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_GeomW3W3_VF_avx_128_fma_single
52 * Electrostatics interaction: Coulomb
53 * VdW interaction: LennardJones
54 * Geometry: Water3-Water3
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCoul_VdwLJ_GeomW3W3_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 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
92 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
93 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
94 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
95 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
96 __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
97 __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
98 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
99 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
100 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
101 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
102 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
103 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
104 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
107 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
110 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
111 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
112 __m128 dummy_mask,cutoff_mask;
113 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
114 __m128 one = _mm_set1_ps(1.0);
115 __m128 two = _mm_set1_ps(2.0);
121 jindex = nlist->jindex;
123 shiftidx = nlist->shift;
125 shiftvec = fr->shift_vec[0];
126 fshift = fr->fshift[0];
127 facel = _mm_set1_ps(fr->epsfac);
128 charge = mdatoms->chargeA;
129 nvdwtype = fr->ntype;
131 vdwtype = mdatoms->typeA;
133 /* Setup water-specific parameters */
134 inr = nlist->iinr[0];
135 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
136 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
137 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
138 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
140 jq0 = _mm_set1_ps(charge[inr+0]);
141 jq1 = _mm_set1_ps(charge[inr+1]);
142 jq2 = _mm_set1_ps(charge[inr+2]);
143 vdwjidx0A = 2*vdwtype[inr+0];
144 qq00 = _mm_mul_ps(iq0,jq0);
145 c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
146 c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
147 qq01 = _mm_mul_ps(iq0,jq1);
148 qq02 = _mm_mul_ps(iq0,jq2);
149 qq10 = _mm_mul_ps(iq1,jq0);
150 qq11 = _mm_mul_ps(iq1,jq1);
151 qq12 = _mm_mul_ps(iq1,jq2);
152 qq20 = _mm_mul_ps(iq2,jq0);
153 qq21 = _mm_mul_ps(iq2,jq1);
154 qq22 = _mm_mul_ps(iq2,jq2);
156 /* Avoid stupid compiler warnings */
157 jnrA = jnrB = jnrC = jnrD = 0;
166 for(iidx=0;iidx<4*DIM;iidx++)
171 /* Start outer loop over neighborlists */
172 for(iidx=0; iidx<nri; iidx++)
174 /* Load shift vector for this list */
175 i_shift_offset = DIM*shiftidx[iidx];
177 /* Load limits for loop over neighbors */
178 j_index_start = jindex[iidx];
179 j_index_end = jindex[iidx+1];
181 /* Get outer coordinate index */
183 i_coord_offset = DIM*inr;
185 /* Load i particle coords and add shift vector */
186 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
187 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
189 fix0 = _mm_setzero_ps();
190 fiy0 = _mm_setzero_ps();
191 fiz0 = _mm_setzero_ps();
192 fix1 = _mm_setzero_ps();
193 fiy1 = _mm_setzero_ps();
194 fiz1 = _mm_setzero_ps();
195 fix2 = _mm_setzero_ps();
196 fiy2 = _mm_setzero_ps();
197 fiz2 = _mm_setzero_ps();
199 /* Reset potential sums */
200 velecsum = _mm_setzero_ps();
201 vvdwsum = _mm_setzero_ps();
203 /* Start inner kernel loop */
204 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
207 /* Get j neighbor index, and coordinate index */
212 j_coord_offsetA = DIM*jnrA;
213 j_coord_offsetB = DIM*jnrB;
214 j_coord_offsetC = DIM*jnrC;
215 j_coord_offsetD = DIM*jnrD;
217 /* load j atom coordinates */
218 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
219 x+j_coord_offsetC,x+j_coord_offsetD,
220 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
222 /* Calculate displacement vector */
223 dx00 = _mm_sub_ps(ix0,jx0);
224 dy00 = _mm_sub_ps(iy0,jy0);
225 dz00 = _mm_sub_ps(iz0,jz0);
226 dx01 = _mm_sub_ps(ix0,jx1);
227 dy01 = _mm_sub_ps(iy0,jy1);
228 dz01 = _mm_sub_ps(iz0,jz1);
229 dx02 = _mm_sub_ps(ix0,jx2);
230 dy02 = _mm_sub_ps(iy0,jy2);
231 dz02 = _mm_sub_ps(iz0,jz2);
232 dx10 = _mm_sub_ps(ix1,jx0);
233 dy10 = _mm_sub_ps(iy1,jy0);
234 dz10 = _mm_sub_ps(iz1,jz0);
235 dx11 = _mm_sub_ps(ix1,jx1);
236 dy11 = _mm_sub_ps(iy1,jy1);
237 dz11 = _mm_sub_ps(iz1,jz1);
238 dx12 = _mm_sub_ps(ix1,jx2);
239 dy12 = _mm_sub_ps(iy1,jy2);
240 dz12 = _mm_sub_ps(iz1,jz2);
241 dx20 = _mm_sub_ps(ix2,jx0);
242 dy20 = _mm_sub_ps(iy2,jy0);
243 dz20 = _mm_sub_ps(iz2,jz0);
244 dx21 = _mm_sub_ps(ix2,jx1);
245 dy21 = _mm_sub_ps(iy2,jy1);
246 dz21 = _mm_sub_ps(iz2,jz1);
247 dx22 = _mm_sub_ps(ix2,jx2);
248 dy22 = _mm_sub_ps(iy2,jy2);
249 dz22 = _mm_sub_ps(iz2,jz2);
251 /* Calculate squared distance and things based on it */
252 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
253 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
254 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
255 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
256 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
257 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
258 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
259 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
260 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
262 rinv00 = gmx_mm_invsqrt_ps(rsq00);
263 rinv01 = gmx_mm_invsqrt_ps(rsq01);
264 rinv02 = gmx_mm_invsqrt_ps(rsq02);
265 rinv10 = gmx_mm_invsqrt_ps(rsq10);
266 rinv11 = gmx_mm_invsqrt_ps(rsq11);
267 rinv12 = gmx_mm_invsqrt_ps(rsq12);
268 rinv20 = gmx_mm_invsqrt_ps(rsq20);
269 rinv21 = gmx_mm_invsqrt_ps(rsq21);
270 rinv22 = gmx_mm_invsqrt_ps(rsq22);
272 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
273 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
274 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
275 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
276 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
277 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
278 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
279 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
280 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
282 fjx0 = _mm_setzero_ps();
283 fjy0 = _mm_setzero_ps();
284 fjz0 = _mm_setzero_ps();
285 fjx1 = _mm_setzero_ps();
286 fjy1 = _mm_setzero_ps();
287 fjz1 = _mm_setzero_ps();
288 fjx2 = _mm_setzero_ps();
289 fjy2 = _mm_setzero_ps();
290 fjz2 = _mm_setzero_ps();
292 /**************************
293 * CALCULATE INTERACTIONS *
294 **************************/
296 /* COULOMB ELECTROSTATICS */
297 velec = _mm_mul_ps(qq00,rinv00);
298 felec = _mm_mul_ps(velec,rinvsq00);
300 /* LENNARD-JONES DISPERSION/REPULSION */
302 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
303 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
304 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
305 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
306 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
308 /* Update potential sum for this i atom from the interaction with this j atom. */
309 velecsum = _mm_add_ps(velecsum,velec);
310 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
312 fscal = _mm_add_ps(felec,fvdw);
314 /* Update vectorial force */
315 fix0 = _mm_macc_ps(dx00,fscal,fix0);
316 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
317 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
319 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
320 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
321 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
323 /**************************
324 * CALCULATE INTERACTIONS *
325 **************************/
327 /* COULOMB ELECTROSTATICS */
328 velec = _mm_mul_ps(qq01,rinv01);
329 felec = _mm_mul_ps(velec,rinvsq01);
331 /* Update potential sum for this i atom from the interaction with this j atom. */
332 velecsum = _mm_add_ps(velecsum,velec);
336 /* Update vectorial force */
337 fix0 = _mm_macc_ps(dx01,fscal,fix0);
338 fiy0 = _mm_macc_ps(dy01,fscal,fiy0);
339 fiz0 = _mm_macc_ps(dz01,fscal,fiz0);
341 fjx1 = _mm_macc_ps(dx01,fscal,fjx1);
342 fjy1 = _mm_macc_ps(dy01,fscal,fjy1);
343 fjz1 = _mm_macc_ps(dz01,fscal,fjz1);
345 /**************************
346 * CALCULATE INTERACTIONS *
347 **************************/
349 /* COULOMB ELECTROSTATICS */
350 velec = _mm_mul_ps(qq02,rinv02);
351 felec = _mm_mul_ps(velec,rinvsq02);
353 /* Update potential sum for this i atom from the interaction with this j atom. */
354 velecsum = _mm_add_ps(velecsum,velec);
358 /* Update vectorial force */
359 fix0 = _mm_macc_ps(dx02,fscal,fix0);
360 fiy0 = _mm_macc_ps(dy02,fscal,fiy0);
361 fiz0 = _mm_macc_ps(dz02,fscal,fiz0);
363 fjx2 = _mm_macc_ps(dx02,fscal,fjx2);
364 fjy2 = _mm_macc_ps(dy02,fscal,fjy2);
365 fjz2 = _mm_macc_ps(dz02,fscal,fjz2);
367 /**************************
368 * CALCULATE INTERACTIONS *
369 **************************/
371 /* COULOMB ELECTROSTATICS */
372 velec = _mm_mul_ps(qq10,rinv10);
373 felec = _mm_mul_ps(velec,rinvsq10);
375 /* Update potential sum for this i atom from the interaction with this j atom. */
376 velecsum = _mm_add_ps(velecsum,velec);
380 /* Update vectorial force */
381 fix1 = _mm_macc_ps(dx10,fscal,fix1);
382 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
383 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
385 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
386 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
387 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
389 /**************************
390 * CALCULATE INTERACTIONS *
391 **************************/
393 /* COULOMB ELECTROSTATICS */
394 velec = _mm_mul_ps(qq11,rinv11);
395 felec = _mm_mul_ps(velec,rinvsq11);
397 /* Update potential sum for this i atom from the interaction with this j atom. */
398 velecsum = _mm_add_ps(velecsum,velec);
402 /* Update vectorial force */
403 fix1 = _mm_macc_ps(dx11,fscal,fix1);
404 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
405 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
407 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
408 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
409 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
411 /**************************
412 * CALCULATE INTERACTIONS *
413 **************************/
415 /* COULOMB ELECTROSTATICS */
416 velec = _mm_mul_ps(qq12,rinv12);
417 felec = _mm_mul_ps(velec,rinvsq12);
419 /* Update potential sum for this i atom from the interaction with this j atom. */
420 velecsum = _mm_add_ps(velecsum,velec);
424 /* Update vectorial force */
425 fix1 = _mm_macc_ps(dx12,fscal,fix1);
426 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
427 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
429 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
430 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
431 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
433 /**************************
434 * CALCULATE INTERACTIONS *
435 **************************/
437 /* COULOMB ELECTROSTATICS */
438 velec = _mm_mul_ps(qq20,rinv20);
439 felec = _mm_mul_ps(velec,rinvsq20);
441 /* Update potential sum for this i atom from the interaction with this j atom. */
442 velecsum = _mm_add_ps(velecsum,velec);
446 /* Update vectorial force */
447 fix2 = _mm_macc_ps(dx20,fscal,fix2);
448 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
449 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
451 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
452 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
453 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
455 /**************************
456 * CALCULATE INTERACTIONS *
457 **************************/
459 /* COULOMB ELECTROSTATICS */
460 velec = _mm_mul_ps(qq21,rinv21);
461 felec = _mm_mul_ps(velec,rinvsq21);
463 /* Update potential sum for this i atom from the interaction with this j atom. */
464 velecsum = _mm_add_ps(velecsum,velec);
468 /* Update vectorial force */
469 fix2 = _mm_macc_ps(dx21,fscal,fix2);
470 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
471 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
473 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
474 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
475 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
477 /**************************
478 * CALCULATE INTERACTIONS *
479 **************************/
481 /* COULOMB ELECTROSTATICS */
482 velec = _mm_mul_ps(qq22,rinv22);
483 felec = _mm_mul_ps(velec,rinvsq22);
485 /* Update potential sum for this i atom from the interaction with this j atom. */
486 velecsum = _mm_add_ps(velecsum,velec);
490 /* Update vectorial force */
491 fix2 = _mm_macc_ps(dx22,fscal,fix2);
492 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
493 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
495 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
496 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
497 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
499 fjptrA = f+j_coord_offsetA;
500 fjptrB = f+j_coord_offsetB;
501 fjptrC = f+j_coord_offsetC;
502 fjptrD = f+j_coord_offsetD;
504 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
505 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
507 /* Inner loop uses 291 flops */
513 /* Get j neighbor index, and coordinate index */
514 jnrlistA = jjnr[jidx];
515 jnrlistB = jjnr[jidx+1];
516 jnrlistC = jjnr[jidx+2];
517 jnrlistD = jjnr[jidx+3];
518 /* Sign of each element will be negative for non-real atoms.
519 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
520 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
522 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
523 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
524 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
525 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
526 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
527 j_coord_offsetA = DIM*jnrA;
528 j_coord_offsetB = DIM*jnrB;
529 j_coord_offsetC = DIM*jnrC;
530 j_coord_offsetD = DIM*jnrD;
532 /* load j atom coordinates */
533 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
534 x+j_coord_offsetC,x+j_coord_offsetD,
535 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
537 /* Calculate displacement vector */
538 dx00 = _mm_sub_ps(ix0,jx0);
539 dy00 = _mm_sub_ps(iy0,jy0);
540 dz00 = _mm_sub_ps(iz0,jz0);
541 dx01 = _mm_sub_ps(ix0,jx1);
542 dy01 = _mm_sub_ps(iy0,jy1);
543 dz01 = _mm_sub_ps(iz0,jz1);
544 dx02 = _mm_sub_ps(ix0,jx2);
545 dy02 = _mm_sub_ps(iy0,jy2);
546 dz02 = _mm_sub_ps(iz0,jz2);
547 dx10 = _mm_sub_ps(ix1,jx0);
548 dy10 = _mm_sub_ps(iy1,jy0);
549 dz10 = _mm_sub_ps(iz1,jz0);
550 dx11 = _mm_sub_ps(ix1,jx1);
551 dy11 = _mm_sub_ps(iy1,jy1);
552 dz11 = _mm_sub_ps(iz1,jz1);
553 dx12 = _mm_sub_ps(ix1,jx2);
554 dy12 = _mm_sub_ps(iy1,jy2);
555 dz12 = _mm_sub_ps(iz1,jz2);
556 dx20 = _mm_sub_ps(ix2,jx0);
557 dy20 = _mm_sub_ps(iy2,jy0);
558 dz20 = _mm_sub_ps(iz2,jz0);
559 dx21 = _mm_sub_ps(ix2,jx1);
560 dy21 = _mm_sub_ps(iy2,jy1);
561 dz21 = _mm_sub_ps(iz2,jz1);
562 dx22 = _mm_sub_ps(ix2,jx2);
563 dy22 = _mm_sub_ps(iy2,jy2);
564 dz22 = _mm_sub_ps(iz2,jz2);
566 /* Calculate squared distance and things based on it */
567 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
568 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
569 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
570 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
571 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
572 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
573 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
574 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
575 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
577 rinv00 = gmx_mm_invsqrt_ps(rsq00);
578 rinv01 = gmx_mm_invsqrt_ps(rsq01);
579 rinv02 = gmx_mm_invsqrt_ps(rsq02);
580 rinv10 = gmx_mm_invsqrt_ps(rsq10);
581 rinv11 = gmx_mm_invsqrt_ps(rsq11);
582 rinv12 = gmx_mm_invsqrt_ps(rsq12);
583 rinv20 = gmx_mm_invsqrt_ps(rsq20);
584 rinv21 = gmx_mm_invsqrt_ps(rsq21);
585 rinv22 = gmx_mm_invsqrt_ps(rsq22);
587 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
588 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
589 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
590 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
591 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
592 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
593 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
594 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
595 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
597 fjx0 = _mm_setzero_ps();
598 fjy0 = _mm_setzero_ps();
599 fjz0 = _mm_setzero_ps();
600 fjx1 = _mm_setzero_ps();
601 fjy1 = _mm_setzero_ps();
602 fjz1 = _mm_setzero_ps();
603 fjx2 = _mm_setzero_ps();
604 fjy2 = _mm_setzero_ps();
605 fjz2 = _mm_setzero_ps();
607 /**************************
608 * CALCULATE INTERACTIONS *
609 **************************/
611 /* COULOMB ELECTROSTATICS */
612 velec = _mm_mul_ps(qq00,rinv00);
613 felec = _mm_mul_ps(velec,rinvsq00);
615 /* LENNARD-JONES DISPERSION/REPULSION */
617 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
618 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
619 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
620 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
621 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
623 /* Update potential sum for this i atom from the interaction with this j atom. */
624 velec = _mm_andnot_ps(dummy_mask,velec);
625 velecsum = _mm_add_ps(velecsum,velec);
626 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
627 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
629 fscal = _mm_add_ps(felec,fvdw);
631 fscal = _mm_andnot_ps(dummy_mask,fscal);
633 /* Update vectorial force */
634 fix0 = _mm_macc_ps(dx00,fscal,fix0);
635 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
636 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
638 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
639 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
640 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
642 /**************************
643 * CALCULATE INTERACTIONS *
644 **************************/
646 /* COULOMB ELECTROSTATICS */
647 velec = _mm_mul_ps(qq01,rinv01);
648 felec = _mm_mul_ps(velec,rinvsq01);
650 /* Update potential sum for this i atom from the interaction with this j atom. */
651 velec = _mm_andnot_ps(dummy_mask,velec);
652 velecsum = _mm_add_ps(velecsum,velec);
656 fscal = _mm_andnot_ps(dummy_mask,fscal);
658 /* Update vectorial force */
659 fix0 = _mm_macc_ps(dx01,fscal,fix0);
660 fiy0 = _mm_macc_ps(dy01,fscal,fiy0);
661 fiz0 = _mm_macc_ps(dz01,fscal,fiz0);
663 fjx1 = _mm_macc_ps(dx01,fscal,fjx1);
664 fjy1 = _mm_macc_ps(dy01,fscal,fjy1);
665 fjz1 = _mm_macc_ps(dz01,fscal,fjz1);
667 /**************************
668 * CALCULATE INTERACTIONS *
669 **************************/
671 /* COULOMB ELECTROSTATICS */
672 velec = _mm_mul_ps(qq02,rinv02);
673 felec = _mm_mul_ps(velec,rinvsq02);
675 /* Update potential sum for this i atom from the interaction with this j atom. */
676 velec = _mm_andnot_ps(dummy_mask,velec);
677 velecsum = _mm_add_ps(velecsum,velec);
681 fscal = _mm_andnot_ps(dummy_mask,fscal);
683 /* Update vectorial force */
684 fix0 = _mm_macc_ps(dx02,fscal,fix0);
685 fiy0 = _mm_macc_ps(dy02,fscal,fiy0);
686 fiz0 = _mm_macc_ps(dz02,fscal,fiz0);
688 fjx2 = _mm_macc_ps(dx02,fscal,fjx2);
689 fjy2 = _mm_macc_ps(dy02,fscal,fjy2);
690 fjz2 = _mm_macc_ps(dz02,fscal,fjz2);
692 /**************************
693 * CALCULATE INTERACTIONS *
694 **************************/
696 /* COULOMB ELECTROSTATICS */
697 velec = _mm_mul_ps(qq10,rinv10);
698 felec = _mm_mul_ps(velec,rinvsq10);
700 /* Update potential sum for this i atom from the interaction with this j atom. */
701 velec = _mm_andnot_ps(dummy_mask,velec);
702 velecsum = _mm_add_ps(velecsum,velec);
706 fscal = _mm_andnot_ps(dummy_mask,fscal);
708 /* Update vectorial force */
709 fix1 = _mm_macc_ps(dx10,fscal,fix1);
710 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
711 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
713 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
714 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
715 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
717 /**************************
718 * CALCULATE INTERACTIONS *
719 **************************/
721 /* COULOMB ELECTROSTATICS */
722 velec = _mm_mul_ps(qq11,rinv11);
723 felec = _mm_mul_ps(velec,rinvsq11);
725 /* Update potential sum for this i atom from the interaction with this j atom. */
726 velec = _mm_andnot_ps(dummy_mask,velec);
727 velecsum = _mm_add_ps(velecsum,velec);
731 fscal = _mm_andnot_ps(dummy_mask,fscal);
733 /* Update vectorial force */
734 fix1 = _mm_macc_ps(dx11,fscal,fix1);
735 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
736 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
738 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
739 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
740 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
742 /**************************
743 * CALCULATE INTERACTIONS *
744 **************************/
746 /* COULOMB ELECTROSTATICS */
747 velec = _mm_mul_ps(qq12,rinv12);
748 felec = _mm_mul_ps(velec,rinvsq12);
750 /* Update potential sum for this i atom from the interaction with this j atom. */
751 velec = _mm_andnot_ps(dummy_mask,velec);
752 velecsum = _mm_add_ps(velecsum,velec);
756 fscal = _mm_andnot_ps(dummy_mask,fscal);
758 /* Update vectorial force */
759 fix1 = _mm_macc_ps(dx12,fscal,fix1);
760 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
761 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
763 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
764 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
765 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
767 /**************************
768 * CALCULATE INTERACTIONS *
769 **************************/
771 /* COULOMB ELECTROSTATICS */
772 velec = _mm_mul_ps(qq20,rinv20);
773 felec = _mm_mul_ps(velec,rinvsq20);
775 /* Update potential sum for this i atom from the interaction with this j atom. */
776 velec = _mm_andnot_ps(dummy_mask,velec);
777 velecsum = _mm_add_ps(velecsum,velec);
781 fscal = _mm_andnot_ps(dummy_mask,fscal);
783 /* Update vectorial force */
784 fix2 = _mm_macc_ps(dx20,fscal,fix2);
785 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
786 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
788 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
789 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
790 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
792 /**************************
793 * CALCULATE INTERACTIONS *
794 **************************/
796 /* COULOMB ELECTROSTATICS */
797 velec = _mm_mul_ps(qq21,rinv21);
798 felec = _mm_mul_ps(velec,rinvsq21);
800 /* Update potential sum for this i atom from the interaction with this j atom. */
801 velec = _mm_andnot_ps(dummy_mask,velec);
802 velecsum = _mm_add_ps(velecsum,velec);
806 fscal = _mm_andnot_ps(dummy_mask,fscal);
808 /* Update vectorial force */
809 fix2 = _mm_macc_ps(dx21,fscal,fix2);
810 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
811 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
813 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
814 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
815 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
817 /**************************
818 * CALCULATE INTERACTIONS *
819 **************************/
821 /* COULOMB ELECTROSTATICS */
822 velec = _mm_mul_ps(qq22,rinv22);
823 felec = _mm_mul_ps(velec,rinvsq22);
825 /* Update potential sum for this i atom from the interaction with this j atom. */
826 velec = _mm_andnot_ps(dummy_mask,velec);
827 velecsum = _mm_add_ps(velecsum,velec);
831 fscal = _mm_andnot_ps(dummy_mask,fscal);
833 /* Update vectorial force */
834 fix2 = _mm_macc_ps(dx22,fscal,fix2);
835 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
836 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
838 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
839 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
840 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
842 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
843 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
844 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
845 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
847 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
848 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
850 /* Inner loop uses 291 flops */
853 /* End of innermost loop */
855 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
856 f+i_coord_offset,fshift+i_shift_offset);
859 /* Update potential energies */
860 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
861 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
863 /* Increment number of inner iterations */
864 inneriter += j_index_end - j_index_start;
866 /* Outer loop uses 20 flops */
869 /* Increment number of outer iterations */
872 /* Update outer/inner flops */
874 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*291);
877 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwLJ_GeomW3W3_F_avx_128_fma_single
878 * Electrostatics interaction: Coulomb
879 * VdW interaction: LennardJones
880 * Geometry: Water3-Water3
881 * Calculate force/pot: Force
884 nb_kernel_ElecCoul_VdwLJ_GeomW3W3_F_avx_128_fma_single
885 (t_nblist * gmx_restrict nlist,
886 rvec * gmx_restrict xx,
887 rvec * gmx_restrict ff,
888 t_forcerec * gmx_restrict fr,
889 t_mdatoms * gmx_restrict mdatoms,
890 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
891 t_nrnb * gmx_restrict nrnb)
893 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
894 * just 0 for non-waters.
895 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
896 * jnr indices corresponding to data put in the four positions in the SIMD register.
898 int i_shift_offset,i_coord_offset,outeriter,inneriter;
899 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
900 int jnrA,jnrB,jnrC,jnrD;
901 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
902 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
903 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
905 real *shiftvec,*fshift,*x,*f;
906 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
908 __m128 fscal,rcutoff,rcutoff2,jidxall;
910 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
912 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
914 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
915 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
916 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
917 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
918 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
919 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
920 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
921 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
922 __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
923 __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
924 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
925 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
926 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
927 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
928 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
929 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
930 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
933 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
936 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
937 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
938 __m128 dummy_mask,cutoff_mask;
939 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
940 __m128 one = _mm_set1_ps(1.0);
941 __m128 two = _mm_set1_ps(2.0);
947 jindex = nlist->jindex;
949 shiftidx = nlist->shift;
951 shiftvec = fr->shift_vec[0];
952 fshift = fr->fshift[0];
953 facel = _mm_set1_ps(fr->epsfac);
954 charge = mdatoms->chargeA;
955 nvdwtype = fr->ntype;
957 vdwtype = mdatoms->typeA;
959 /* Setup water-specific parameters */
960 inr = nlist->iinr[0];
961 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
962 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
963 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
964 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
966 jq0 = _mm_set1_ps(charge[inr+0]);
967 jq1 = _mm_set1_ps(charge[inr+1]);
968 jq2 = _mm_set1_ps(charge[inr+2]);
969 vdwjidx0A = 2*vdwtype[inr+0];
970 qq00 = _mm_mul_ps(iq0,jq0);
971 c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
972 c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
973 qq01 = _mm_mul_ps(iq0,jq1);
974 qq02 = _mm_mul_ps(iq0,jq2);
975 qq10 = _mm_mul_ps(iq1,jq0);
976 qq11 = _mm_mul_ps(iq1,jq1);
977 qq12 = _mm_mul_ps(iq1,jq2);
978 qq20 = _mm_mul_ps(iq2,jq0);
979 qq21 = _mm_mul_ps(iq2,jq1);
980 qq22 = _mm_mul_ps(iq2,jq2);
982 /* Avoid stupid compiler warnings */
983 jnrA = jnrB = jnrC = jnrD = 0;
992 for(iidx=0;iidx<4*DIM;iidx++)
997 /* Start outer loop over neighborlists */
998 for(iidx=0; iidx<nri; iidx++)
1000 /* Load shift vector for this list */
1001 i_shift_offset = DIM*shiftidx[iidx];
1003 /* Load limits for loop over neighbors */
1004 j_index_start = jindex[iidx];
1005 j_index_end = jindex[iidx+1];
1007 /* Get outer coordinate index */
1009 i_coord_offset = DIM*inr;
1011 /* Load i particle coords and add shift vector */
1012 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1013 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1015 fix0 = _mm_setzero_ps();
1016 fiy0 = _mm_setzero_ps();
1017 fiz0 = _mm_setzero_ps();
1018 fix1 = _mm_setzero_ps();
1019 fiy1 = _mm_setzero_ps();
1020 fiz1 = _mm_setzero_ps();
1021 fix2 = _mm_setzero_ps();
1022 fiy2 = _mm_setzero_ps();
1023 fiz2 = _mm_setzero_ps();
1025 /* Start inner kernel loop */
1026 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1029 /* Get j neighbor index, and coordinate index */
1031 jnrB = jjnr[jidx+1];
1032 jnrC = jjnr[jidx+2];
1033 jnrD = jjnr[jidx+3];
1034 j_coord_offsetA = DIM*jnrA;
1035 j_coord_offsetB = DIM*jnrB;
1036 j_coord_offsetC = DIM*jnrC;
1037 j_coord_offsetD = DIM*jnrD;
1039 /* load j atom coordinates */
1040 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1041 x+j_coord_offsetC,x+j_coord_offsetD,
1042 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1044 /* Calculate displacement vector */
1045 dx00 = _mm_sub_ps(ix0,jx0);
1046 dy00 = _mm_sub_ps(iy0,jy0);
1047 dz00 = _mm_sub_ps(iz0,jz0);
1048 dx01 = _mm_sub_ps(ix0,jx1);
1049 dy01 = _mm_sub_ps(iy0,jy1);
1050 dz01 = _mm_sub_ps(iz0,jz1);
1051 dx02 = _mm_sub_ps(ix0,jx2);
1052 dy02 = _mm_sub_ps(iy0,jy2);
1053 dz02 = _mm_sub_ps(iz0,jz2);
1054 dx10 = _mm_sub_ps(ix1,jx0);
1055 dy10 = _mm_sub_ps(iy1,jy0);
1056 dz10 = _mm_sub_ps(iz1,jz0);
1057 dx11 = _mm_sub_ps(ix1,jx1);
1058 dy11 = _mm_sub_ps(iy1,jy1);
1059 dz11 = _mm_sub_ps(iz1,jz1);
1060 dx12 = _mm_sub_ps(ix1,jx2);
1061 dy12 = _mm_sub_ps(iy1,jy2);
1062 dz12 = _mm_sub_ps(iz1,jz2);
1063 dx20 = _mm_sub_ps(ix2,jx0);
1064 dy20 = _mm_sub_ps(iy2,jy0);
1065 dz20 = _mm_sub_ps(iz2,jz0);
1066 dx21 = _mm_sub_ps(ix2,jx1);
1067 dy21 = _mm_sub_ps(iy2,jy1);
1068 dz21 = _mm_sub_ps(iz2,jz1);
1069 dx22 = _mm_sub_ps(ix2,jx2);
1070 dy22 = _mm_sub_ps(iy2,jy2);
1071 dz22 = _mm_sub_ps(iz2,jz2);
1073 /* Calculate squared distance and things based on it */
1074 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1075 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1076 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1077 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1078 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1079 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1080 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1081 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1082 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1084 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1085 rinv01 = gmx_mm_invsqrt_ps(rsq01);
1086 rinv02 = gmx_mm_invsqrt_ps(rsq02);
1087 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1088 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1089 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1090 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1091 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1092 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1094 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1095 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
1096 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
1097 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1098 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
1099 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
1100 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1101 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
1102 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
1104 fjx0 = _mm_setzero_ps();
1105 fjy0 = _mm_setzero_ps();
1106 fjz0 = _mm_setzero_ps();
1107 fjx1 = _mm_setzero_ps();
1108 fjy1 = _mm_setzero_ps();
1109 fjz1 = _mm_setzero_ps();
1110 fjx2 = _mm_setzero_ps();
1111 fjy2 = _mm_setzero_ps();
1112 fjz2 = _mm_setzero_ps();
1114 /**************************
1115 * CALCULATE INTERACTIONS *
1116 **************************/
1118 /* COULOMB ELECTROSTATICS */
1119 velec = _mm_mul_ps(qq00,rinv00);
1120 felec = _mm_mul_ps(velec,rinvsq00);
1122 /* LENNARD-JONES DISPERSION/REPULSION */
1124 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1125 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1127 fscal = _mm_add_ps(felec,fvdw);
1129 /* Update vectorial force */
1130 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1131 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1132 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1134 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1135 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1136 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1138 /**************************
1139 * CALCULATE INTERACTIONS *
1140 **************************/
1142 /* COULOMB ELECTROSTATICS */
1143 velec = _mm_mul_ps(qq01,rinv01);
1144 felec = _mm_mul_ps(velec,rinvsq01);
1148 /* Update vectorial force */
1149 fix0 = _mm_macc_ps(dx01,fscal,fix0);
1150 fiy0 = _mm_macc_ps(dy01,fscal,fiy0);
1151 fiz0 = _mm_macc_ps(dz01,fscal,fiz0);
1153 fjx1 = _mm_macc_ps(dx01,fscal,fjx1);
1154 fjy1 = _mm_macc_ps(dy01,fscal,fjy1);
1155 fjz1 = _mm_macc_ps(dz01,fscal,fjz1);
1157 /**************************
1158 * CALCULATE INTERACTIONS *
1159 **************************/
1161 /* COULOMB ELECTROSTATICS */
1162 velec = _mm_mul_ps(qq02,rinv02);
1163 felec = _mm_mul_ps(velec,rinvsq02);
1167 /* Update vectorial force */
1168 fix0 = _mm_macc_ps(dx02,fscal,fix0);
1169 fiy0 = _mm_macc_ps(dy02,fscal,fiy0);
1170 fiz0 = _mm_macc_ps(dz02,fscal,fiz0);
1172 fjx2 = _mm_macc_ps(dx02,fscal,fjx2);
1173 fjy2 = _mm_macc_ps(dy02,fscal,fjy2);
1174 fjz2 = _mm_macc_ps(dz02,fscal,fjz2);
1176 /**************************
1177 * CALCULATE INTERACTIONS *
1178 **************************/
1180 /* COULOMB ELECTROSTATICS */
1181 velec = _mm_mul_ps(qq10,rinv10);
1182 felec = _mm_mul_ps(velec,rinvsq10);
1186 /* Update vectorial force */
1187 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1188 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1189 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1191 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1192 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1193 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1195 /**************************
1196 * CALCULATE INTERACTIONS *
1197 **************************/
1199 /* COULOMB ELECTROSTATICS */
1200 velec = _mm_mul_ps(qq11,rinv11);
1201 felec = _mm_mul_ps(velec,rinvsq11);
1205 /* Update vectorial force */
1206 fix1 = _mm_macc_ps(dx11,fscal,fix1);
1207 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
1208 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
1210 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
1211 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
1212 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
1214 /**************************
1215 * CALCULATE INTERACTIONS *
1216 **************************/
1218 /* COULOMB ELECTROSTATICS */
1219 velec = _mm_mul_ps(qq12,rinv12);
1220 felec = _mm_mul_ps(velec,rinvsq12);
1224 /* Update vectorial force */
1225 fix1 = _mm_macc_ps(dx12,fscal,fix1);
1226 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
1227 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
1229 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
1230 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
1231 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
1233 /**************************
1234 * CALCULATE INTERACTIONS *
1235 **************************/
1237 /* COULOMB ELECTROSTATICS */
1238 velec = _mm_mul_ps(qq20,rinv20);
1239 felec = _mm_mul_ps(velec,rinvsq20);
1243 /* Update vectorial force */
1244 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1245 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1246 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1248 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1249 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1250 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1252 /**************************
1253 * CALCULATE INTERACTIONS *
1254 **************************/
1256 /* COULOMB ELECTROSTATICS */
1257 velec = _mm_mul_ps(qq21,rinv21);
1258 felec = _mm_mul_ps(velec,rinvsq21);
1262 /* Update vectorial force */
1263 fix2 = _mm_macc_ps(dx21,fscal,fix2);
1264 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
1265 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
1267 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
1268 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
1269 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
1271 /**************************
1272 * CALCULATE INTERACTIONS *
1273 **************************/
1275 /* COULOMB ELECTROSTATICS */
1276 velec = _mm_mul_ps(qq22,rinv22);
1277 felec = _mm_mul_ps(velec,rinvsq22);
1281 /* Update vectorial force */
1282 fix2 = _mm_macc_ps(dx22,fscal,fix2);
1283 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
1284 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
1286 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
1287 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
1288 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
1290 fjptrA = f+j_coord_offsetA;
1291 fjptrB = f+j_coord_offsetB;
1292 fjptrC = f+j_coord_offsetC;
1293 fjptrD = f+j_coord_offsetD;
1295 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1296 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1298 /* Inner loop uses 277 flops */
1301 if(jidx<j_index_end)
1304 /* Get j neighbor index, and coordinate index */
1305 jnrlistA = jjnr[jidx];
1306 jnrlistB = jjnr[jidx+1];
1307 jnrlistC = jjnr[jidx+2];
1308 jnrlistD = jjnr[jidx+3];
1309 /* Sign of each element will be negative for non-real atoms.
1310 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1311 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1313 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1314 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1315 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1316 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1317 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1318 j_coord_offsetA = DIM*jnrA;
1319 j_coord_offsetB = DIM*jnrB;
1320 j_coord_offsetC = DIM*jnrC;
1321 j_coord_offsetD = DIM*jnrD;
1323 /* load j atom coordinates */
1324 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1325 x+j_coord_offsetC,x+j_coord_offsetD,
1326 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1328 /* Calculate displacement vector */
1329 dx00 = _mm_sub_ps(ix0,jx0);
1330 dy00 = _mm_sub_ps(iy0,jy0);
1331 dz00 = _mm_sub_ps(iz0,jz0);
1332 dx01 = _mm_sub_ps(ix0,jx1);
1333 dy01 = _mm_sub_ps(iy0,jy1);
1334 dz01 = _mm_sub_ps(iz0,jz1);
1335 dx02 = _mm_sub_ps(ix0,jx2);
1336 dy02 = _mm_sub_ps(iy0,jy2);
1337 dz02 = _mm_sub_ps(iz0,jz2);
1338 dx10 = _mm_sub_ps(ix1,jx0);
1339 dy10 = _mm_sub_ps(iy1,jy0);
1340 dz10 = _mm_sub_ps(iz1,jz0);
1341 dx11 = _mm_sub_ps(ix1,jx1);
1342 dy11 = _mm_sub_ps(iy1,jy1);
1343 dz11 = _mm_sub_ps(iz1,jz1);
1344 dx12 = _mm_sub_ps(ix1,jx2);
1345 dy12 = _mm_sub_ps(iy1,jy2);
1346 dz12 = _mm_sub_ps(iz1,jz2);
1347 dx20 = _mm_sub_ps(ix2,jx0);
1348 dy20 = _mm_sub_ps(iy2,jy0);
1349 dz20 = _mm_sub_ps(iz2,jz0);
1350 dx21 = _mm_sub_ps(ix2,jx1);
1351 dy21 = _mm_sub_ps(iy2,jy1);
1352 dz21 = _mm_sub_ps(iz2,jz1);
1353 dx22 = _mm_sub_ps(ix2,jx2);
1354 dy22 = _mm_sub_ps(iy2,jy2);
1355 dz22 = _mm_sub_ps(iz2,jz2);
1357 /* Calculate squared distance and things based on it */
1358 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1359 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1360 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1361 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1362 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1363 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1364 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1365 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1366 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1368 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1369 rinv01 = gmx_mm_invsqrt_ps(rsq01);
1370 rinv02 = gmx_mm_invsqrt_ps(rsq02);
1371 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1372 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1373 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1374 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1375 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1376 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1378 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1379 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
1380 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
1381 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1382 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
1383 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
1384 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1385 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
1386 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
1388 fjx0 = _mm_setzero_ps();
1389 fjy0 = _mm_setzero_ps();
1390 fjz0 = _mm_setzero_ps();
1391 fjx1 = _mm_setzero_ps();
1392 fjy1 = _mm_setzero_ps();
1393 fjz1 = _mm_setzero_ps();
1394 fjx2 = _mm_setzero_ps();
1395 fjy2 = _mm_setzero_ps();
1396 fjz2 = _mm_setzero_ps();
1398 /**************************
1399 * CALCULATE INTERACTIONS *
1400 **************************/
1402 /* COULOMB ELECTROSTATICS */
1403 velec = _mm_mul_ps(qq00,rinv00);
1404 felec = _mm_mul_ps(velec,rinvsq00);
1406 /* LENNARD-JONES DISPERSION/REPULSION */
1408 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1409 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1411 fscal = _mm_add_ps(felec,fvdw);
1413 fscal = _mm_andnot_ps(dummy_mask,fscal);
1415 /* Update vectorial force */
1416 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1417 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1418 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1420 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1421 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1422 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1424 /**************************
1425 * CALCULATE INTERACTIONS *
1426 **************************/
1428 /* COULOMB ELECTROSTATICS */
1429 velec = _mm_mul_ps(qq01,rinv01);
1430 felec = _mm_mul_ps(velec,rinvsq01);
1434 fscal = _mm_andnot_ps(dummy_mask,fscal);
1436 /* Update vectorial force */
1437 fix0 = _mm_macc_ps(dx01,fscal,fix0);
1438 fiy0 = _mm_macc_ps(dy01,fscal,fiy0);
1439 fiz0 = _mm_macc_ps(dz01,fscal,fiz0);
1441 fjx1 = _mm_macc_ps(dx01,fscal,fjx1);
1442 fjy1 = _mm_macc_ps(dy01,fscal,fjy1);
1443 fjz1 = _mm_macc_ps(dz01,fscal,fjz1);
1445 /**************************
1446 * CALCULATE INTERACTIONS *
1447 **************************/
1449 /* COULOMB ELECTROSTATICS */
1450 velec = _mm_mul_ps(qq02,rinv02);
1451 felec = _mm_mul_ps(velec,rinvsq02);
1455 fscal = _mm_andnot_ps(dummy_mask,fscal);
1457 /* Update vectorial force */
1458 fix0 = _mm_macc_ps(dx02,fscal,fix0);
1459 fiy0 = _mm_macc_ps(dy02,fscal,fiy0);
1460 fiz0 = _mm_macc_ps(dz02,fscal,fiz0);
1462 fjx2 = _mm_macc_ps(dx02,fscal,fjx2);
1463 fjy2 = _mm_macc_ps(dy02,fscal,fjy2);
1464 fjz2 = _mm_macc_ps(dz02,fscal,fjz2);
1466 /**************************
1467 * CALCULATE INTERACTIONS *
1468 **************************/
1470 /* COULOMB ELECTROSTATICS */
1471 velec = _mm_mul_ps(qq10,rinv10);
1472 felec = _mm_mul_ps(velec,rinvsq10);
1476 fscal = _mm_andnot_ps(dummy_mask,fscal);
1478 /* Update vectorial force */
1479 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1480 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1481 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1483 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1484 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1485 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1487 /**************************
1488 * CALCULATE INTERACTIONS *
1489 **************************/
1491 /* COULOMB ELECTROSTATICS */
1492 velec = _mm_mul_ps(qq11,rinv11);
1493 felec = _mm_mul_ps(velec,rinvsq11);
1497 fscal = _mm_andnot_ps(dummy_mask,fscal);
1499 /* Update vectorial force */
1500 fix1 = _mm_macc_ps(dx11,fscal,fix1);
1501 fiy1 = _mm_macc_ps(dy11,fscal,fiy1);
1502 fiz1 = _mm_macc_ps(dz11,fscal,fiz1);
1504 fjx1 = _mm_macc_ps(dx11,fscal,fjx1);
1505 fjy1 = _mm_macc_ps(dy11,fscal,fjy1);
1506 fjz1 = _mm_macc_ps(dz11,fscal,fjz1);
1508 /**************************
1509 * CALCULATE INTERACTIONS *
1510 **************************/
1512 /* COULOMB ELECTROSTATICS */
1513 velec = _mm_mul_ps(qq12,rinv12);
1514 felec = _mm_mul_ps(velec,rinvsq12);
1518 fscal = _mm_andnot_ps(dummy_mask,fscal);
1520 /* Update vectorial force */
1521 fix1 = _mm_macc_ps(dx12,fscal,fix1);
1522 fiy1 = _mm_macc_ps(dy12,fscal,fiy1);
1523 fiz1 = _mm_macc_ps(dz12,fscal,fiz1);
1525 fjx2 = _mm_macc_ps(dx12,fscal,fjx2);
1526 fjy2 = _mm_macc_ps(dy12,fscal,fjy2);
1527 fjz2 = _mm_macc_ps(dz12,fscal,fjz2);
1529 /**************************
1530 * CALCULATE INTERACTIONS *
1531 **************************/
1533 /* COULOMB ELECTROSTATICS */
1534 velec = _mm_mul_ps(qq20,rinv20);
1535 felec = _mm_mul_ps(velec,rinvsq20);
1539 fscal = _mm_andnot_ps(dummy_mask,fscal);
1541 /* Update vectorial force */
1542 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1543 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1544 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1546 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1547 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1548 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1550 /**************************
1551 * CALCULATE INTERACTIONS *
1552 **************************/
1554 /* COULOMB ELECTROSTATICS */
1555 velec = _mm_mul_ps(qq21,rinv21);
1556 felec = _mm_mul_ps(velec,rinvsq21);
1560 fscal = _mm_andnot_ps(dummy_mask,fscal);
1562 /* Update vectorial force */
1563 fix2 = _mm_macc_ps(dx21,fscal,fix2);
1564 fiy2 = _mm_macc_ps(dy21,fscal,fiy2);
1565 fiz2 = _mm_macc_ps(dz21,fscal,fiz2);
1567 fjx1 = _mm_macc_ps(dx21,fscal,fjx1);
1568 fjy1 = _mm_macc_ps(dy21,fscal,fjy1);
1569 fjz1 = _mm_macc_ps(dz21,fscal,fjz1);
1571 /**************************
1572 * CALCULATE INTERACTIONS *
1573 **************************/
1575 /* COULOMB ELECTROSTATICS */
1576 velec = _mm_mul_ps(qq22,rinv22);
1577 felec = _mm_mul_ps(velec,rinvsq22);
1581 fscal = _mm_andnot_ps(dummy_mask,fscal);
1583 /* Update vectorial force */
1584 fix2 = _mm_macc_ps(dx22,fscal,fix2);
1585 fiy2 = _mm_macc_ps(dy22,fscal,fiy2);
1586 fiz2 = _mm_macc_ps(dz22,fscal,fiz2);
1588 fjx2 = _mm_macc_ps(dx22,fscal,fjx2);
1589 fjy2 = _mm_macc_ps(dy22,fscal,fjy2);
1590 fjz2 = _mm_macc_ps(dz22,fscal,fjz2);
1592 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1593 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1594 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1595 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1597 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1598 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1600 /* Inner loop uses 277 flops */
1603 /* End of innermost loop */
1605 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1606 f+i_coord_offset,fshift+i_shift_offset);
1608 /* Increment number of inner iterations */
1609 inneriter += j_index_end - j_index_start;
1611 /* Outer loop uses 18 flops */
1614 /* Increment number of outer iterations */
1617 /* Update outer/inner flops */
1619 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*277);