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_VdwNone_GeomW3P1_VF_sse2_single
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: None
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_sse2_single
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int 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;
72 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
73 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
75 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
76 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
77 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
79 __m128 dummy_mask,cutoff_mask;
80 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
81 __m128 one = _mm_set1_ps(1.0);
82 __m128 two = _mm_set1_ps(2.0);
88 jindex = nlist->jindex;
90 shiftidx = nlist->shift;
92 shiftvec = fr->shift_vec[0];
93 fshift = fr->fshift[0];
94 facel = _mm_set1_ps(fr->epsfac);
95 charge = mdatoms->chargeA;
97 /* Setup water-specific parameters */
99 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
100 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
101 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
103 /* Avoid stupid compiler warnings */
104 jnrA = jnrB = jnrC = jnrD = 0;
113 /* Start outer loop over neighborlists */
114 for(iidx=0; iidx<nri; iidx++)
116 /* Load shift vector for this list */
117 i_shift_offset = DIM*shiftidx[iidx];
118 shX = shiftvec[i_shift_offset+XX];
119 shY = shiftvec[i_shift_offset+YY];
120 shZ = shiftvec[i_shift_offset+ZZ];
122 /* Load limits for loop over neighbors */
123 j_index_start = jindex[iidx];
124 j_index_end = jindex[iidx+1];
126 /* Get outer coordinate index */
128 i_coord_offset = DIM*inr;
130 /* Load i particle coords and add shift vector */
131 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
132 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
133 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
134 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
135 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
136 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
137 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
138 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
139 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
141 fix0 = _mm_setzero_ps();
142 fiy0 = _mm_setzero_ps();
143 fiz0 = _mm_setzero_ps();
144 fix1 = _mm_setzero_ps();
145 fiy1 = _mm_setzero_ps();
146 fiz1 = _mm_setzero_ps();
147 fix2 = _mm_setzero_ps();
148 fiy2 = _mm_setzero_ps();
149 fiz2 = _mm_setzero_ps();
151 /* Reset potential sums */
152 velecsum = _mm_setzero_ps();
154 /* Start inner kernel loop */
155 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
158 /* 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 dx00 = _mm_sub_ps(ix0,jx0);
176 dy00 = _mm_sub_ps(iy0,jy0);
177 dz00 = _mm_sub_ps(iz0,jz0);
178 dx10 = _mm_sub_ps(ix1,jx0);
179 dy10 = _mm_sub_ps(iy1,jy0);
180 dz10 = _mm_sub_ps(iz1,jz0);
181 dx20 = _mm_sub_ps(ix2,jx0);
182 dy20 = _mm_sub_ps(iy2,jy0);
183 dz20 = _mm_sub_ps(iz2,jz0);
185 /* Calculate squared distance and things based on it */
186 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
187 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
188 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
190 rinv00 = gmx_mm_invsqrt_ps(rsq00);
191 rinv10 = gmx_mm_invsqrt_ps(rsq10);
192 rinv20 = gmx_mm_invsqrt_ps(rsq20);
194 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
195 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
196 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
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 /**************************
203 * CALCULATE INTERACTIONS *
204 **************************/
206 /* Compute parameters for interactions between i and j atoms */
207 qq00 = _mm_mul_ps(iq0,jq0);
209 /* COULOMB ELECTROSTATICS */
210 velec = _mm_mul_ps(qq00,rinv00);
211 felec = _mm_mul_ps(velec,rinvsq00);
213 /* Update potential sum for this i atom from the interaction with this j atom. */
214 velecsum = _mm_add_ps(velecsum,velec);
218 /* Calculate temporary vectorial force */
219 tx = _mm_mul_ps(fscal,dx00);
220 ty = _mm_mul_ps(fscal,dy00);
221 tz = _mm_mul_ps(fscal,dz00);
223 /* Update vectorial force */
224 fix0 = _mm_add_ps(fix0,tx);
225 fiy0 = _mm_add_ps(fiy0,ty);
226 fiz0 = _mm_add_ps(fiz0,tz);
228 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
229 f+j_coord_offsetC,f+j_coord_offsetD,
232 /**************************
233 * CALCULATE INTERACTIONS *
234 **************************/
236 /* Compute parameters for interactions between i and j atoms */
237 qq10 = _mm_mul_ps(iq1,jq0);
239 /* COULOMB ELECTROSTATICS */
240 velec = _mm_mul_ps(qq10,rinv10);
241 felec = _mm_mul_ps(velec,rinvsq10);
243 /* Update potential sum for this i atom from the interaction with this j atom. */
244 velecsum = _mm_add_ps(velecsum,velec);
248 /* Calculate temporary vectorial force */
249 tx = _mm_mul_ps(fscal,dx10);
250 ty = _mm_mul_ps(fscal,dy10);
251 tz = _mm_mul_ps(fscal,dz10);
253 /* Update vectorial force */
254 fix1 = _mm_add_ps(fix1,tx);
255 fiy1 = _mm_add_ps(fiy1,ty);
256 fiz1 = _mm_add_ps(fiz1,tz);
258 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
259 f+j_coord_offsetC,f+j_coord_offsetD,
262 /**************************
263 * CALCULATE INTERACTIONS *
264 **************************/
266 /* Compute parameters for interactions between i and j atoms */
267 qq20 = _mm_mul_ps(iq2,jq0);
269 /* COULOMB ELECTROSTATICS */
270 velec = _mm_mul_ps(qq20,rinv20);
271 felec = _mm_mul_ps(velec,rinvsq20);
273 /* Update potential sum for this i atom from the interaction with this j atom. */
274 velecsum = _mm_add_ps(velecsum,velec);
278 /* Calculate temporary vectorial force */
279 tx = _mm_mul_ps(fscal,dx20);
280 ty = _mm_mul_ps(fscal,dy20);
281 tz = _mm_mul_ps(fscal,dz20);
283 /* Update vectorial force */
284 fix2 = _mm_add_ps(fix2,tx);
285 fiy2 = _mm_add_ps(fiy2,ty);
286 fiz2 = _mm_add_ps(fiz2,tz);
288 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
289 f+j_coord_offsetC,f+j_coord_offsetD,
292 /* Inner loop uses 84 flops */
298 /* Get j neighbor index, and coordinate index */
304 /* Sign of each element will be negative for non-real atoms.
305 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
306 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
308 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
309 jnrA = (jnrA>=0) ? jnrA : 0;
310 jnrB = (jnrB>=0) ? jnrB : 0;
311 jnrC = (jnrC>=0) ? jnrC : 0;
312 jnrD = (jnrD>=0) ? jnrD : 0;
314 j_coord_offsetA = DIM*jnrA;
315 j_coord_offsetB = DIM*jnrB;
316 j_coord_offsetC = DIM*jnrC;
317 j_coord_offsetD = DIM*jnrD;
319 /* load j atom coordinates */
320 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
321 x+j_coord_offsetC,x+j_coord_offsetD,
324 /* Calculate displacement vector */
325 dx00 = _mm_sub_ps(ix0,jx0);
326 dy00 = _mm_sub_ps(iy0,jy0);
327 dz00 = _mm_sub_ps(iz0,jz0);
328 dx10 = _mm_sub_ps(ix1,jx0);
329 dy10 = _mm_sub_ps(iy1,jy0);
330 dz10 = _mm_sub_ps(iz1,jz0);
331 dx20 = _mm_sub_ps(ix2,jx0);
332 dy20 = _mm_sub_ps(iy2,jy0);
333 dz20 = _mm_sub_ps(iz2,jz0);
335 /* Calculate squared distance and things based on it */
336 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
337 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
338 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
340 rinv00 = gmx_mm_invsqrt_ps(rsq00);
341 rinv10 = gmx_mm_invsqrt_ps(rsq10);
342 rinv20 = gmx_mm_invsqrt_ps(rsq20);
344 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
345 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
346 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
348 /* Load parameters for j particles */
349 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
350 charge+jnrC+0,charge+jnrD+0);
352 /**************************
353 * CALCULATE INTERACTIONS *
354 **************************/
356 /* Compute parameters for interactions between i and j atoms */
357 qq00 = _mm_mul_ps(iq0,jq0);
359 /* COULOMB ELECTROSTATICS */
360 velec = _mm_mul_ps(qq00,rinv00);
361 felec = _mm_mul_ps(velec,rinvsq00);
363 /* Update potential sum for this i atom from the interaction with this j atom. */
364 velec = _mm_andnot_ps(dummy_mask,velec);
365 velecsum = _mm_add_ps(velecsum,velec);
369 fscal = _mm_andnot_ps(dummy_mask,fscal);
371 /* Calculate temporary vectorial force */
372 tx = _mm_mul_ps(fscal,dx00);
373 ty = _mm_mul_ps(fscal,dy00);
374 tz = _mm_mul_ps(fscal,dz00);
376 /* Update vectorial force */
377 fix0 = _mm_add_ps(fix0,tx);
378 fiy0 = _mm_add_ps(fiy0,ty);
379 fiz0 = _mm_add_ps(fiz0,tz);
381 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
382 f+j_coord_offsetC,f+j_coord_offsetD,
385 /**************************
386 * CALCULATE INTERACTIONS *
387 **************************/
389 /* Compute parameters for interactions between i and j atoms */
390 qq10 = _mm_mul_ps(iq1,jq0);
392 /* COULOMB ELECTROSTATICS */
393 velec = _mm_mul_ps(qq10,rinv10);
394 felec = _mm_mul_ps(velec,rinvsq10);
396 /* Update potential sum for this i atom from the interaction with this j atom. */
397 velec = _mm_andnot_ps(dummy_mask,velec);
398 velecsum = _mm_add_ps(velecsum,velec);
402 fscal = _mm_andnot_ps(dummy_mask,fscal);
404 /* Calculate temporary vectorial force */
405 tx = _mm_mul_ps(fscal,dx10);
406 ty = _mm_mul_ps(fscal,dy10);
407 tz = _mm_mul_ps(fscal,dz10);
409 /* Update vectorial force */
410 fix1 = _mm_add_ps(fix1,tx);
411 fiy1 = _mm_add_ps(fiy1,ty);
412 fiz1 = _mm_add_ps(fiz1,tz);
414 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
415 f+j_coord_offsetC,f+j_coord_offsetD,
418 /**************************
419 * CALCULATE INTERACTIONS *
420 **************************/
422 /* Compute parameters for interactions between i and j atoms */
423 qq20 = _mm_mul_ps(iq2,jq0);
425 /* COULOMB ELECTROSTATICS */
426 velec = _mm_mul_ps(qq20,rinv20);
427 felec = _mm_mul_ps(velec,rinvsq20);
429 /* Update potential sum for this i atom from the interaction with this j atom. */
430 velec = _mm_andnot_ps(dummy_mask,velec);
431 velecsum = _mm_add_ps(velecsum,velec);
435 fscal = _mm_andnot_ps(dummy_mask,fscal);
437 /* Calculate temporary vectorial force */
438 tx = _mm_mul_ps(fscal,dx20);
439 ty = _mm_mul_ps(fscal,dy20);
440 tz = _mm_mul_ps(fscal,dz20);
442 /* Update vectorial force */
443 fix2 = _mm_add_ps(fix2,tx);
444 fiy2 = _mm_add_ps(fiy2,ty);
445 fiz2 = _mm_add_ps(fiz2,tz);
447 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
448 f+j_coord_offsetC,f+j_coord_offsetD,
451 /* Inner loop uses 84 flops */
454 /* End of innermost loop */
456 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
457 f+i_coord_offset,fshift+i_shift_offset);
460 /* Update potential energies */
461 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
463 /* Increment number of inner iterations */
464 inneriter += j_index_end - j_index_start;
466 /* Outer loop uses 28 flops */
469 /* Increment number of outer iterations */
472 /* Update outer/inner flops */
474 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*28 + inneriter*84);
477 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_sse2_single
478 * Electrostatics interaction: Coulomb
479 * VdW interaction: None
480 * Geometry: Water3-Particle
481 * Calculate force/pot: Force
484 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_sse2_single
485 (t_nblist * gmx_restrict nlist,
486 rvec * gmx_restrict xx,
487 rvec * gmx_restrict ff,
488 t_forcerec * gmx_restrict fr,
489 t_mdatoms * gmx_restrict mdatoms,
490 nb_kernel_data_t * gmx_restrict kernel_data,
491 t_nrnb * gmx_restrict nrnb)
493 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
494 * just 0 for non-waters.
495 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
496 * jnr indices corresponding to data put in the four positions in the SIMD register.
498 int i_shift_offset,i_coord_offset,outeriter,inneriter;
499 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
500 int jnrA,jnrB,jnrC,jnrD;
501 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
502 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
503 real shX,shY,shZ,rcutoff_scalar;
504 real *shiftvec,*fshift,*x,*f;
505 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
507 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
509 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
511 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
512 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
513 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
514 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
515 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
516 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
517 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
519 __m128 dummy_mask,cutoff_mask;
520 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
521 __m128 one = _mm_set1_ps(1.0);
522 __m128 two = _mm_set1_ps(2.0);
528 jindex = nlist->jindex;
530 shiftidx = nlist->shift;
532 shiftvec = fr->shift_vec[0];
533 fshift = fr->fshift[0];
534 facel = _mm_set1_ps(fr->epsfac);
535 charge = mdatoms->chargeA;
537 /* Setup water-specific parameters */
538 inr = nlist->iinr[0];
539 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
540 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
541 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
543 /* Avoid stupid compiler warnings */
544 jnrA = jnrB = jnrC = jnrD = 0;
553 /* Start outer loop over neighborlists */
554 for(iidx=0; iidx<nri; iidx++)
556 /* Load shift vector for this list */
557 i_shift_offset = DIM*shiftidx[iidx];
558 shX = shiftvec[i_shift_offset+XX];
559 shY = shiftvec[i_shift_offset+YY];
560 shZ = shiftvec[i_shift_offset+ZZ];
562 /* Load limits for loop over neighbors */
563 j_index_start = jindex[iidx];
564 j_index_end = jindex[iidx+1];
566 /* Get outer coordinate index */
568 i_coord_offset = DIM*inr;
570 /* Load i particle coords and add shift vector */
571 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
572 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
573 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
574 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
575 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
576 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
577 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
578 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
579 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
581 fix0 = _mm_setzero_ps();
582 fiy0 = _mm_setzero_ps();
583 fiz0 = _mm_setzero_ps();
584 fix1 = _mm_setzero_ps();
585 fiy1 = _mm_setzero_ps();
586 fiz1 = _mm_setzero_ps();
587 fix2 = _mm_setzero_ps();
588 fiy2 = _mm_setzero_ps();
589 fiz2 = _mm_setzero_ps();
591 /* Start inner kernel loop */
592 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
595 /* Get j neighbor index, and coordinate index */
601 j_coord_offsetA = DIM*jnrA;
602 j_coord_offsetB = DIM*jnrB;
603 j_coord_offsetC = DIM*jnrC;
604 j_coord_offsetD = DIM*jnrD;
606 /* load j atom coordinates */
607 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
608 x+j_coord_offsetC,x+j_coord_offsetD,
611 /* Calculate displacement vector */
612 dx00 = _mm_sub_ps(ix0,jx0);
613 dy00 = _mm_sub_ps(iy0,jy0);
614 dz00 = _mm_sub_ps(iz0,jz0);
615 dx10 = _mm_sub_ps(ix1,jx0);
616 dy10 = _mm_sub_ps(iy1,jy0);
617 dz10 = _mm_sub_ps(iz1,jz0);
618 dx20 = _mm_sub_ps(ix2,jx0);
619 dy20 = _mm_sub_ps(iy2,jy0);
620 dz20 = _mm_sub_ps(iz2,jz0);
622 /* Calculate squared distance and things based on it */
623 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
624 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
625 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
627 rinv00 = gmx_mm_invsqrt_ps(rsq00);
628 rinv10 = gmx_mm_invsqrt_ps(rsq10);
629 rinv20 = gmx_mm_invsqrt_ps(rsq20);
631 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
632 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
633 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
635 /* Load parameters for j particles */
636 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
637 charge+jnrC+0,charge+jnrD+0);
639 /**************************
640 * CALCULATE INTERACTIONS *
641 **************************/
643 /* Compute parameters for interactions between i and j atoms */
644 qq00 = _mm_mul_ps(iq0,jq0);
646 /* COULOMB ELECTROSTATICS */
647 velec = _mm_mul_ps(qq00,rinv00);
648 felec = _mm_mul_ps(velec,rinvsq00);
652 /* Calculate temporary vectorial force */
653 tx = _mm_mul_ps(fscal,dx00);
654 ty = _mm_mul_ps(fscal,dy00);
655 tz = _mm_mul_ps(fscal,dz00);
657 /* Update vectorial force */
658 fix0 = _mm_add_ps(fix0,tx);
659 fiy0 = _mm_add_ps(fiy0,ty);
660 fiz0 = _mm_add_ps(fiz0,tz);
662 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
663 f+j_coord_offsetC,f+j_coord_offsetD,
666 /**************************
667 * CALCULATE INTERACTIONS *
668 **************************/
670 /* Compute parameters for interactions between i and j atoms */
671 qq10 = _mm_mul_ps(iq1,jq0);
673 /* COULOMB ELECTROSTATICS */
674 velec = _mm_mul_ps(qq10,rinv10);
675 felec = _mm_mul_ps(velec,rinvsq10);
679 /* Calculate temporary vectorial force */
680 tx = _mm_mul_ps(fscal,dx10);
681 ty = _mm_mul_ps(fscal,dy10);
682 tz = _mm_mul_ps(fscal,dz10);
684 /* Update vectorial force */
685 fix1 = _mm_add_ps(fix1,tx);
686 fiy1 = _mm_add_ps(fiy1,ty);
687 fiz1 = _mm_add_ps(fiz1,tz);
689 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
690 f+j_coord_offsetC,f+j_coord_offsetD,
693 /**************************
694 * CALCULATE INTERACTIONS *
695 **************************/
697 /* Compute parameters for interactions between i and j atoms */
698 qq20 = _mm_mul_ps(iq2,jq0);
700 /* COULOMB ELECTROSTATICS */
701 velec = _mm_mul_ps(qq20,rinv20);
702 felec = _mm_mul_ps(velec,rinvsq20);
706 /* Calculate temporary vectorial force */
707 tx = _mm_mul_ps(fscal,dx20);
708 ty = _mm_mul_ps(fscal,dy20);
709 tz = _mm_mul_ps(fscal,dz20);
711 /* Update vectorial force */
712 fix2 = _mm_add_ps(fix2,tx);
713 fiy2 = _mm_add_ps(fiy2,ty);
714 fiz2 = _mm_add_ps(fiz2,tz);
716 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
717 f+j_coord_offsetC,f+j_coord_offsetD,
720 /* Inner loop uses 81 flops */
726 /* Get j neighbor index, and coordinate index */
732 /* Sign of each element will be negative for non-real atoms.
733 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
734 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
736 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
737 jnrA = (jnrA>=0) ? jnrA : 0;
738 jnrB = (jnrB>=0) ? jnrB : 0;
739 jnrC = (jnrC>=0) ? jnrC : 0;
740 jnrD = (jnrD>=0) ? jnrD : 0;
742 j_coord_offsetA = DIM*jnrA;
743 j_coord_offsetB = DIM*jnrB;
744 j_coord_offsetC = DIM*jnrC;
745 j_coord_offsetD = DIM*jnrD;
747 /* load j atom coordinates */
748 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
749 x+j_coord_offsetC,x+j_coord_offsetD,
752 /* Calculate displacement vector */
753 dx00 = _mm_sub_ps(ix0,jx0);
754 dy00 = _mm_sub_ps(iy0,jy0);
755 dz00 = _mm_sub_ps(iz0,jz0);
756 dx10 = _mm_sub_ps(ix1,jx0);
757 dy10 = _mm_sub_ps(iy1,jy0);
758 dz10 = _mm_sub_ps(iz1,jz0);
759 dx20 = _mm_sub_ps(ix2,jx0);
760 dy20 = _mm_sub_ps(iy2,jy0);
761 dz20 = _mm_sub_ps(iz2,jz0);
763 /* Calculate squared distance and things based on it */
764 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
765 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
766 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
768 rinv00 = gmx_mm_invsqrt_ps(rsq00);
769 rinv10 = gmx_mm_invsqrt_ps(rsq10);
770 rinv20 = gmx_mm_invsqrt_ps(rsq20);
772 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
773 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
774 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
776 /* Load parameters for j particles */
777 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
778 charge+jnrC+0,charge+jnrD+0);
780 /**************************
781 * CALCULATE INTERACTIONS *
782 **************************/
784 /* Compute parameters for interactions between i and j atoms */
785 qq00 = _mm_mul_ps(iq0,jq0);
787 /* COULOMB ELECTROSTATICS */
788 velec = _mm_mul_ps(qq00,rinv00);
789 felec = _mm_mul_ps(velec,rinvsq00);
793 fscal = _mm_andnot_ps(dummy_mask,fscal);
795 /* Calculate temporary vectorial force */
796 tx = _mm_mul_ps(fscal,dx00);
797 ty = _mm_mul_ps(fscal,dy00);
798 tz = _mm_mul_ps(fscal,dz00);
800 /* Update vectorial force */
801 fix0 = _mm_add_ps(fix0,tx);
802 fiy0 = _mm_add_ps(fiy0,ty);
803 fiz0 = _mm_add_ps(fiz0,tz);
805 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
806 f+j_coord_offsetC,f+j_coord_offsetD,
809 /**************************
810 * CALCULATE INTERACTIONS *
811 **************************/
813 /* Compute parameters for interactions between i and j atoms */
814 qq10 = _mm_mul_ps(iq1,jq0);
816 /* COULOMB ELECTROSTATICS */
817 velec = _mm_mul_ps(qq10,rinv10);
818 felec = _mm_mul_ps(velec,rinvsq10);
822 fscal = _mm_andnot_ps(dummy_mask,fscal);
824 /* Calculate temporary vectorial force */
825 tx = _mm_mul_ps(fscal,dx10);
826 ty = _mm_mul_ps(fscal,dy10);
827 tz = _mm_mul_ps(fscal,dz10);
829 /* Update vectorial force */
830 fix1 = _mm_add_ps(fix1,tx);
831 fiy1 = _mm_add_ps(fiy1,ty);
832 fiz1 = _mm_add_ps(fiz1,tz);
834 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
835 f+j_coord_offsetC,f+j_coord_offsetD,
838 /**************************
839 * CALCULATE INTERACTIONS *
840 **************************/
842 /* Compute parameters for interactions between i and j atoms */
843 qq20 = _mm_mul_ps(iq2,jq0);
845 /* COULOMB ELECTROSTATICS */
846 velec = _mm_mul_ps(qq20,rinv20);
847 felec = _mm_mul_ps(velec,rinvsq20);
851 fscal = _mm_andnot_ps(dummy_mask,fscal);
853 /* Calculate temporary vectorial force */
854 tx = _mm_mul_ps(fscal,dx20);
855 ty = _mm_mul_ps(fscal,dy20);
856 tz = _mm_mul_ps(fscal,dz20);
858 /* Update vectorial force */
859 fix2 = _mm_add_ps(fix2,tx);
860 fiy2 = _mm_add_ps(fiy2,ty);
861 fiz2 = _mm_add_ps(fiz2,tz);
863 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
864 f+j_coord_offsetC,f+j_coord_offsetD,
867 /* Inner loop uses 81 flops */
870 /* End of innermost loop */
872 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
873 f+i_coord_offset,fshift+i_shift_offset);
875 /* Increment number of inner iterations */
876 inneriter += j_index_end - j_index_start;
878 /* Outer loop uses 27 flops */
881 /* Increment number of outer iterations */
884 /* Update outer/inner flops */
886 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*27 + inneriter*81);