecsum,velec);
395 fscal = _mm_andnot_ps(dummy_mask,fscal);
397 /* Update vectorial force */
398 fix2 = _mm_macc_ps(dx20,fscal,fix2);
399 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
400 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
402 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
403 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
404 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
406 /**************************
407 * CALCULATE INTERACTIONS *
408 **************************/
410 /* Compute parameters for interactions between i and j atoms */
411 qq30 = _mm_mul_ps(iq3,jq0);
413 /* REACTION-FIELD ELECTROSTATICS */
414 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
415 felec = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
417 /* Update potential sum for this i atom from the interaction with this j atom. */
418 velec = _mm_andnot_ps(dummy_mask,velec);
419 velecsum = _mm_add_ps(velecsum,velec);
423 fscal = _mm_andnot_ps(dummy_mask,fscal);
425 /* Update vectorial force */
426 fix3 = _mm_macc_ps(dx30,fscal,fix3);
427 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
428 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
430 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
431 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
432 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
434 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
435 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
436 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
437 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
439 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
441 /* Inner loop uses 105 flops */
444 /* End of innermost loop */
446 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
447 f+i_coord_offset+DIM,fshift+i_shift_offset);
450 /* Update potential energies */
451 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
453 /* Increment number of inner iterations */
454 inneriter += j_index_end - j_index_start;
456 /* Outer loop uses 19 flops */
459 /* Increment number of outer iterations */
462 /* Update outer/inner flops */
464 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*105);
467 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_single
468 * Electrostatics interaction: ReactionField
469 * VdW interaction: None
470 * Geometry: Water4-Particle
471 * Calculate force/pot: Force
474 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_avx_128_fma_single
475 (t_nblist * gmx_restrict nlist,
476 rvec * gmx_restrict xx,
477 rvec * gmx_restrict ff,
478 t_forcerec * gmx_restrict fr,
479 t_mdatoms * gmx_restrict mdatoms,
480 nb_kernel_data_t * gmx_restrict kernel_data,
481 t_nrnb * gmx_restrict nrnb)
483 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
484 * just 0 for non-waters.
485 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
486 * jnr indices corresponding to data put in the four positions in the SIMD register.
488 int i_shift_offset,i_coord_offset,outeriter,inneriter;
489 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
490 int jnrA,jnrB,jnrC,jnrD;
491 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
492 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
493 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
495 real *shiftvec,*fshift,*x,*f;
496 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
498 __m128 fscal,rcutoff,rcutoff2,jidxall;
500 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
502 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
504 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
505 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
506 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
507 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
508 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
509 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
510 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
512 __m128 dummy_mask,cutoff_mask;
513 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
514 __m128 one = _mm_set1_ps(1.0);
515 __m128 two = _mm_set1_ps(2.0);
521 jindex = nlist->jindex;
523 shiftidx = nlist->shift;
525 shiftvec = fr->shift_vec[0];
526 fshift = fr->fshift[0];
527 facel = _mm_set1_ps(fr->epsfac);
528 charge = mdatoms->chargeA;
529 krf = _mm_set1_ps(fr->ic->k_rf);
530 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
531 crf = _mm_set1_ps(fr->ic->c_rf);
533 /* Setup water-specific parameters */
534 inr = nlist->iinr[0];
535 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
536 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
537 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
539 /* Avoid stupid compiler warnings */
540 jnrA = jnrB = jnrC = jnrD = 0;
549 for(iidx=0;iidx<4*DIM;iidx++)
554 /* Start outer loop over neighborlists */
555 for(iidx=0; iidx<nri; iidx++)
557 /* Load shift vector for this list */
558 i_shift_offset = DIM*shiftidx[iidx];
560 /* Load limits for loop over neighbors */
561 j_index_start = jindex[iidx];
562 j_index_end = jindex[iidx+1];
564 /* Get outer coordinate index */
566 i_coord_offset = DIM*inr;
568 /* Load i particle coords and add shift vector */
569 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
570 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
572 fix1 = _mm_setzero_ps();
573 fiy1 = _mm_setzero_ps();
574 fiz1 = _mm_setzero_ps();
575 fix2 = _mm_setzero_ps();
576 fiy2 = _mm_setzero_ps();
577 fiz2 = _mm_setzero_ps();
578 fix3 = _mm_setzero_ps();
579 fiy3 = _mm_setzero_ps();
580 fiz3 = _mm_setzero_ps();
582 /* Start inner kernel loop */
583 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
586 /* Get j neighbor index, and coordinate index */
591 j_coord_offsetA = DIM*jnrA;
592 j_coord_offsetB = DIM*jnrB;
593 j_coord_offsetC = DIM*jnrC;
594 j_coord_offsetD = DIM*jnrD;
596 /* load j atom coordinates */
597 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
598 x+j_coord_offsetC,x+j_coord_offsetD,
601 /* Calculate displacement vector */
602 dx10 = _mm_sub_ps(ix1,jx0);
603 dy10 = _mm_sub_ps(iy1,jy0);
604 dz10 = _mm_sub_ps(iz1,jz0);
605 dx20 = _mm_sub_ps(ix2,jx0);
606 dy20 = _mm_sub_ps(iy2,jy0);
607 dz20 = _mm_sub_ps(iz2,jz0);
608 dx30 = _mm_sub_ps(ix3,jx0);
609 dy30 = _mm_sub_ps(iy3,jy0);
610 dz30 = _mm_sub_ps(iz3,jz0);
612 /* Calculate squared distance and things based on it */
613 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
614 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
615 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
617 rinv10 = gmx_mm_invsqrt_ps(rsq10);
618 rinv20 = gmx_mm_invsqrt_ps(rsq20);
619 rinv30 = gmx_mm_invsqrt_ps(rsq30);
621 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
622 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
623 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
625 /* Load parameters for j particles */
626 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
627 charge+jnrC+0,charge+jnrD+0);
629 fjx0 = _mm_setzero_ps();
630 fjy0 = _mm_setzero_ps();
631 fjz0 = _mm_setzero_ps();
633 /**************************
634 * CALCULATE INTERACTIONS *
635 **************************/
637 /* Compute parameters for interactions between i and j atoms */
638 qq10 = _mm_mul_ps(iq1,jq0);
640 /* REACTION-FIELD ELECTROSTATICS */
641 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
645 /* Update vectorial force */
646 fix1 = _mm_macc_ps(dx10,fscal,fix1);
647 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
648 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
650 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
651 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
652 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
654 /**************************
655 * CALCULATE INTERACTIONS *
656 **************************/
658 /* Compute parameters for interactions between i and j atoms */
659 qq20 = _mm_mul_ps(iq2,jq0);
661 /* REACTION-FIELD ELECTROSTATICS */
662 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
666 /* Update vectorial force */
667 fix2 = _mm_macc_ps(dx20,fscal,fix2);
668 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
669 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
671 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
672 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
673 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
675 /**************************
676 * CALCULATE INTERACTIONS *
677 **************************/
679 /* Compute parameters for interactions between i and j atoms */
680 qq30 = _mm_mul_ps(iq3,jq0);
682 /* REACTION-FIELD ELECTROSTATICS */
683 felec = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
687 /* Update vectorial force */
688 fix3 = _mm_macc_ps(dx30,fscal,fix3);
689 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
690 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
692 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
693 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
694 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
696 fjptrA = f+j_coord_offsetA;
697 fjptrB = f+j_coord_offsetB;
698 fjptrC = f+j_coord_offsetC;
699 fjptrD = f+j_coord_offsetD;
701 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
703 /* Inner loop uses 90 flops */
709 /* Get j neighbor index, and coordinate index */
710 jnrlistA = jjnr[jidx];
711 jnrlistB = jjnr[jidx+1];
712 jnrlistC = jjnr[jidx+2];
713 jnrlistD = jjnr[jidx+3];
714 /* Sign of each element will be negative for non-real atoms.
715 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
716 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
718 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
719 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
720 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
721 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
722 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
723 j_coord_offsetA = DIM*jnrA;
724 j_coord_offsetB = DIM*jnrB;
725 j_coord_offsetC = DIM*jnrC;
726 j_coord_offsetD = DIM*jnrD;
728 /* load j atom coordinates */
729 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
730 x+j_coord_offsetC,x+j_coord_offsetD,
733 /* Calculate displacement vector */
734 dx10 = _mm_sub_ps(ix1,jx0);
735 dy10 = _mm_sub_ps(iy1,jy0);
736 dz10 = _mm_sub_ps(iz1,jz0);
737 dx20 = _mm_sub_ps(ix2,jx0);
738 dy20 = _mm_sub_ps(iy2,jy0);
739 dz20 = _mm_sub_ps(iz2,jz0);
740 dx30 = _mm_sub_ps(ix3,jx0);
741 dy30 = _mm_sub_ps(iy3,jy0);
742 dz30 = _mm_sub_ps(iz3,jz0);
744 /* Calculate squared distance and things based on it */
745 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
746 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
747 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
749 rinv10 = gmx_mm_invsqrt_ps(rsq10);
750 rinv20 = gmx_mm_invsqrt_ps(rsq20);
751 rinv30 = gmx_mm_invsqrt_ps(rsq30);
753 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
754 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
755 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
757 /* Load parameters for j particles */
758 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
759 charge+jnrC+0,charge+jnrD+0);
761 fjx0 = _mm_setzero_ps();
762 fjy0 = _mm_setzero_ps();
763 fjz0 = _mm_setzero_ps();
765 /**************************
766 * CALCULATE INTERACTIONS *
767 **************************/
769 /* Compute parameters for interactions between i and j atoms */
770 qq10 = _mm_mul_ps(iq1,jq0);
772 /* REACTION-FIELD ELECTROSTATICS */
773 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
777 fscal = _mm_andnot_ps(dummy_mask,fscal);
779 /* Update vectorial force */
780 fix1 = _mm_macc_ps(dx10,fscal,fix1);
781 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
782 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
784 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
785 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
786 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
788 /**************************
789 * CALCULATE INTERACTIONS *
790 **************************/
792 /* Compute parameters for interactions between i and j atoms */
793 qq20 = _mm_mul_ps(iq2,jq0);
795 /* REACTION-FIELD ELECTROSTATICS */
796 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
800 fscal = _mm_andnot_ps(dummy_mask,fscal);
802 /* Update vectorial force */
803 fix2 = _mm_macc_ps(dx20,fscal,fix2);
804 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
805 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
807 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
808 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
809 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
811 /**************************
812 * CALCULATE INTERACTIONS *
813 **************************/
815 /* Compute parameters for interactions between i and j atoms */
816 qq30 = _mm_mul_ps(iq3,jq0);
818 /* REACTION-FIELD ELECTROSTATICS */
819 felec = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
823 fscal = _mm_andnot_ps(dummy_mask,fscal);
825 /* Update vectorial force */
826 fix3 = _mm_macc_ps(dx30,fscal,fix3);
827 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
828 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
830 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
831 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
832 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
834 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
835 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
836 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
837 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
839 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
841 /* Inner loop uses 90 flops */
844 /* End of innermost loop */
846 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
847 f+i_coord_offset+DIM,fshift+i_shift_offset);
849 /* Increment number of inner iterations */
850 inneriter += j_index_end - j_index_start;
852 /* Outer loop uses 18 flops */
855 /* Increment number of outer iterations */
858 /* Update outer/inner flops */
860 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*90);