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 sse4_1_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_sse4_1_single.h"
50 #include "kernelutil_x86_sse4_1_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_sse4_1_single
54 * Electrostatics interaction: Ewald
55 * VdW interaction: None
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_sse4_1_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
84 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
92 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
94 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
95 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
96 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
99 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
101 __m128 dummy_mask,cutoff_mask;
102 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
103 __m128 one = _mm_set1_ps(1.0);
104 __m128 two = _mm_set1_ps(2.0);
110 jindex = nlist->jindex;
112 shiftidx = nlist->shift;
114 shiftvec = fr->shift_vec[0];
115 fshift = fr->fshift[0];
116 facel = _mm_set1_ps(fr->epsfac);
117 charge = mdatoms->chargeA;
119 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
120 ewtab = fr->ic->tabq_coul_FDV0;
121 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
122 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
124 /* Setup water-specific parameters */
125 inr = nlist->iinr[0];
126 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
127 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
128 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
130 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
131 rcutoff_scalar = fr->rcoulomb;
132 rcutoff = _mm_set1_ps(rcutoff_scalar);
133 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
135 /* Avoid stupid compiler warnings */
136 jnrA = jnrB = jnrC = jnrD = 0;
145 for(iidx=0;iidx<4*DIM;iidx++)
150 /* Start outer loop over neighborlists */
151 for(iidx=0; iidx<nri; iidx++)
153 /* Load shift vector for this list */
154 i_shift_offset = DIM*shiftidx[iidx];
156 /* Load limits for loop over neighbors */
157 j_index_start = jindex[iidx];
158 j_index_end = jindex[iidx+1];
160 /* Get outer coordinate index */
162 i_coord_offset = DIM*inr;
164 /* Load i particle coords and add shift vector */
165 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
166 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
168 fix0 = _mm_setzero_ps();
169 fiy0 = _mm_setzero_ps();
170 fiz0 = _mm_setzero_ps();
171 fix1 = _mm_setzero_ps();
172 fiy1 = _mm_setzero_ps();
173 fiz1 = _mm_setzero_ps();
174 fix2 = _mm_setzero_ps();
175 fiy2 = _mm_setzero_ps();
176 fiz2 = _mm_setzero_ps();
178 /* Reset potential sums */
179 velecsum = _mm_setzero_ps();
181 /* Start inner kernel loop */
182 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
185 /* Get j neighbor index, and coordinate index */
190 j_coord_offsetA = DIM*jnrA;
191 j_coord_offsetB = DIM*jnrB;
192 j_coord_offsetC = DIM*jnrC;
193 j_coord_offsetD = DIM*jnrD;
195 /* load j atom coordinates */
196 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
197 x+j_coord_offsetC,x+j_coord_offsetD,
200 /* Calculate displacement vector */
201 dx00 = _mm_sub_ps(ix0,jx0);
202 dy00 = _mm_sub_ps(iy0,jy0);
203 dz00 = _mm_sub_ps(iz0,jz0);
204 dx10 = _mm_sub_ps(ix1,jx0);
205 dy10 = _mm_sub_ps(iy1,jy0);
206 dz10 = _mm_sub_ps(iz1,jz0);
207 dx20 = _mm_sub_ps(ix2,jx0);
208 dy20 = _mm_sub_ps(iy2,jy0);
209 dz20 = _mm_sub_ps(iz2,jz0);
211 /* Calculate squared distance and things based on it */
212 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
213 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
214 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
216 rinv00 = gmx_mm_invsqrt_ps(rsq00);
217 rinv10 = gmx_mm_invsqrt_ps(rsq10);
218 rinv20 = gmx_mm_invsqrt_ps(rsq20);
220 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
221 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
222 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
224 /* Load parameters for j particles */
225 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
226 charge+jnrC+0,charge+jnrD+0);
228 fjx0 = _mm_setzero_ps();
229 fjy0 = _mm_setzero_ps();
230 fjz0 = _mm_setzero_ps();
232 /**************************
233 * CALCULATE INTERACTIONS *
234 **************************/
236 if (gmx_mm_any_lt(rsq00,rcutoff2))
239 r00 = _mm_mul_ps(rsq00,rinv00);
241 /* Compute parameters for interactions between i and j atoms */
242 qq00 = _mm_mul_ps(iq0,jq0);
244 /* EWALD ELECTROSTATICS */
246 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
247 ewrt = _mm_mul_ps(r00,ewtabscale);
248 ewitab = _mm_cvttps_epi32(ewrt);
249 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
250 ewitab = _mm_slli_epi32(ewitab,2);
251 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
252 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
253 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
254 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
255 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
256 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
257 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
258 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
259 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
261 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
263 /* Update potential sum for this i atom from the interaction with this j atom. */
264 velec = _mm_and_ps(velec,cutoff_mask);
265 velecsum = _mm_add_ps(velecsum,velec);
269 fscal = _mm_and_ps(fscal,cutoff_mask);
271 /* Calculate temporary vectorial force */
272 tx = _mm_mul_ps(fscal,dx00);
273 ty = _mm_mul_ps(fscal,dy00);
274 tz = _mm_mul_ps(fscal,dz00);
276 /* Update vectorial force */
277 fix0 = _mm_add_ps(fix0,tx);
278 fiy0 = _mm_add_ps(fiy0,ty);
279 fiz0 = _mm_add_ps(fiz0,tz);
281 fjx0 = _mm_add_ps(fjx0,tx);
282 fjy0 = _mm_add_ps(fjy0,ty);
283 fjz0 = _mm_add_ps(fjz0,tz);
287 /**************************
288 * CALCULATE INTERACTIONS *
289 **************************/
291 if (gmx_mm_any_lt(rsq10,rcutoff2))
294 r10 = _mm_mul_ps(rsq10,rinv10);
296 /* Compute parameters for interactions between i and j atoms */
297 qq10 = _mm_mul_ps(iq1,jq0);
299 /* EWALD ELECTROSTATICS */
301 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
302 ewrt = _mm_mul_ps(r10,ewtabscale);
303 ewitab = _mm_cvttps_epi32(ewrt);
304 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
305 ewitab = _mm_slli_epi32(ewitab,2);
306 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
307 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
308 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
309 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
310 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
311 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
312 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
313 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
314 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
316 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
318 /* Update potential sum for this i atom from the interaction with this j atom. */
319 velec = _mm_and_ps(velec,cutoff_mask);
320 velecsum = _mm_add_ps(velecsum,velec);
324 fscal = _mm_and_ps(fscal,cutoff_mask);
326 /* Calculate temporary vectorial force */
327 tx = _mm_mul_ps(fscal,dx10);
328 ty = _mm_mul_ps(fscal,dy10);
329 tz = _mm_mul_ps(fscal,dz10);
331 /* Update vectorial force */
332 fix1 = _mm_add_ps(fix1,tx);
333 fiy1 = _mm_add_ps(fiy1,ty);
334 fiz1 = _mm_add_ps(fiz1,tz);
336 fjx0 = _mm_add_ps(fjx0,tx);
337 fjy0 = _mm_add_ps(fjy0,ty);
338 fjz0 = _mm_add_ps(fjz0,tz);
342 /**************************
343 * CALCULATE INTERACTIONS *
344 **************************/
346 if (gmx_mm_any_lt(rsq20,rcutoff2))
349 r20 = _mm_mul_ps(rsq20,rinv20);
351 /* Compute parameters for interactions between i and j atoms */
352 qq20 = _mm_mul_ps(iq2,jq0);
354 /* EWALD ELECTROSTATICS */
356 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
357 ewrt = _mm_mul_ps(r20,ewtabscale);
358 ewitab = _mm_cvttps_epi32(ewrt);
359 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
360 ewitab = _mm_slli_epi32(ewitab,2);
361 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
362 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
363 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
364 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
365 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
366 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
367 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
368 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
369 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
371 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
373 /* Update potential sum for this i atom from the interaction with this j atom. */
374 velec = _mm_and_ps(velec,cutoff_mask);
375 velecsum = _mm_add_ps(velecsum,velec);
379 fscal = _mm_and_ps(fscal,cutoff_mask);
381 /* Calculate temporary vectorial force */
382 tx = _mm_mul_ps(fscal,dx20);
383 ty = _mm_mul_ps(fscal,dy20);
384 tz = _mm_mul_ps(fscal,dz20);
386 /* Update vectorial force */
387 fix2 = _mm_add_ps(fix2,tx);
388 fiy2 = _mm_add_ps(fiy2,ty);
389 fiz2 = _mm_add_ps(fiz2,tz);
391 fjx0 = _mm_add_ps(fjx0,tx);
392 fjy0 = _mm_add_ps(fjy0,ty);
393 fjz0 = _mm_add_ps(fjz0,tz);
397 fjptrA = f+j_coord_offsetA;
398 fjptrB = f+j_coord_offsetB;
399 fjptrC = f+j_coord_offsetC;
400 fjptrD = f+j_coord_offsetD;
402 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
404 /* Inner loop uses 138 flops */
410 /* Get j neighbor index, and coordinate index */
411 jnrlistA = jjnr[jidx];
412 jnrlistB = jjnr[jidx+1];
413 jnrlistC = jjnr[jidx+2];
414 jnrlistD = jjnr[jidx+3];
415 /* Sign of each element will be negative for non-real atoms.
416 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
417 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
419 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
420 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
421 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
422 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
423 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
424 j_coord_offsetA = DIM*jnrA;
425 j_coord_offsetB = DIM*jnrB;
426 j_coord_offsetC = DIM*jnrC;
427 j_coord_offsetD = DIM*jnrD;
429 /* load j atom coordinates */
430 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
431 x+j_coord_offsetC,x+j_coord_offsetD,
434 /* Calculate displacement vector */
435 dx00 = _mm_sub_ps(ix0,jx0);
436 dy00 = _mm_sub_ps(iy0,jy0);
437 dz00 = _mm_sub_ps(iz0,jz0);
438 dx10 = _mm_sub_ps(ix1,jx0);
439 dy10 = _mm_sub_ps(iy1,jy0);
440 dz10 = _mm_sub_ps(iz1,jz0);
441 dx20 = _mm_sub_ps(ix2,jx0);
442 dy20 = _mm_sub_ps(iy2,jy0);
443 dz20 = _mm_sub_ps(iz2,jz0);
445 /* Calculate squared distance and things based on it */
446 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
447 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
448 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
450 rinv00 = gmx_mm_invsqrt_ps(rsq00);
451 rinv10 = gmx_mm_invsqrt_ps(rsq10);
452 rinv20 = gmx_mm_invsqrt_ps(rsq20);
454 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
455 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
456 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
458 /* Load parameters for j particles */
459 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
460 charge+jnrC+0,charge+jnrD+0);
462 fjx0 = _mm_setzero_ps();
463 fjy0 = _mm_setzero_ps();
464 fjz0 = _mm_setzero_ps();
466 /**************************
467 * CALCULATE INTERACTIONS *
468 **************************/
470 if (gmx_mm_any_lt(rsq00,rcutoff2))
473 r00 = _mm_mul_ps(rsq00,rinv00);
474 r00 = _mm_andnot_ps(dummy_mask,r00);
476 /* Compute parameters for interactions between i and j atoms */
477 qq00 = _mm_mul_ps(iq0,jq0);
479 /* EWALD ELECTROSTATICS */
481 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
482 ewrt = _mm_mul_ps(r00,ewtabscale);
483 ewitab = _mm_cvttps_epi32(ewrt);
484 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
485 ewitab = _mm_slli_epi32(ewitab,2);
486 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
487 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
488 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
489 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
490 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
491 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
492 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
493 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
494 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
496 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
498 /* Update potential sum for this i atom from the interaction with this j atom. */
499 velec = _mm_and_ps(velec,cutoff_mask);
500 velec = _mm_andnot_ps(dummy_mask,velec);
501 velecsum = _mm_add_ps(velecsum,velec);
505 fscal = _mm_and_ps(fscal,cutoff_mask);
507 fscal = _mm_andnot_ps(dummy_mask,fscal);
509 /* Calculate temporary vectorial force */
510 tx = _mm_mul_ps(fscal,dx00);
511 ty = _mm_mul_ps(fscal,dy00);
512 tz = _mm_mul_ps(fscal,dz00);
514 /* Update vectorial force */
515 fix0 = _mm_add_ps(fix0,tx);
516 fiy0 = _mm_add_ps(fiy0,ty);
517 fiz0 = _mm_add_ps(fiz0,tz);
519 fjx0 = _mm_add_ps(fjx0,tx);
520 fjy0 = _mm_add_ps(fjy0,ty);
521 fjz0 = _mm_add_ps(fjz0,tz);
525 /**************************
526 * CALCULATE INTERACTIONS *
527 **************************/
529 if (gmx_mm_any_lt(rsq10,rcutoff2))
532 r10 = _mm_mul_ps(rsq10,rinv10);
533 r10 = _mm_andnot_ps(dummy_mask,r10);
535 /* Compute parameters for interactions between i and j atoms */
536 qq10 = _mm_mul_ps(iq1,jq0);
538 /* EWALD ELECTROSTATICS */
540 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
541 ewrt = _mm_mul_ps(r10,ewtabscale);
542 ewitab = _mm_cvttps_epi32(ewrt);
543 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
544 ewitab = _mm_slli_epi32(ewitab,2);
545 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
546 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
547 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
548 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
549 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
550 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
551 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
552 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
553 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
555 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
557 /* Update potential sum for this i atom from the interaction with this j atom. */
558 velec = _mm_and_ps(velec,cutoff_mask);
559 velec = _mm_andnot_ps(dummy_mask,velec);
560 velecsum = _mm_add_ps(velecsum,velec);
564 fscal = _mm_and_ps(fscal,cutoff_mask);
566 fscal = _mm_andnot_ps(dummy_mask,fscal);
568 /* Calculate temporary vectorial force */
569 tx = _mm_mul_ps(fscal,dx10);
570 ty = _mm_mul_ps(fscal,dy10);
571 tz = _mm_mul_ps(fscal,dz10);
573 /* Update vectorial force */
574 fix1 = _mm_add_ps(fix1,tx);
575 fiy1 = _mm_add_ps(fiy1,ty);
576 fiz1 = _mm_add_ps(fiz1,tz);
578 fjx0 = _mm_add_ps(fjx0,tx);
579 fjy0 = _mm_add_ps(fjy0,ty);
580 fjz0 = _mm_add_ps(fjz0,tz);
584 /**************************
585 * CALCULATE INTERACTIONS *
586 **************************/
588 if (gmx_mm_any_lt(rsq20,rcutoff2))
591 r20 = _mm_mul_ps(rsq20,rinv20);
592 r20 = _mm_andnot_ps(dummy_mask,r20);
594 /* Compute parameters for interactions between i and j atoms */
595 qq20 = _mm_mul_ps(iq2,jq0);
597 /* EWALD ELECTROSTATICS */
599 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
600 ewrt = _mm_mul_ps(r20,ewtabscale);
601 ewitab = _mm_cvttps_epi32(ewrt);
602 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
603 ewitab = _mm_slli_epi32(ewitab,2);
604 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
605 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
606 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
607 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
608 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
609 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
610 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
611 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
612 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
614 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
616 /* Update potential sum for this i atom from the interaction with this j atom. */
617 velec = _mm_and_ps(velec,cutoff_mask);
618 velec = _mm_andnot_ps(dummy_mask,velec);
619 velecsum = _mm_add_ps(velecsum,velec);
623 fscal = _mm_and_ps(fscal,cutoff_mask);
625 fscal = _mm_andnot_ps(dummy_mask,fscal);
627 /* Calculate temporary vectorial force */
628 tx = _mm_mul_ps(fscal,dx20);
629 ty = _mm_mul_ps(fscal,dy20);
630 tz = _mm_mul_ps(fscal,dz20);
632 /* Update vectorial force */
633 fix2 = _mm_add_ps(fix2,tx);
634 fiy2 = _mm_add_ps(fiy2,ty);
635 fiz2 = _mm_add_ps(fiz2,tz);
637 fjx0 = _mm_add_ps(fjx0,tx);
638 fjy0 = _mm_add_ps(fjy0,ty);
639 fjz0 = _mm_add_ps(fjz0,tz);
643 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
644 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
645 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
646 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
648 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
650 /* Inner loop uses 141 flops */
653 /* End of innermost loop */
655 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
656 f+i_coord_offset,fshift+i_shift_offset);
659 /* Update potential energies */
660 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
662 /* Increment number of inner iterations */
663 inneriter += j_index_end - j_index_start;
665 /* Outer loop uses 19 flops */
668 /* Increment number of outer iterations */
671 /* Update outer/inner flops */
673 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*141);
676 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_sse4_1_single
677 * Electrostatics interaction: Ewald
678 * VdW interaction: None
679 * Geometry: Water3-Particle
680 * Calculate force/pot: Force
683 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_sse4_1_single
684 (t_nblist * gmx_restrict nlist,
685 rvec * gmx_restrict xx,
686 rvec * gmx_restrict ff,
687 t_forcerec * gmx_restrict fr,
688 t_mdatoms * gmx_restrict mdatoms,
689 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
690 t_nrnb * gmx_restrict nrnb)
692 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
693 * just 0 for non-waters.
694 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
695 * jnr indices corresponding to data put in the four positions in the SIMD register.
697 int i_shift_offset,i_coord_offset,outeriter,inneriter;
698 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
699 int jnrA,jnrB,jnrC,jnrD;
700 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
701 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
702 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
704 real *shiftvec,*fshift,*x,*f;
705 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
707 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
709 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
711 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
713 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
714 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
715 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
716 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
717 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
718 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
719 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
722 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
724 __m128 dummy_mask,cutoff_mask;
725 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
726 __m128 one = _mm_set1_ps(1.0);
727 __m128 two = _mm_set1_ps(2.0);
733 jindex = nlist->jindex;
735 shiftidx = nlist->shift;
737 shiftvec = fr->shift_vec[0];
738 fshift = fr->fshift[0];
739 facel = _mm_set1_ps(fr->epsfac);
740 charge = mdatoms->chargeA;
742 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
743 ewtab = fr->ic->tabq_coul_F;
744 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
745 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
747 /* Setup water-specific parameters */
748 inr = nlist->iinr[0];
749 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
750 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
751 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
753 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
754 rcutoff_scalar = fr->rcoulomb;
755 rcutoff = _mm_set1_ps(rcutoff_scalar);
756 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
758 /* Avoid stupid compiler warnings */
759 jnrA = jnrB = jnrC = jnrD = 0;
768 for(iidx=0;iidx<4*DIM;iidx++)
773 /* Start outer loop over neighborlists */
774 for(iidx=0; iidx<nri; iidx++)
776 /* Load shift vector for this list */
777 i_shift_offset = DIM*shiftidx[iidx];
779 /* Load limits for loop over neighbors */
780 j_index_start = jindex[iidx];
781 j_index_end = jindex[iidx+1];
783 /* Get outer coordinate index */
785 i_coord_offset = DIM*inr;
787 /* Load i particle coords and add shift vector */
788 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
789 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
791 fix0 = _mm_setzero_ps();
792 fiy0 = _mm_setzero_ps();
793 fiz0 = _mm_setzero_ps();
794 fix1 = _mm_setzero_ps();
795 fiy1 = _mm_setzero_ps();
796 fiz1 = _mm_setzero_ps();
797 fix2 = _mm_setzero_ps();
798 fiy2 = _mm_setzero_ps();
799 fiz2 = _mm_setzero_ps();
801 /* Start inner kernel loop */
802 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
805 /* Get j neighbor index, and coordinate index */
810 j_coord_offsetA = DIM*jnrA;
811 j_coord_offsetB = DIM*jnrB;
812 j_coord_offsetC = DIM*jnrC;
813 j_coord_offsetD = DIM*jnrD;
815 /* load j atom coordinates */
816 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
817 x+j_coord_offsetC,x+j_coord_offsetD,
820 /* Calculate displacement vector */
821 dx00 = _mm_sub_ps(ix0,jx0);
822 dy00 = _mm_sub_ps(iy0,jy0);
823 dz00 = _mm_sub_ps(iz0,jz0);
824 dx10 = _mm_sub_ps(ix1,jx0);
825 dy10 = _mm_sub_ps(iy1,jy0);
826 dz10 = _mm_sub_ps(iz1,jz0);
827 dx20 = _mm_sub_ps(ix2,jx0);
828 dy20 = _mm_sub_ps(iy2,jy0);
829 dz20 = _mm_sub_ps(iz2,jz0);
831 /* Calculate squared distance and things based on it */
832 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
833 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
834 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
836 rinv00 = gmx_mm_invsqrt_ps(rsq00);
837 rinv10 = gmx_mm_invsqrt_ps(rsq10);
838 rinv20 = gmx_mm_invsqrt_ps(rsq20);
840 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
841 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
842 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
844 /* Load parameters for j particles */
845 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
846 charge+jnrC+0,charge+jnrD+0);
848 fjx0 = _mm_setzero_ps();
849 fjy0 = _mm_setzero_ps();
850 fjz0 = _mm_setzero_ps();
852 /**************************
853 * CALCULATE INTERACTIONS *
854 **************************/
856 if (gmx_mm_any_lt(rsq00,rcutoff2))
859 r00 = _mm_mul_ps(rsq00,rinv00);
861 /* Compute parameters for interactions between i and j atoms */
862 qq00 = _mm_mul_ps(iq0,jq0);
864 /* EWALD ELECTROSTATICS */
866 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
867 ewrt = _mm_mul_ps(r00,ewtabscale);
868 ewitab = _mm_cvttps_epi32(ewrt);
869 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
870 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
871 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
873 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
874 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
876 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
880 fscal = _mm_and_ps(fscal,cutoff_mask);
882 /* Calculate temporary vectorial force */
883 tx = _mm_mul_ps(fscal,dx00);
884 ty = _mm_mul_ps(fscal,dy00);
885 tz = _mm_mul_ps(fscal,dz00);
887 /* Update vectorial force */
888 fix0 = _mm_add_ps(fix0,tx);
889 fiy0 = _mm_add_ps(fiy0,ty);
890 fiz0 = _mm_add_ps(fiz0,tz);
892 fjx0 = _mm_add_ps(fjx0,tx);
893 fjy0 = _mm_add_ps(fjy0,ty);
894 fjz0 = _mm_add_ps(fjz0,tz);
898 /**************************
899 * CALCULATE INTERACTIONS *
900 **************************/
902 if (gmx_mm_any_lt(rsq10,rcutoff2))
905 r10 = _mm_mul_ps(rsq10,rinv10);
907 /* Compute parameters for interactions between i and j atoms */
908 qq10 = _mm_mul_ps(iq1,jq0);
910 /* EWALD ELECTROSTATICS */
912 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
913 ewrt = _mm_mul_ps(r10,ewtabscale);
914 ewitab = _mm_cvttps_epi32(ewrt);
915 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
916 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
917 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
919 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
920 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
922 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
926 fscal = _mm_and_ps(fscal,cutoff_mask);
928 /* Calculate temporary vectorial force */
929 tx = _mm_mul_ps(fscal,dx10);
930 ty = _mm_mul_ps(fscal,dy10);
931 tz = _mm_mul_ps(fscal,dz10);
933 /* Update vectorial force */
934 fix1 = _mm_add_ps(fix1,tx);
935 fiy1 = _mm_add_ps(fiy1,ty);
936 fiz1 = _mm_add_ps(fiz1,tz);
938 fjx0 = _mm_add_ps(fjx0,tx);
939 fjy0 = _mm_add_ps(fjy0,ty);
940 fjz0 = _mm_add_ps(fjz0,tz);
944 /**************************
945 * CALCULATE INTERACTIONS *
946 **************************/
948 if (gmx_mm_any_lt(rsq20,rcutoff2))
951 r20 = _mm_mul_ps(rsq20,rinv20);
953 /* Compute parameters for interactions between i and j atoms */
954 qq20 = _mm_mul_ps(iq2,jq0);
956 /* EWALD ELECTROSTATICS */
958 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
959 ewrt = _mm_mul_ps(r20,ewtabscale);
960 ewitab = _mm_cvttps_epi32(ewrt);
961 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
962 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
963 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
965 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
966 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
968 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
972 fscal = _mm_and_ps(fscal,cutoff_mask);
974 /* Calculate temporary vectorial force */
975 tx = _mm_mul_ps(fscal,dx20);
976 ty = _mm_mul_ps(fscal,dy20);
977 tz = _mm_mul_ps(fscal,dz20);
979 /* Update vectorial force */
980 fix2 = _mm_add_ps(fix2,tx);
981 fiy2 = _mm_add_ps(fiy2,ty);
982 fiz2 = _mm_add_ps(fiz2,tz);
984 fjx0 = _mm_add_ps(fjx0,tx);
985 fjy0 = _mm_add_ps(fjy0,ty);
986 fjz0 = _mm_add_ps(fjz0,tz);
990 fjptrA = f+j_coord_offsetA;
991 fjptrB = f+j_coord_offsetB;
992 fjptrC = f+j_coord_offsetC;
993 fjptrD = f+j_coord_offsetD;
995 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
997 /* Inner loop uses 117 flops */
1000 if(jidx<j_index_end)
1003 /* Get j neighbor index, and coordinate index */
1004 jnrlistA = jjnr[jidx];
1005 jnrlistB = jjnr[jidx+1];
1006 jnrlistC = jjnr[jidx+2];
1007 jnrlistD = jjnr[jidx+3];
1008 /* Sign of each element will be negative for non-real atoms.
1009 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1010 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1012 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1013 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1014 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1015 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1016 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1017 j_coord_offsetA = DIM*jnrA;
1018 j_coord_offsetB = DIM*jnrB;
1019 j_coord_offsetC = DIM*jnrC;
1020 j_coord_offsetD = DIM*jnrD;
1022 /* load j atom coordinates */
1023 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1024 x+j_coord_offsetC,x+j_coord_offsetD,
1027 /* Calculate displacement vector */
1028 dx00 = _mm_sub_ps(ix0,jx0);
1029 dy00 = _mm_sub_ps(iy0,jy0);
1030 dz00 = _mm_sub_ps(iz0,jz0);
1031 dx10 = _mm_sub_ps(ix1,jx0);
1032 dy10 = _mm_sub_ps(iy1,jy0);
1033 dz10 = _mm_sub_ps(iz1,jz0);
1034 dx20 = _mm_sub_ps(ix2,jx0);
1035 dy20 = _mm_sub_ps(iy2,jy0);
1036 dz20 = _mm_sub_ps(iz2,jz0);
1038 /* Calculate squared distance and things based on it */
1039 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1040 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1041 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1043 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1044 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1045 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1047 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1048 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1049 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1051 /* Load parameters for j particles */
1052 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1053 charge+jnrC+0,charge+jnrD+0);
1055 fjx0 = _mm_setzero_ps();
1056 fjy0 = _mm_setzero_ps();
1057 fjz0 = _mm_setzero_ps();
1059 /**************************
1060 * CALCULATE INTERACTIONS *
1061 **************************/
1063 if (gmx_mm_any_lt(rsq00,rcutoff2))
1066 r00 = _mm_mul_ps(rsq00,rinv00);
1067 r00 = _mm_andnot_ps(dummy_mask,r00);
1069 /* Compute parameters for interactions between i and j atoms */
1070 qq00 = _mm_mul_ps(iq0,jq0);
1072 /* EWALD ELECTROSTATICS */
1074 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1075 ewrt = _mm_mul_ps(r00,ewtabscale);
1076 ewitab = _mm_cvttps_epi32(ewrt);
1077 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1078 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1079 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1081 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1082 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1084 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1088 fscal = _mm_and_ps(fscal,cutoff_mask);
1090 fscal = _mm_andnot_ps(dummy_mask,fscal);
1092 /* Calculate temporary vectorial force */
1093 tx = _mm_mul_ps(fscal,dx00);
1094 ty = _mm_mul_ps(fscal,dy00);
1095 tz = _mm_mul_ps(fscal,dz00);
1097 /* Update vectorial force */
1098 fix0 = _mm_add_ps(fix0,tx);
1099 fiy0 = _mm_add_ps(fiy0,ty);
1100 fiz0 = _mm_add_ps(fiz0,tz);
1102 fjx0 = _mm_add_ps(fjx0,tx);
1103 fjy0 = _mm_add_ps(fjy0,ty);
1104 fjz0 = _mm_add_ps(fjz0,tz);
1108 /**************************
1109 * CALCULATE INTERACTIONS *
1110 **************************/
1112 if (gmx_mm_any_lt(rsq10,rcutoff2))
1115 r10 = _mm_mul_ps(rsq10,rinv10);
1116 r10 = _mm_andnot_ps(dummy_mask,r10);
1118 /* Compute parameters for interactions between i and j atoms */
1119 qq10 = _mm_mul_ps(iq1,jq0);
1121 /* EWALD ELECTROSTATICS */
1123 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1124 ewrt = _mm_mul_ps(r10,ewtabscale);
1125 ewitab = _mm_cvttps_epi32(ewrt);
1126 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1127 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1128 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1130 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1131 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1133 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1137 fscal = _mm_and_ps(fscal,cutoff_mask);
1139 fscal = _mm_andnot_ps(dummy_mask,fscal);
1141 /* Calculate temporary vectorial force */
1142 tx = _mm_mul_ps(fscal,dx10);
1143 ty = _mm_mul_ps(fscal,dy10);
1144 tz = _mm_mul_ps(fscal,dz10);
1146 /* Update vectorial force */
1147 fix1 = _mm_add_ps(fix1,tx);
1148 fiy1 = _mm_add_ps(fiy1,ty);
1149 fiz1 = _mm_add_ps(fiz1,tz);
1151 fjx0 = _mm_add_ps(fjx0,tx);
1152 fjy0 = _mm_add_ps(fjy0,ty);
1153 fjz0 = _mm_add_ps(fjz0,tz);
1157 /**************************
1158 * CALCULATE INTERACTIONS *
1159 **************************/
1161 if (gmx_mm_any_lt(rsq20,rcutoff2))
1164 r20 = _mm_mul_ps(rsq20,rinv20);
1165 r20 = _mm_andnot_ps(dummy_mask,r20);
1167 /* Compute parameters for interactions between i and j atoms */
1168 qq20 = _mm_mul_ps(iq2,jq0);
1170 /* EWALD ELECTROSTATICS */
1172 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1173 ewrt = _mm_mul_ps(r20,ewtabscale);
1174 ewitab = _mm_cvttps_epi32(ewrt);
1175 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1176 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1177 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1179 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1180 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1182 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1186 fscal = _mm_and_ps(fscal,cutoff_mask);
1188 fscal = _mm_andnot_ps(dummy_mask,fscal);
1190 /* Calculate temporary vectorial force */
1191 tx = _mm_mul_ps(fscal,dx20);
1192 ty = _mm_mul_ps(fscal,dy20);
1193 tz = _mm_mul_ps(fscal,dz20);
1195 /* Update vectorial force */
1196 fix2 = _mm_add_ps(fix2,tx);
1197 fiy2 = _mm_add_ps(fiy2,ty);
1198 fiz2 = _mm_add_ps(fiz2,tz);
1200 fjx0 = _mm_add_ps(fjx0,tx);
1201 fjy0 = _mm_add_ps(fjy0,ty);
1202 fjz0 = _mm_add_ps(fjz0,tz);
1206 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1207 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1208 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1209 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1211 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1213 /* Inner loop uses 120 flops */
1216 /* End of innermost loop */
1218 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1219 f+i_coord_offset,fshift+i_shift_offset);
1221 /* Increment number of inner iterations */
1222 inneriter += j_index_end - j_index_start;
1224 /* Outer loop uses 18 flops */
1227 /* Increment number of outer iterations */
1230 /* Update outer/inner flops */
1232 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*120);