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_256_single kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_256_single.h"
48 #include "kernelutil_x86_avx_256_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_single
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: LennardJones
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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 jnrE,jnrF,jnrG,jnrH;
76 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
77 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
85 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr0;
87 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 real * vdwioffsetptr1;
89 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 real * vdwioffsetptr2;
91 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
93 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
94 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
95 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
96 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
97 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
100 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
103 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
104 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
105 __m256 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
106 real rswitch_scalar,d_scalar;
107 __m256 dummy_mask,cutoff_mask;
108 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
109 __m256 one = _mm256_set1_ps(1.0);
110 __m256 two = _mm256_set1_ps(2.0);
116 jindex = nlist->jindex;
118 shiftidx = nlist->shift;
120 shiftvec = fr->shift_vec[0];
121 fshift = fr->fshift[0];
122 facel = _mm256_set1_ps(fr->epsfac);
123 charge = mdatoms->chargeA;
124 krf = _mm256_set1_ps(fr->ic->k_rf);
125 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
126 crf = _mm256_set1_ps(fr->ic->c_rf);
127 nvdwtype = fr->ntype;
129 vdwtype = mdatoms->typeA;
131 /* Setup water-specific parameters */
132 inr = nlist->iinr[0];
133 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
134 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
135 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
136 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
138 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
139 rcutoff_scalar = fr->rcoulomb;
140 rcutoff = _mm256_set1_ps(rcutoff_scalar);
141 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
143 rswitch_scalar = fr->rvdw_switch;
144 rswitch = _mm256_set1_ps(rswitch_scalar);
145 /* Setup switch parameters */
146 d_scalar = rcutoff_scalar-rswitch_scalar;
147 d = _mm256_set1_ps(d_scalar);
148 swV3 = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
149 swV4 = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
150 swV5 = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
151 swF2 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
152 swF3 = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
153 swF4 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
155 /* Avoid stupid compiler warnings */
156 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
169 for(iidx=0;iidx<4*DIM;iidx++)
174 /* Start outer loop over neighborlists */
175 for(iidx=0; iidx<nri; iidx++)
177 /* Load shift vector for this list */
178 i_shift_offset = DIM*shiftidx[iidx];
180 /* Load limits for loop over neighbors */
181 j_index_start = jindex[iidx];
182 j_index_end = jindex[iidx+1];
184 /* Get outer coordinate index */
186 i_coord_offset = DIM*inr;
188 /* Load i particle coords and add shift vector */
189 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
190 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
192 fix0 = _mm256_setzero_ps();
193 fiy0 = _mm256_setzero_ps();
194 fiz0 = _mm256_setzero_ps();
195 fix1 = _mm256_setzero_ps();
196 fiy1 = _mm256_setzero_ps();
197 fiz1 = _mm256_setzero_ps();
198 fix2 = _mm256_setzero_ps();
199 fiy2 = _mm256_setzero_ps();
200 fiz2 = _mm256_setzero_ps();
202 /* Reset potential sums */
203 velecsum = _mm256_setzero_ps();
204 vvdwsum = _mm256_setzero_ps();
206 /* Start inner kernel loop */
207 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
210 /* Get j neighbor index, and coordinate index */
219 j_coord_offsetA = DIM*jnrA;
220 j_coord_offsetB = DIM*jnrB;
221 j_coord_offsetC = DIM*jnrC;
222 j_coord_offsetD = DIM*jnrD;
223 j_coord_offsetE = DIM*jnrE;
224 j_coord_offsetF = DIM*jnrF;
225 j_coord_offsetG = DIM*jnrG;
226 j_coord_offsetH = DIM*jnrH;
228 /* load j atom coordinates */
229 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
230 x+j_coord_offsetC,x+j_coord_offsetD,
231 x+j_coord_offsetE,x+j_coord_offsetF,
232 x+j_coord_offsetG,x+j_coord_offsetH,
235 /* Calculate displacement vector */
236 dx00 = _mm256_sub_ps(ix0,jx0);
237 dy00 = _mm256_sub_ps(iy0,jy0);
238 dz00 = _mm256_sub_ps(iz0,jz0);
239 dx10 = _mm256_sub_ps(ix1,jx0);
240 dy10 = _mm256_sub_ps(iy1,jy0);
241 dz10 = _mm256_sub_ps(iz1,jz0);
242 dx20 = _mm256_sub_ps(ix2,jx0);
243 dy20 = _mm256_sub_ps(iy2,jy0);
244 dz20 = _mm256_sub_ps(iz2,jz0);
246 /* Calculate squared distance and things based on it */
247 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
248 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
249 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
251 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
252 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
253 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
255 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
256 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
257 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
259 /* Load parameters for j particles */
260 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
261 charge+jnrC+0,charge+jnrD+0,
262 charge+jnrE+0,charge+jnrF+0,
263 charge+jnrG+0,charge+jnrH+0);
264 vdwjidx0A = 2*vdwtype[jnrA+0];
265 vdwjidx0B = 2*vdwtype[jnrB+0];
266 vdwjidx0C = 2*vdwtype[jnrC+0];
267 vdwjidx0D = 2*vdwtype[jnrD+0];
268 vdwjidx0E = 2*vdwtype[jnrE+0];
269 vdwjidx0F = 2*vdwtype[jnrF+0];
270 vdwjidx0G = 2*vdwtype[jnrG+0];
271 vdwjidx0H = 2*vdwtype[jnrH+0];
273 fjx0 = _mm256_setzero_ps();
274 fjy0 = _mm256_setzero_ps();
275 fjz0 = _mm256_setzero_ps();
277 /**************************
278 * CALCULATE INTERACTIONS *
279 **************************/
281 if (gmx_mm256_any_lt(rsq00,rcutoff2))
284 r00 = _mm256_mul_ps(rsq00,rinv00);
286 /* Compute parameters for interactions between i and j atoms */
287 qq00 = _mm256_mul_ps(iq0,jq0);
288 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
289 vdwioffsetptr0+vdwjidx0B,
290 vdwioffsetptr0+vdwjidx0C,
291 vdwioffsetptr0+vdwjidx0D,
292 vdwioffsetptr0+vdwjidx0E,
293 vdwioffsetptr0+vdwjidx0F,
294 vdwioffsetptr0+vdwjidx0G,
295 vdwioffsetptr0+vdwjidx0H,
298 /* REACTION-FIELD ELECTROSTATICS */
299 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
300 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
302 /* LENNARD-JONES DISPERSION/REPULSION */
304 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
305 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
306 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
307 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
308 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
310 d = _mm256_sub_ps(r00,rswitch);
311 d = _mm256_max_ps(d,_mm256_setzero_ps());
312 d2 = _mm256_mul_ps(d,d);
313 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
315 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
317 /* Evaluate switch function */
318 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
319 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
320 vvdw = _mm256_mul_ps(vvdw,sw);
321 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
323 /* Update potential sum for this i atom from the interaction with this j atom. */
324 velec = _mm256_and_ps(velec,cutoff_mask);
325 velecsum = _mm256_add_ps(velecsum,velec);
326 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
327 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
329 fscal = _mm256_add_ps(felec,fvdw);
331 fscal = _mm256_and_ps(fscal,cutoff_mask);
333 /* Calculate temporary vectorial force */
334 tx = _mm256_mul_ps(fscal,dx00);
335 ty = _mm256_mul_ps(fscal,dy00);
336 tz = _mm256_mul_ps(fscal,dz00);
338 /* Update vectorial force */
339 fix0 = _mm256_add_ps(fix0,tx);
340 fiy0 = _mm256_add_ps(fiy0,ty);
341 fiz0 = _mm256_add_ps(fiz0,tz);
343 fjx0 = _mm256_add_ps(fjx0,tx);
344 fjy0 = _mm256_add_ps(fjy0,ty);
345 fjz0 = _mm256_add_ps(fjz0,tz);
349 /**************************
350 * CALCULATE INTERACTIONS *
351 **************************/
353 if (gmx_mm256_any_lt(rsq10,rcutoff2))
356 /* Compute parameters for interactions between i and j atoms */
357 qq10 = _mm256_mul_ps(iq1,jq0);
359 /* REACTION-FIELD ELECTROSTATICS */
360 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
361 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
363 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
365 /* Update potential sum for this i atom from the interaction with this j atom. */
366 velec = _mm256_and_ps(velec,cutoff_mask);
367 velecsum = _mm256_add_ps(velecsum,velec);
371 fscal = _mm256_and_ps(fscal,cutoff_mask);
373 /* Calculate temporary vectorial force */
374 tx = _mm256_mul_ps(fscal,dx10);
375 ty = _mm256_mul_ps(fscal,dy10);
376 tz = _mm256_mul_ps(fscal,dz10);
378 /* Update vectorial force */
379 fix1 = _mm256_add_ps(fix1,tx);
380 fiy1 = _mm256_add_ps(fiy1,ty);
381 fiz1 = _mm256_add_ps(fiz1,tz);
383 fjx0 = _mm256_add_ps(fjx0,tx);
384 fjy0 = _mm256_add_ps(fjy0,ty);
385 fjz0 = _mm256_add_ps(fjz0,tz);
389 /**************************
390 * CALCULATE INTERACTIONS *
391 **************************/
393 if (gmx_mm256_any_lt(rsq20,rcutoff2))
396 /* Compute parameters for interactions between i and j atoms */
397 qq20 = _mm256_mul_ps(iq2,jq0);
399 /* REACTION-FIELD ELECTROSTATICS */
400 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
401 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
403 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
405 /* Update potential sum for this i atom from the interaction with this j atom. */
406 velec = _mm256_and_ps(velec,cutoff_mask);
407 velecsum = _mm256_add_ps(velecsum,velec);
411 fscal = _mm256_and_ps(fscal,cutoff_mask);
413 /* Calculate temporary vectorial force */
414 tx = _mm256_mul_ps(fscal,dx20);
415 ty = _mm256_mul_ps(fscal,dy20);
416 tz = _mm256_mul_ps(fscal,dz20);
418 /* Update vectorial force */
419 fix2 = _mm256_add_ps(fix2,tx);
420 fiy2 = _mm256_add_ps(fiy2,ty);
421 fiz2 = _mm256_add_ps(fiz2,tz);
423 fjx0 = _mm256_add_ps(fjx0,tx);
424 fjy0 = _mm256_add_ps(fjy0,ty);
425 fjz0 = _mm256_add_ps(fjz0,tz);
429 fjptrA = f+j_coord_offsetA;
430 fjptrB = f+j_coord_offsetB;
431 fjptrC = f+j_coord_offsetC;
432 fjptrD = f+j_coord_offsetD;
433 fjptrE = f+j_coord_offsetE;
434 fjptrF = f+j_coord_offsetF;
435 fjptrG = f+j_coord_offsetG;
436 fjptrH = f+j_coord_offsetH;
438 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
440 /* Inner loop uses 145 flops */
446 /* Get j neighbor index, and coordinate index */
447 jnrlistA = jjnr[jidx];
448 jnrlistB = jjnr[jidx+1];
449 jnrlistC = jjnr[jidx+2];
450 jnrlistD = jjnr[jidx+3];
451 jnrlistE = jjnr[jidx+4];
452 jnrlistF = jjnr[jidx+5];
453 jnrlistG = jjnr[jidx+6];
454 jnrlistH = jjnr[jidx+7];
455 /* Sign of each element will be negative for non-real atoms.
456 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
457 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
459 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
460 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
462 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
463 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
464 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
465 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
466 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
467 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
468 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
469 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
470 j_coord_offsetA = DIM*jnrA;
471 j_coord_offsetB = DIM*jnrB;
472 j_coord_offsetC = DIM*jnrC;
473 j_coord_offsetD = DIM*jnrD;
474 j_coord_offsetE = DIM*jnrE;
475 j_coord_offsetF = DIM*jnrF;
476 j_coord_offsetG = DIM*jnrG;
477 j_coord_offsetH = DIM*jnrH;
479 /* load j atom coordinates */
480 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
481 x+j_coord_offsetC,x+j_coord_offsetD,
482 x+j_coord_offsetE,x+j_coord_offsetF,
483 x+j_coord_offsetG,x+j_coord_offsetH,
486 /* Calculate displacement vector */
487 dx00 = _mm256_sub_ps(ix0,jx0);
488 dy00 = _mm256_sub_ps(iy0,jy0);
489 dz00 = _mm256_sub_ps(iz0,jz0);
490 dx10 = _mm256_sub_ps(ix1,jx0);
491 dy10 = _mm256_sub_ps(iy1,jy0);
492 dz10 = _mm256_sub_ps(iz1,jz0);
493 dx20 = _mm256_sub_ps(ix2,jx0);
494 dy20 = _mm256_sub_ps(iy2,jy0);
495 dz20 = _mm256_sub_ps(iz2,jz0);
497 /* Calculate squared distance and things based on it */
498 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
499 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
500 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
502 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
503 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
504 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
506 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
507 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
508 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
510 /* Load parameters for j particles */
511 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
512 charge+jnrC+0,charge+jnrD+0,
513 charge+jnrE+0,charge+jnrF+0,
514 charge+jnrG+0,charge+jnrH+0);
515 vdwjidx0A = 2*vdwtype[jnrA+0];
516 vdwjidx0B = 2*vdwtype[jnrB+0];
517 vdwjidx0C = 2*vdwtype[jnrC+0];
518 vdwjidx0D = 2*vdwtype[jnrD+0];
519 vdwjidx0E = 2*vdwtype[jnrE+0];
520 vdwjidx0F = 2*vdwtype[jnrF+0];
521 vdwjidx0G = 2*vdwtype[jnrG+0];
522 vdwjidx0H = 2*vdwtype[jnrH+0];
524 fjx0 = _mm256_setzero_ps();
525 fjy0 = _mm256_setzero_ps();
526 fjz0 = _mm256_setzero_ps();
528 /**************************
529 * CALCULATE INTERACTIONS *
530 **************************/
532 if (gmx_mm256_any_lt(rsq00,rcutoff2))
535 r00 = _mm256_mul_ps(rsq00,rinv00);
536 r00 = _mm256_andnot_ps(dummy_mask,r00);
538 /* Compute parameters for interactions between i and j atoms */
539 qq00 = _mm256_mul_ps(iq0,jq0);
540 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
541 vdwioffsetptr0+vdwjidx0B,
542 vdwioffsetptr0+vdwjidx0C,
543 vdwioffsetptr0+vdwjidx0D,
544 vdwioffsetptr0+vdwjidx0E,
545 vdwioffsetptr0+vdwjidx0F,
546 vdwioffsetptr0+vdwjidx0G,
547 vdwioffsetptr0+vdwjidx0H,
550 /* REACTION-FIELD ELECTROSTATICS */
551 velec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
552 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
554 /* LENNARD-JONES DISPERSION/REPULSION */
556 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
557 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
558 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
559 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
560 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
562 d = _mm256_sub_ps(r00,rswitch);
563 d = _mm256_max_ps(d,_mm256_setzero_ps());
564 d2 = _mm256_mul_ps(d,d);
565 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
567 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
569 /* Evaluate switch function */
570 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
571 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
572 vvdw = _mm256_mul_ps(vvdw,sw);
573 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
575 /* Update potential sum for this i atom from the interaction with this j atom. */
576 velec = _mm256_and_ps(velec,cutoff_mask);
577 velec = _mm256_andnot_ps(dummy_mask,velec);
578 velecsum = _mm256_add_ps(velecsum,velec);
579 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
580 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
581 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
583 fscal = _mm256_add_ps(felec,fvdw);
585 fscal = _mm256_and_ps(fscal,cutoff_mask);
587 fscal = _mm256_andnot_ps(dummy_mask,fscal);
589 /* Calculate temporary vectorial force */
590 tx = _mm256_mul_ps(fscal,dx00);
591 ty = _mm256_mul_ps(fscal,dy00);
592 tz = _mm256_mul_ps(fscal,dz00);
594 /* Update vectorial force */
595 fix0 = _mm256_add_ps(fix0,tx);
596 fiy0 = _mm256_add_ps(fiy0,ty);
597 fiz0 = _mm256_add_ps(fiz0,tz);
599 fjx0 = _mm256_add_ps(fjx0,tx);
600 fjy0 = _mm256_add_ps(fjy0,ty);
601 fjz0 = _mm256_add_ps(fjz0,tz);
605 /**************************
606 * CALCULATE INTERACTIONS *
607 **************************/
609 if (gmx_mm256_any_lt(rsq10,rcutoff2))
612 /* Compute parameters for interactions between i and j atoms */
613 qq10 = _mm256_mul_ps(iq1,jq0);
615 /* REACTION-FIELD ELECTROSTATICS */
616 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
617 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
619 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
621 /* Update potential sum for this i atom from the interaction with this j atom. */
622 velec = _mm256_and_ps(velec,cutoff_mask);
623 velec = _mm256_andnot_ps(dummy_mask,velec);
624 velecsum = _mm256_add_ps(velecsum,velec);
628 fscal = _mm256_and_ps(fscal,cutoff_mask);
630 fscal = _mm256_andnot_ps(dummy_mask,fscal);
632 /* Calculate temporary vectorial force */
633 tx = _mm256_mul_ps(fscal,dx10);
634 ty = _mm256_mul_ps(fscal,dy10);
635 tz = _mm256_mul_ps(fscal,dz10);
637 /* Update vectorial force */
638 fix1 = _mm256_add_ps(fix1,tx);
639 fiy1 = _mm256_add_ps(fiy1,ty);
640 fiz1 = _mm256_add_ps(fiz1,tz);
642 fjx0 = _mm256_add_ps(fjx0,tx);
643 fjy0 = _mm256_add_ps(fjy0,ty);
644 fjz0 = _mm256_add_ps(fjz0,tz);
648 /**************************
649 * CALCULATE INTERACTIONS *
650 **************************/
652 if (gmx_mm256_any_lt(rsq20,rcutoff2))
655 /* Compute parameters for interactions between i and j atoms */
656 qq20 = _mm256_mul_ps(iq2,jq0);
658 /* REACTION-FIELD ELECTROSTATICS */
659 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
660 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
662 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
664 /* Update potential sum for this i atom from the interaction with this j atom. */
665 velec = _mm256_and_ps(velec,cutoff_mask);
666 velec = _mm256_andnot_ps(dummy_mask,velec);
667 velecsum = _mm256_add_ps(velecsum,velec);
671 fscal = _mm256_and_ps(fscal,cutoff_mask);
673 fscal = _mm256_andnot_ps(dummy_mask,fscal);
675 /* Calculate temporary vectorial force */
676 tx = _mm256_mul_ps(fscal,dx20);
677 ty = _mm256_mul_ps(fscal,dy20);
678 tz = _mm256_mul_ps(fscal,dz20);
680 /* Update vectorial force */
681 fix2 = _mm256_add_ps(fix2,tx);
682 fiy2 = _mm256_add_ps(fiy2,ty);
683 fiz2 = _mm256_add_ps(fiz2,tz);
685 fjx0 = _mm256_add_ps(fjx0,tx);
686 fjy0 = _mm256_add_ps(fjy0,ty);
687 fjz0 = _mm256_add_ps(fjz0,tz);
691 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
692 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
693 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
694 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
695 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
696 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
697 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
698 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
700 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
702 /* Inner loop uses 146 flops */
705 /* End of innermost loop */
707 gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
708 f+i_coord_offset,fshift+i_shift_offset);
711 /* Update potential energies */
712 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
713 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
715 /* Increment number of inner iterations */
716 inneriter += j_index_end - j_index_start;
718 /* Outer loop uses 20 flops */
721 /* Increment number of outer iterations */
724 /* Update outer/inner flops */
726 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*146);
729 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_single
730 * Electrostatics interaction: ReactionField
731 * VdW interaction: LennardJones
732 * Geometry: Water3-Particle
733 * Calculate force/pot: Force
736 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_single
737 (t_nblist * gmx_restrict nlist,
738 rvec * gmx_restrict xx,
739 rvec * gmx_restrict ff,
740 t_forcerec * gmx_restrict fr,
741 t_mdatoms * gmx_restrict mdatoms,
742 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
743 t_nrnb * gmx_restrict nrnb)
745 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
746 * just 0 for non-waters.
747 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
748 * jnr indices corresponding to data put in the four positions in the SIMD register.
750 int i_shift_offset,i_coord_offset,outeriter,inneriter;
751 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
752 int jnrA,jnrB,jnrC,jnrD;
753 int jnrE,jnrF,jnrG,jnrH;
754 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
755 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
756 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
757 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
758 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
760 real *shiftvec,*fshift,*x,*f;
761 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
763 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
764 real * vdwioffsetptr0;
765 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
766 real * vdwioffsetptr1;
767 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
768 real * vdwioffsetptr2;
769 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
770 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
771 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
772 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
773 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
774 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
775 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
778 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
781 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
782 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
783 __m256 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
784 real rswitch_scalar,d_scalar;
785 __m256 dummy_mask,cutoff_mask;
786 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
787 __m256 one = _mm256_set1_ps(1.0);
788 __m256 two = _mm256_set1_ps(2.0);
794 jindex = nlist->jindex;
796 shiftidx = nlist->shift;
798 shiftvec = fr->shift_vec[0];
799 fshift = fr->fshift[0];
800 facel = _mm256_set1_ps(fr->epsfac);
801 charge = mdatoms->chargeA;
802 krf = _mm256_set1_ps(fr->ic->k_rf);
803 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
804 crf = _mm256_set1_ps(fr->ic->c_rf);
805 nvdwtype = fr->ntype;
807 vdwtype = mdatoms->typeA;
809 /* Setup water-specific parameters */
810 inr = nlist->iinr[0];
811 iq0 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
812 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
813 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
814 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
816 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
817 rcutoff_scalar = fr->rcoulomb;
818 rcutoff = _mm256_set1_ps(rcutoff_scalar);
819 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
821 rswitch_scalar = fr->rvdw_switch;
822 rswitch = _mm256_set1_ps(rswitch_scalar);
823 /* Setup switch parameters */
824 d_scalar = rcutoff_scalar-rswitch_scalar;
825 d = _mm256_set1_ps(d_scalar);
826 swV3 = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
827 swV4 = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
828 swV5 = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
829 swF2 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
830 swF3 = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
831 swF4 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
833 /* Avoid stupid compiler warnings */
834 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
847 for(iidx=0;iidx<4*DIM;iidx++)
852 /* Start outer loop over neighborlists */
853 for(iidx=0; iidx<nri; iidx++)
855 /* Load shift vector for this list */
856 i_shift_offset = DIM*shiftidx[iidx];
858 /* Load limits for loop over neighbors */
859 j_index_start = jindex[iidx];
860 j_index_end = jindex[iidx+1];
862 /* Get outer coordinate index */
864 i_coord_offset = DIM*inr;
866 /* Load i particle coords and add shift vector */
867 gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
868 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
870 fix0 = _mm256_setzero_ps();
871 fiy0 = _mm256_setzero_ps();
872 fiz0 = _mm256_setzero_ps();
873 fix1 = _mm256_setzero_ps();
874 fiy1 = _mm256_setzero_ps();
875 fiz1 = _mm256_setzero_ps();
876 fix2 = _mm256_setzero_ps();
877 fiy2 = _mm256_setzero_ps();
878 fiz2 = _mm256_setzero_ps();
880 /* Start inner kernel loop */
881 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
884 /* Get j neighbor index, and coordinate index */
893 j_coord_offsetA = DIM*jnrA;
894 j_coord_offsetB = DIM*jnrB;
895 j_coord_offsetC = DIM*jnrC;
896 j_coord_offsetD = DIM*jnrD;
897 j_coord_offsetE = DIM*jnrE;
898 j_coord_offsetF = DIM*jnrF;
899 j_coord_offsetG = DIM*jnrG;
900 j_coord_offsetH = DIM*jnrH;
902 /* load j atom coordinates */
903 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
904 x+j_coord_offsetC,x+j_coord_offsetD,
905 x+j_coord_offsetE,x+j_coord_offsetF,
906 x+j_coord_offsetG,x+j_coord_offsetH,
909 /* Calculate displacement vector */
910 dx00 = _mm256_sub_ps(ix0,jx0);
911 dy00 = _mm256_sub_ps(iy0,jy0);
912 dz00 = _mm256_sub_ps(iz0,jz0);
913 dx10 = _mm256_sub_ps(ix1,jx0);
914 dy10 = _mm256_sub_ps(iy1,jy0);
915 dz10 = _mm256_sub_ps(iz1,jz0);
916 dx20 = _mm256_sub_ps(ix2,jx0);
917 dy20 = _mm256_sub_ps(iy2,jy0);
918 dz20 = _mm256_sub_ps(iz2,jz0);
920 /* Calculate squared distance and things based on it */
921 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
922 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
923 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
925 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
926 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
927 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
929 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
930 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
931 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
933 /* Load parameters for j particles */
934 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
935 charge+jnrC+0,charge+jnrD+0,
936 charge+jnrE+0,charge+jnrF+0,
937 charge+jnrG+0,charge+jnrH+0);
938 vdwjidx0A = 2*vdwtype[jnrA+0];
939 vdwjidx0B = 2*vdwtype[jnrB+0];
940 vdwjidx0C = 2*vdwtype[jnrC+0];
941 vdwjidx0D = 2*vdwtype[jnrD+0];
942 vdwjidx0E = 2*vdwtype[jnrE+0];
943 vdwjidx0F = 2*vdwtype[jnrF+0];
944 vdwjidx0G = 2*vdwtype[jnrG+0];
945 vdwjidx0H = 2*vdwtype[jnrH+0];
947 fjx0 = _mm256_setzero_ps();
948 fjy0 = _mm256_setzero_ps();
949 fjz0 = _mm256_setzero_ps();
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
955 if (gmx_mm256_any_lt(rsq00,rcutoff2))
958 r00 = _mm256_mul_ps(rsq00,rinv00);
960 /* Compute parameters for interactions between i and j atoms */
961 qq00 = _mm256_mul_ps(iq0,jq0);
962 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
963 vdwioffsetptr0+vdwjidx0B,
964 vdwioffsetptr0+vdwjidx0C,
965 vdwioffsetptr0+vdwjidx0D,
966 vdwioffsetptr0+vdwjidx0E,
967 vdwioffsetptr0+vdwjidx0F,
968 vdwioffsetptr0+vdwjidx0G,
969 vdwioffsetptr0+vdwjidx0H,
972 /* REACTION-FIELD ELECTROSTATICS */
973 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
975 /* LENNARD-JONES DISPERSION/REPULSION */
977 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
978 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
979 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
980 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
981 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
983 d = _mm256_sub_ps(r00,rswitch);
984 d = _mm256_max_ps(d,_mm256_setzero_ps());
985 d2 = _mm256_mul_ps(d,d);
986 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
988 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
990 /* Evaluate switch function */
991 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
992 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
993 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
995 fscal = _mm256_add_ps(felec,fvdw);
997 fscal = _mm256_and_ps(fscal,cutoff_mask);
999 /* Calculate temporary vectorial force */
1000 tx = _mm256_mul_ps(fscal,dx00);
1001 ty = _mm256_mul_ps(fscal,dy00);
1002 tz = _mm256_mul_ps(fscal,dz00);
1004 /* Update vectorial force */
1005 fix0 = _mm256_add_ps(fix0,tx);
1006 fiy0 = _mm256_add_ps(fiy0,ty);
1007 fiz0 = _mm256_add_ps(fiz0,tz);
1009 fjx0 = _mm256_add_ps(fjx0,tx);
1010 fjy0 = _mm256_add_ps(fjy0,ty);
1011 fjz0 = _mm256_add_ps(fjz0,tz);
1015 /**************************
1016 * CALCULATE INTERACTIONS *
1017 **************************/
1019 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1022 /* Compute parameters for interactions between i and j atoms */
1023 qq10 = _mm256_mul_ps(iq1,jq0);
1025 /* REACTION-FIELD ELECTROSTATICS */
1026 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1028 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1032 fscal = _mm256_and_ps(fscal,cutoff_mask);
1034 /* Calculate temporary vectorial force */
1035 tx = _mm256_mul_ps(fscal,dx10);
1036 ty = _mm256_mul_ps(fscal,dy10);
1037 tz = _mm256_mul_ps(fscal,dz10);
1039 /* Update vectorial force */
1040 fix1 = _mm256_add_ps(fix1,tx);
1041 fiy1 = _mm256_add_ps(fiy1,ty);
1042 fiz1 = _mm256_add_ps(fiz1,tz);
1044 fjx0 = _mm256_add_ps(fjx0,tx);
1045 fjy0 = _mm256_add_ps(fjy0,ty);
1046 fjz0 = _mm256_add_ps(fjz0,tz);
1050 /**************************
1051 * CALCULATE INTERACTIONS *
1052 **************************/
1054 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1057 /* Compute parameters for interactions between i and j atoms */
1058 qq20 = _mm256_mul_ps(iq2,jq0);
1060 /* REACTION-FIELD ELECTROSTATICS */
1061 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1063 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1067 fscal = _mm256_and_ps(fscal,cutoff_mask);
1069 /* Calculate temporary vectorial force */
1070 tx = _mm256_mul_ps(fscal,dx20);
1071 ty = _mm256_mul_ps(fscal,dy20);
1072 tz = _mm256_mul_ps(fscal,dz20);
1074 /* Update vectorial force */
1075 fix2 = _mm256_add_ps(fix2,tx);
1076 fiy2 = _mm256_add_ps(fiy2,ty);
1077 fiz2 = _mm256_add_ps(fiz2,tz);
1079 fjx0 = _mm256_add_ps(fjx0,tx);
1080 fjy0 = _mm256_add_ps(fjy0,ty);
1081 fjz0 = _mm256_add_ps(fjz0,tz);
1085 fjptrA = f+j_coord_offsetA;
1086 fjptrB = f+j_coord_offsetB;
1087 fjptrC = f+j_coord_offsetC;
1088 fjptrD = f+j_coord_offsetD;
1089 fjptrE = f+j_coord_offsetE;
1090 fjptrF = f+j_coord_offsetF;
1091 fjptrG = f+j_coord_offsetG;
1092 fjptrH = f+j_coord_offsetH;
1094 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1096 /* Inner loop uses 124 flops */
1099 if(jidx<j_index_end)
1102 /* Get j neighbor index, and coordinate index */
1103 jnrlistA = jjnr[jidx];
1104 jnrlistB = jjnr[jidx+1];
1105 jnrlistC = jjnr[jidx+2];
1106 jnrlistD = jjnr[jidx+3];
1107 jnrlistE = jjnr[jidx+4];
1108 jnrlistF = jjnr[jidx+5];
1109 jnrlistG = jjnr[jidx+6];
1110 jnrlistH = jjnr[jidx+7];
1111 /* Sign of each element will be negative for non-real atoms.
1112 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1113 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1115 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1116 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1118 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1119 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1120 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1121 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1122 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
1123 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
1124 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
1125 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
1126 j_coord_offsetA = DIM*jnrA;
1127 j_coord_offsetB = DIM*jnrB;
1128 j_coord_offsetC = DIM*jnrC;
1129 j_coord_offsetD = DIM*jnrD;
1130 j_coord_offsetE = DIM*jnrE;
1131 j_coord_offsetF = DIM*jnrF;
1132 j_coord_offsetG = DIM*jnrG;
1133 j_coord_offsetH = DIM*jnrH;
1135 /* load j atom coordinates */
1136 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1137 x+j_coord_offsetC,x+j_coord_offsetD,
1138 x+j_coord_offsetE,x+j_coord_offsetF,
1139 x+j_coord_offsetG,x+j_coord_offsetH,
1142 /* Calculate displacement vector */
1143 dx00 = _mm256_sub_ps(ix0,jx0);
1144 dy00 = _mm256_sub_ps(iy0,jy0);
1145 dz00 = _mm256_sub_ps(iz0,jz0);
1146 dx10 = _mm256_sub_ps(ix1,jx0);
1147 dy10 = _mm256_sub_ps(iy1,jy0);
1148 dz10 = _mm256_sub_ps(iz1,jz0);
1149 dx20 = _mm256_sub_ps(ix2,jx0);
1150 dy20 = _mm256_sub_ps(iy2,jy0);
1151 dz20 = _mm256_sub_ps(iz2,jz0);
1153 /* Calculate squared distance and things based on it */
1154 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1155 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1156 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1158 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
1159 rinv10 = gmx_mm256_invsqrt_ps(rsq10);
1160 rinv20 = gmx_mm256_invsqrt_ps(rsq20);
1162 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
1163 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1164 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1166 /* Load parameters for j particles */
1167 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1168 charge+jnrC+0,charge+jnrD+0,
1169 charge+jnrE+0,charge+jnrF+0,
1170 charge+jnrG+0,charge+jnrH+0);
1171 vdwjidx0A = 2*vdwtype[jnrA+0];
1172 vdwjidx0B = 2*vdwtype[jnrB+0];
1173 vdwjidx0C = 2*vdwtype[jnrC+0];
1174 vdwjidx0D = 2*vdwtype[jnrD+0];
1175 vdwjidx0E = 2*vdwtype[jnrE+0];
1176 vdwjidx0F = 2*vdwtype[jnrF+0];
1177 vdwjidx0G = 2*vdwtype[jnrG+0];
1178 vdwjidx0H = 2*vdwtype[jnrH+0];
1180 fjx0 = _mm256_setzero_ps();
1181 fjy0 = _mm256_setzero_ps();
1182 fjz0 = _mm256_setzero_ps();
1184 /**************************
1185 * CALCULATE INTERACTIONS *
1186 **************************/
1188 if (gmx_mm256_any_lt(rsq00,rcutoff2))
1191 r00 = _mm256_mul_ps(rsq00,rinv00);
1192 r00 = _mm256_andnot_ps(dummy_mask,r00);
1194 /* Compute parameters for interactions between i and j atoms */
1195 qq00 = _mm256_mul_ps(iq0,jq0);
1196 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1197 vdwioffsetptr0+vdwjidx0B,
1198 vdwioffsetptr0+vdwjidx0C,
1199 vdwioffsetptr0+vdwjidx0D,
1200 vdwioffsetptr0+vdwjidx0E,
1201 vdwioffsetptr0+vdwjidx0F,
1202 vdwioffsetptr0+vdwjidx0G,
1203 vdwioffsetptr0+vdwjidx0H,
1206 /* REACTION-FIELD ELECTROSTATICS */
1207 felec = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1209 /* LENNARD-JONES DISPERSION/REPULSION */
1211 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1212 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
1213 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1214 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1215 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1217 d = _mm256_sub_ps(r00,rswitch);
1218 d = _mm256_max_ps(d,_mm256_setzero_ps());
1219 d2 = _mm256_mul_ps(d,d);
1220 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1222 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1224 /* Evaluate switch function */
1225 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1226 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1227 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1229 fscal = _mm256_add_ps(felec,fvdw);
1231 fscal = _mm256_and_ps(fscal,cutoff_mask);
1233 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1235 /* Calculate temporary vectorial force */
1236 tx = _mm256_mul_ps(fscal,dx00);
1237 ty = _mm256_mul_ps(fscal,dy00);
1238 tz = _mm256_mul_ps(fscal,dz00);
1240 /* Update vectorial force */
1241 fix0 = _mm256_add_ps(fix0,tx);
1242 fiy0 = _mm256_add_ps(fiy0,ty);
1243 fiz0 = _mm256_add_ps(fiz0,tz);
1245 fjx0 = _mm256_add_ps(fjx0,tx);
1246 fjy0 = _mm256_add_ps(fjy0,ty);
1247 fjz0 = _mm256_add_ps(fjz0,tz);
1251 /**************************
1252 * CALCULATE INTERACTIONS *
1253 **************************/
1255 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1258 /* Compute parameters for interactions between i and j atoms */
1259 qq10 = _mm256_mul_ps(iq1,jq0);
1261 /* REACTION-FIELD ELECTROSTATICS */
1262 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1264 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1268 fscal = _mm256_and_ps(fscal,cutoff_mask);
1270 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1272 /* Calculate temporary vectorial force */
1273 tx = _mm256_mul_ps(fscal,dx10);
1274 ty = _mm256_mul_ps(fscal,dy10);
1275 tz = _mm256_mul_ps(fscal,dz10);
1277 /* Update vectorial force */
1278 fix1 = _mm256_add_ps(fix1,tx);
1279 fiy1 = _mm256_add_ps(fiy1,ty);
1280 fiz1 = _mm256_add_ps(fiz1,tz);
1282 fjx0 = _mm256_add_ps(fjx0,tx);
1283 fjy0 = _mm256_add_ps(fjy0,ty);
1284 fjz0 = _mm256_add_ps(fjz0,tz);
1288 /**************************
1289 * CALCULATE INTERACTIONS *
1290 **************************/
1292 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1295 /* Compute parameters for interactions between i and j atoms */
1296 qq20 = _mm256_mul_ps(iq2,jq0);
1298 /* REACTION-FIELD ELECTROSTATICS */
1299 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1301 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1305 fscal = _mm256_and_ps(fscal,cutoff_mask);
1307 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1309 /* Calculate temporary vectorial force */
1310 tx = _mm256_mul_ps(fscal,dx20);
1311 ty = _mm256_mul_ps(fscal,dy20);
1312 tz = _mm256_mul_ps(fscal,dz20);
1314 /* Update vectorial force */
1315 fix2 = _mm256_add_ps(fix2,tx);
1316 fiy2 = _mm256_add_ps(fiy2,ty);
1317 fiz2 = _mm256_add_ps(fiz2,tz);
1319 fjx0 = _mm256_add_ps(fjx0,tx);
1320 fjy0 = _mm256_add_ps(fjy0,ty);
1321 fjz0 = _mm256_add_ps(fjz0,tz);
1325 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1326 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1327 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1328 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1329 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1330 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1331 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1332 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1334 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1336 /* Inner loop uses 125 flops */
1339 /* End of innermost loop */
1341 gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1342 f+i_coord_offset,fshift+i_shift_offset);
1344 /* Increment number of inner iterations */
1345 inneriter += j_index_end - j_index_start;
1347 /* Outer loop uses 18 flops */
1350 /* Increment number of outer iterations */
1353 /* Update outer/inner flops */
1355 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*125);