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 sse2_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_sse2_single.h"
48 #include "kernelutil_x86_sse2_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_sse2_single
52 * Electrostatics interaction: Ewald
53 * VdW interaction: None
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_sse2_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 SSE, 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 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
86 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
89 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
90 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
92 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
93 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
94 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
97 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
99 __m128 dummy_mask,cutoff_mask;
100 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
101 __m128 one = _mm_set1_ps(1.0);
102 __m128 two = _mm_set1_ps(2.0);
108 jindex = nlist->jindex;
110 shiftidx = nlist->shift;
112 shiftvec = fr->shift_vec[0];
113 fshift = fr->fshift[0];
114 facel = _mm_set1_ps(fr->epsfac);
115 charge = mdatoms->chargeA;
117 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
118 ewtab = fr->ic->tabq_coul_FDV0;
119 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
120 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
122 /* Setup water-specific parameters */
123 inr = nlist->iinr[0];
124 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
125 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
126 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
128 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
129 rcutoff_scalar = fr->rcoulomb;
130 rcutoff = _mm_set1_ps(rcutoff_scalar);
131 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
133 /* Avoid stupid compiler warnings */
134 jnrA = jnrB = jnrC = jnrD = 0;
143 for(iidx=0;iidx<4*DIM;iidx++)
148 /* Start outer loop over neighborlists */
149 for(iidx=0; iidx<nri; iidx++)
151 /* Load shift vector for this list */
152 i_shift_offset = DIM*shiftidx[iidx];
154 /* Load limits for loop over neighbors */
155 j_index_start = jindex[iidx];
156 j_index_end = jindex[iidx+1];
158 /* Get outer coordinate index */
160 i_coord_offset = DIM*inr;
162 /* Load i particle coords and add shift vector */
163 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
164 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
166 fix1 = _mm_setzero_ps();
167 fiy1 = _mm_setzero_ps();
168 fiz1 = _mm_setzero_ps();
169 fix2 = _mm_setzero_ps();
170 fiy2 = _mm_setzero_ps();
171 fiz2 = _mm_setzero_ps();
172 fix3 = _mm_setzero_ps();
173 fiy3 = _mm_setzero_ps();
174 fiz3 = _mm_setzero_ps();
176 /* Reset potential sums */
177 velecsum = _mm_setzero_ps();
179 /* Start inner kernel loop */
180 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
183 /* Get j neighbor index, and coordinate index */
188 j_coord_offsetA = DIM*jnrA;
189 j_coord_offsetB = DIM*jnrB;
190 j_coord_offsetC = DIM*jnrC;
191 j_coord_offsetD = DIM*jnrD;
193 /* load j atom coordinates */
194 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
195 x+j_coord_offsetC,x+j_coord_offsetD,
198 /* Calculate displacement vector */
199 dx10 = _mm_sub_ps(ix1,jx0);
200 dy10 = _mm_sub_ps(iy1,jy0);
201 dz10 = _mm_sub_ps(iz1,jz0);
202 dx20 = _mm_sub_ps(ix2,jx0);
203 dy20 = _mm_sub_ps(iy2,jy0);
204 dz20 = _mm_sub_ps(iz2,jz0);
205 dx30 = _mm_sub_ps(ix3,jx0);
206 dy30 = _mm_sub_ps(iy3,jy0);
207 dz30 = _mm_sub_ps(iz3,jz0);
209 /* Calculate squared distance and things based on it */
210 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
211 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
212 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
214 rinv10 = gmx_mm_invsqrt_ps(rsq10);
215 rinv20 = gmx_mm_invsqrt_ps(rsq20);
216 rinv30 = gmx_mm_invsqrt_ps(rsq30);
218 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
219 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
220 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
222 /* Load parameters for j particles */
223 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
224 charge+jnrC+0,charge+jnrD+0);
226 fjx0 = _mm_setzero_ps();
227 fjy0 = _mm_setzero_ps();
228 fjz0 = _mm_setzero_ps();
230 /**************************
231 * CALCULATE INTERACTIONS *
232 **************************/
234 if (gmx_mm_any_lt(rsq10,rcutoff2))
237 r10 = _mm_mul_ps(rsq10,rinv10);
239 /* Compute parameters for interactions between i and j atoms */
240 qq10 = _mm_mul_ps(iq1,jq0);
242 /* EWALD ELECTROSTATICS */
244 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
245 ewrt = _mm_mul_ps(r10,ewtabscale);
246 ewitab = _mm_cvttps_epi32(ewrt);
247 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
248 ewitab = _mm_slli_epi32(ewitab,2);
249 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
250 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
251 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
252 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
253 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
254 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
255 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
256 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
257 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
259 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
261 /* Update potential sum for this i atom from the interaction with this j atom. */
262 velec = _mm_and_ps(velec,cutoff_mask);
263 velecsum = _mm_add_ps(velecsum,velec);
267 fscal = _mm_and_ps(fscal,cutoff_mask);
269 /* Calculate temporary vectorial force */
270 tx = _mm_mul_ps(fscal,dx10);
271 ty = _mm_mul_ps(fscal,dy10);
272 tz = _mm_mul_ps(fscal,dz10);
274 /* Update vectorial force */
275 fix1 = _mm_add_ps(fix1,tx);
276 fiy1 = _mm_add_ps(fiy1,ty);
277 fiz1 = _mm_add_ps(fiz1,tz);
279 fjx0 = _mm_add_ps(fjx0,tx);
280 fjy0 = _mm_add_ps(fjy0,ty);
281 fjz0 = _mm_add_ps(fjz0,tz);
285 /**************************
286 * CALCULATE INTERACTIONS *
287 **************************/
289 if (gmx_mm_any_lt(rsq20,rcutoff2))
292 r20 = _mm_mul_ps(rsq20,rinv20);
294 /* Compute parameters for interactions between i and j atoms */
295 qq20 = _mm_mul_ps(iq2,jq0);
297 /* EWALD ELECTROSTATICS */
299 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
300 ewrt = _mm_mul_ps(r20,ewtabscale);
301 ewitab = _mm_cvttps_epi32(ewrt);
302 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
303 ewitab = _mm_slli_epi32(ewitab,2);
304 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
305 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
306 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
307 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
308 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
309 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
310 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
311 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
312 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
314 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
316 /* Update potential sum for this i atom from the interaction with this j atom. */
317 velec = _mm_and_ps(velec,cutoff_mask);
318 velecsum = _mm_add_ps(velecsum,velec);
322 fscal = _mm_and_ps(fscal,cutoff_mask);
324 /* Calculate temporary vectorial force */
325 tx = _mm_mul_ps(fscal,dx20);
326 ty = _mm_mul_ps(fscal,dy20);
327 tz = _mm_mul_ps(fscal,dz20);
329 /* Update vectorial force */
330 fix2 = _mm_add_ps(fix2,tx);
331 fiy2 = _mm_add_ps(fiy2,ty);
332 fiz2 = _mm_add_ps(fiz2,tz);
334 fjx0 = _mm_add_ps(fjx0,tx);
335 fjy0 = _mm_add_ps(fjy0,ty);
336 fjz0 = _mm_add_ps(fjz0,tz);
340 /**************************
341 * CALCULATE INTERACTIONS *
342 **************************/
344 if (gmx_mm_any_lt(rsq30,rcutoff2))
347 r30 = _mm_mul_ps(rsq30,rinv30);
349 /* Compute parameters for interactions between i and j atoms */
350 qq30 = _mm_mul_ps(iq3,jq0);
352 /* EWALD ELECTROSTATICS */
354 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
355 ewrt = _mm_mul_ps(r30,ewtabscale);
356 ewitab = _mm_cvttps_epi32(ewrt);
357 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
358 ewitab = _mm_slli_epi32(ewitab,2);
359 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
360 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
361 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
362 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
363 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
364 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
365 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
366 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
367 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
369 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
371 /* Update potential sum for this i atom from the interaction with this j atom. */
372 velec = _mm_and_ps(velec,cutoff_mask);
373 velecsum = _mm_add_ps(velecsum,velec);
377 fscal = _mm_and_ps(fscal,cutoff_mask);
379 /* Calculate temporary vectorial force */
380 tx = _mm_mul_ps(fscal,dx30);
381 ty = _mm_mul_ps(fscal,dy30);
382 tz = _mm_mul_ps(fscal,dz30);
384 /* Update vectorial force */
385 fix3 = _mm_add_ps(fix3,tx);
386 fiy3 = _mm_add_ps(fiy3,ty);
387 fiz3 = _mm_add_ps(fiz3,tz);
389 fjx0 = _mm_add_ps(fjx0,tx);
390 fjy0 = _mm_add_ps(fjy0,ty);
391 fjz0 = _mm_add_ps(fjz0,tz);
395 fjptrA = f+j_coord_offsetA;
396 fjptrB = f+j_coord_offsetB;
397 fjptrC = f+j_coord_offsetC;
398 fjptrD = f+j_coord_offsetD;
400 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
402 /* Inner loop uses 138 flops */
408 /* Get j neighbor index, and coordinate index */
409 jnrlistA = jjnr[jidx];
410 jnrlistB = jjnr[jidx+1];
411 jnrlistC = jjnr[jidx+2];
412 jnrlistD = jjnr[jidx+3];
413 /* Sign of each element will be negative for non-real atoms.
414 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
415 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
417 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
418 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
419 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
420 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
421 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
422 j_coord_offsetA = DIM*jnrA;
423 j_coord_offsetB = DIM*jnrB;
424 j_coord_offsetC = DIM*jnrC;
425 j_coord_offsetD = DIM*jnrD;
427 /* load j atom coordinates */
428 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
429 x+j_coord_offsetC,x+j_coord_offsetD,
432 /* Calculate displacement vector */
433 dx10 = _mm_sub_ps(ix1,jx0);
434 dy10 = _mm_sub_ps(iy1,jy0);
435 dz10 = _mm_sub_ps(iz1,jz0);
436 dx20 = _mm_sub_ps(ix2,jx0);
437 dy20 = _mm_sub_ps(iy2,jy0);
438 dz20 = _mm_sub_ps(iz2,jz0);
439 dx30 = _mm_sub_ps(ix3,jx0);
440 dy30 = _mm_sub_ps(iy3,jy0);
441 dz30 = _mm_sub_ps(iz3,jz0);
443 /* Calculate squared distance and things based on it */
444 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
445 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
446 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
448 rinv10 = gmx_mm_invsqrt_ps(rsq10);
449 rinv20 = gmx_mm_invsqrt_ps(rsq20);
450 rinv30 = gmx_mm_invsqrt_ps(rsq30);
452 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
453 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
454 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
456 /* Load parameters for j particles */
457 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
458 charge+jnrC+0,charge+jnrD+0);
460 fjx0 = _mm_setzero_ps();
461 fjy0 = _mm_setzero_ps();
462 fjz0 = _mm_setzero_ps();
464 /**************************
465 * CALCULATE INTERACTIONS *
466 **************************/
468 if (gmx_mm_any_lt(rsq10,rcutoff2))
471 r10 = _mm_mul_ps(rsq10,rinv10);
472 r10 = _mm_andnot_ps(dummy_mask,r10);
474 /* Compute parameters for interactions between i and j atoms */
475 qq10 = _mm_mul_ps(iq1,jq0);
477 /* EWALD ELECTROSTATICS */
479 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
480 ewrt = _mm_mul_ps(r10,ewtabscale);
481 ewitab = _mm_cvttps_epi32(ewrt);
482 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
483 ewitab = _mm_slli_epi32(ewitab,2);
484 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
485 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
486 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
487 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
488 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
489 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
490 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
491 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
492 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
494 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
496 /* Update potential sum for this i atom from the interaction with this j atom. */
497 velec = _mm_and_ps(velec,cutoff_mask);
498 velec = _mm_andnot_ps(dummy_mask,velec);
499 velecsum = _mm_add_ps(velecsum,velec);
503 fscal = _mm_and_ps(fscal,cutoff_mask);
505 fscal = _mm_andnot_ps(dummy_mask,fscal);
507 /* Calculate temporary vectorial force */
508 tx = _mm_mul_ps(fscal,dx10);
509 ty = _mm_mul_ps(fscal,dy10);
510 tz = _mm_mul_ps(fscal,dz10);
512 /* Update vectorial force */
513 fix1 = _mm_add_ps(fix1,tx);
514 fiy1 = _mm_add_ps(fiy1,ty);
515 fiz1 = _mm_add_ps(fiz1,tz);
517 fjx0 = _mm_add_ps(fjx0,tx);
518 fjy0 = _mm_add_ps(fjy0,ty);
519 fjz0 = _mm_add_ps(fjz0,tz);
523 /**************************
524 * CALCULATE INTERACTIONS *
525 **************************/
527 if (gmx_mm_any_lt(rsq20,rcutoff2))
530 r20 = _mm_mul_ps(rsq20,rinv20);
531 r20 = _mm_andnot_ps(dummy_mask,r20);
533 /* Compute parameters for interactions between i and j atoms */
534 qq20 = _mm_mul_ps(iq2,jq0);
536 /* EWALD ELECTROSTATICS */
538 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
539 ewrt = _mm_mul_ps(r20,ewtabscale);
540 ewitab = _mm_cvttps_epi32(ewrt);
541 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
542 ewitab = _mm_slli_epi32(ewitab,2);
543 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
544 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
545 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
546 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
547 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
548 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
549 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
550 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
551 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
553 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
555 /* Update potential sum for this i atom from the interaction with this j atom. */
556 velec = _mm_and_ps(velec,cutoff_mask);
557 velec = _mm_andnot_ps(dummy_mask,velec);
558 velecsum = _mm_add_ps(velecsum,velec);
562 fscal = _mm_and_ps(fscal,cutoff_mask);
564 fscal = _mm_andnot_ps(dummy_mask,fscal);
566 /* Calculate temporary vectorial force */
567 tx = _mm_mul_ps(fscal,dx20);
568 ty = _mm_mul_ps(fscal,dy20);
569 tz = _mm_mul_ps(fscal,dz20);
571 /* Update vectorial force */
572 fix2 = _mm_add_ps(fix2,tx);
573 fiy2 = _mm_add_ps(fiy2,ty);
574 fiz2 = _mm_add_ps(fiz2,tz);
576 fjx0 = _mm_add_ps(fjx0,tx);
577 fjy0 = _mm_add_ps(fjy0,ty);
578 fjz0 = _mm_add_ps(fjz0,tz);
582 /**************************
583 * CALCULATE INTERACTIONS *
584 **************************/
586 if (gmx_mm_any_lt(rsq30,rcutoff2))
589 r30 = _mm_mul_ps(rsq30,rinv30);
590 r30 = _mm_andnot_ps(dummy_mask,r30);
592 /* Compute parameters for interactions between i and j atoms */
593 qq30 = _mm_mul_ps(iq3,jq0);
595 /* EWALD ELECTROSTATICS */
597 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
598 ewrt = _mm_mul_ps(r30,ewtabscale);
599 ewitab = _mm_cvttps_epi32(ewrt);
600 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
601 ewitab = _mm_slli_epi32(ewitab,2);
602 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
603 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
604 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
605 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
606 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
607 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
608 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
609 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
610 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
612 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
614 /* Update potential sum for this i atom from the interaction with this j atom. */
615 velec = _mm_and_ps(velec,cutoff_mask);
616 velec = _mm_andnot_ps(dummy_mask,velec);
617 velecsum = _mm_add_ps(velecsum,velec);
621 fscal = _mm_and_ps(fscal,cutoff_mask);
623 fscal = _mm_andnot_ps(dummy_mask,fscal);
625 /* Calculate temporary vectorial force */
626 tx = _mm_mul_ps(fscal,dx30);
627 ty = _mm_mul_ps(fscal,dy30);
628 tz = _mm_mul_ps(fscal,dz30);
630 /* Update vectorial force */
631 fix3 = _mm_add_ps(fix3,tx);
632 fiy3 = _mm_add_ps(fiy3,ty);
633 fiz3 = _mm_add_ps(fiz3,tz);
635 fjx0 = _mm_add_ps(fjx0,tx);
636 fjy0 = _mm_add_ps(fjy0,ty);
637 fjz0 = _mm_add_ps(fjz0,tz);
641 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
642 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
643 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
644 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
646 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
648 /* Inner loop uses 141 flops */
651 /* End of innermost loop */
653 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
654 f+i_coord_offset+DIM,fshift+i_shift_offset);
657 /* Update potential energies */
658 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
660 /* Increment number of inner iterations */
661 inneriter += j_index_end - j_index_start;
663 /* Outer loop uses 19 flops */
666 /* Increment number of outer iterations */
669 /* Update outer/inner flops */
671 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*141);
674 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_sse2_single
675 * Electrostatics interaction: Ewald
676 * VdW interaction: None
677 * Geometry: Water4-Particle
678 * Calculate force/pot: Force
681 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_sse2_single
682 (t_nblist * gmx_restrict nlist,
683 rvec * gmx_restrict xx,
684 rvec * gmx_restrict ff,
685 t_forcerec * gmx_restrict fr,
686 t_mdatoms * gmx_restrict mdatoms,
687 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
688 t_nrnb * gmx_restrict nrnb)
690 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
691 * just 0 for non-waters.
692 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
693 * jnr indices corresponding to data put in the four positions in the SIMD register.
695 int i_shift_offset,i_coord_offset,outeriter,inneriter;
696 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
697 int jnrA,jnrB,jnrC,jnrD;
698 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
699 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
700 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
702 real *shiftvec,*fshift,*x,*f;
703 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
705 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
707 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
709 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
711 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
712 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
713 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
714 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
715 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
716 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
717 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
720 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
722 __m128 dummy_mask,cutoff_mask;
723 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
724 __m128 one = _mm_set1_ps(1.0);
725 __m128 two = _mm_set1_ps(2.0);
731 jindex = nlist->jindex;
733 shiftidx = nlist->shift;
735 shiftvec = fr->shift_vec[0];
736 fshift = fr->fshift[0];
737 facel = _mm_set1_ps(fr->epsfac);
738 charge = mdatoms->chargeA;
740 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
741 ewtab = fr->ic->tabq_coul_F;
742 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
743 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
745 /* Setup water-specific parameters */
746 inr = nlist->iinr[0];
747 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
748 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
749 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
751 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
752 rcutoff_scalar = fr->rcoulomb;
753 rcutoff = _mm_set1_ps(rcutoff_scalar);
754 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
756 /* Avoid stupid compiler warnings */
757 jnrA = jnrB = jnrC = jnrD = 0;
766 for(iidx=0;iidx<4*DIM;iidx++)
771 /* Start outer loop over neighborlists */
772 for(iidx=0; iidx<nri; iidx++)
774 /* Load shift vector for this list */
775 i_shift_offset = DIM*shiftidx[iidx];
777 /* Load limits for loop over neighbors */
778 j_index_start = jindex[iidx];
779 j_index_end = jindex[iidx+1];
781 /* Get outer coordinate index */
783 i_coord_offset = DIM*inr;
785 /* Load i particle coords and add shift vector */
786 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
787 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
789 fix1 = _mm_setzero_ps();
790 fiy1 = _mm_setzero_ps();
791 fiz1 = _mm_setzero_ps();
792 fix2 = _mm_setzero_ps();
793 fiy2 = _mm_setzero_ps();
794 fiz2 = _mm_setzero_ps();
795 fix3 = _mm_setzero_ps();
796 fiy3 = _mm_setzero_ps();
797 fiz3 = _mm_setzero_ps();
799 /* Start inner kernel loop */
800 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
803 /* Get j neighbor index, and coordinate index */
808 j_coord_offsetA = DIM*jnrA;
809 j_coord_offsetB = DIM*jnrB;
810 j_coord_offsetC = DIM*jnrC;
811 j_coord_offsetD = DIM*jnrD;
813 /* load j atom coordinates */
814 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
815 x+j_coord_offsetC,x+j_coord_offsetD,
818 /* Calculate displacement vector */
819 dx10 = _mm_sub_ps(ix1,jx0);
820 dy10 = _mm_sub_ps(iy1,jy0);
821 dz10 = _mm_sub_ps(iz1,jz0);
822 dx20 = _mm_sub_ps(ix2,jx0);
823 dy20 = _mm_sub_ps(iy2,jy0);
824 dz20 = _mm_sub_ps(iz2,jz0);
825 dx30 = _mm_sub_ps(ix3,jx0);
826 dy30 = _mm_sub_ps(iy3,jy0);
827 dz30 = _mm_sub_ps(iz3,jz0);
829 /* Calculate squared distance and things based on it */
830 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
831 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
832 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
834 rinv10 = gmx_mm_invsqrt_ps(rsq10);
835 rinv20 = gmx_mm_invsqrt_ps(rsq20);
836 rinv30 = gmx_mm_invsqrt_ps(rsq30);
838 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
839 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
840 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
842 /* Load parameters for j particles */
843 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
844 charge+jnrC+0,charge+jnrD+0);
846 fjx0 = _mm_setzero_ps();
847 fjy0 = _mm_setzero_ps();
848 fjz0 = _mm_setzero_ps();
850 /**************************
851 * CALCULATE INTERACTIONS *
852 **************************/
854 if (gmx_mm_any_lt(rsq10,rcutoff2))
857 r10 = _mm_mul_ps(rsq10,rinv10);
859 /* Compute parameters for interactions between i and j atoms */
860 qq10 = _mm_mul_ps(iq1,jq0);
862 /* EWALD ELECTROSTATICS */
864 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
865 ewrt = _mm_mul_ps(r10,ewtabscale);
866 ewitab = _mm_cvttps_epi32(ewrt);
867 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
868 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
869 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
871 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
872 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
874 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
878 fscal = _mm_and_ps(fscal,cutoff_mask);
880 /* Calculate temporary vectorial force */
881 tx = _mm_mul_ps(fscal,dx10);
882 ty = _mm_mul_ps(fscal,dy10);
883 tz = _mm_mul_ps(fscal,dz10);
885 /* Update vectorial force */
886 fix1 = _mm_add_ps(fix1,tx);
887 fiy1 = _mm_add_ps(fiy1,ty);
888 fiz1 = _mm_add_ps(fiz1,tz);
890 fjx0 = _mm_add_ps(fjx0,tx);
891 fjy0 = _mm_add_ps(fjy0,ty);
892 fjz0 = _mm_add_ps(fjz0,tz);
896 /**************************
897 * CALCULATE INTERACTIONS *
898 **************************/
900 if (gmx_mm_any_lt(rsq20,rcutoff2))
903 r20 = _mm_mul_ps(rsq20,rinv20);
905 /* Compute parameters for interactions between i and j atoms */
906 qq20 = _mm_mul_ps(iq2,jq0);
908 /* EWALD ELECTROSTATICS */
910 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
911 ewrt = _mm_mul_ps(r20,ewtabscale);
912 ewitab = _mm_cvttps_epi32(ewrt);
913 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
914 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
915 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
917 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
918 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
920 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
924 fscal = _mm_and_ps(fscal,cutoff_mask);
926 /* Calculate temporary vectorial force */
927 tx = _mm_mul_ps(fscal,dx20);
928 ty = _mm_mul_ps(fscal,dy20);
929 tz = _mm_mul_ps(fscal,dz20);
931 /* Update vectorial force */
932 fix2 = _mm_add_ps(fix2,tx);
933 fiy2 = _mm_add_ps(fiy2,ty);
934 fiz2 = _mm_add_ps(fiz2,tz);
936 fjx0 = _mm_add_ps(fjx0,tx);
937 fjy0 = _mm_add_ps(fjy0,ty);
938 fjz0 = _mm_add_ps(fjz0,tz);
942 /**************************
943 * CALCULATE INTERACTIONS *
944 **************************/
946 if (gmx_mm_any_lt(rsq30,rcutoff2))
949 r30 = _mm_mul_ps(rsq30,rinv30);
951 /* Compute parameters for interactions between i and j atoms */
952 qq30 = _mm_mul_ps(iq3,jq0);
954 /* EWALD ELECTROSTATICS */
956 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
957 ewrt = _mm_mul_ps(r30,ewtabscale);
958 ewitab = _mm_cvttps_epi32(ewrt);
959 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
960 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
961 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
963 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
964 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
966 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
970 fscal = _mm_and_ps(fscal,cutoff_mask);
972 /* Calculate temporary vectorial force */
973 tx = _mm_mul_ps(fscal,dx30);
974 ty = _mm_mul_ps(fscal,dy30);
975 tz = _mm_mul_ps(fscal,dz30);
977 /* Update vectorial force */
978 fix3 = _mm_add_ps(fix3,tx);
979 fiy3 = _mm_add_ps(fiy3,ty);
980 fiz3 = _mm_add_ps(fiz3,tz);
982 fjx0 = _mm_add_ps(fjx0,tx);
983 fjy0 = _mm_add_ps(fjy0,ty);
984 fjz0 = _mm_add_ps(fjz0,tz);
988 fjptrA = f+j_coord_offsetA;
989 fjptrB = f+j_coord_offsetB;
990 fjptrC = f+j_coord_offsetC;
991 fjptrD = f+j_coord_offsetD;
993 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
995 /* Inner loop uses 117 flops */
1001 /* Get j neighbor index, and coordinate index */
1002 jnrlistA = jjnr[jidx];
1003 jnrlistB = jjnr[jidx+1];
1004 jnrlistC = jjnr[jidx+2];
1005 jnrlistD = jjnr[jidx+3];
1006 /* Sign of each element will be negative for non-real atoms.
1007 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1008 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1010 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1011 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1012 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1013 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1014 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1015 j_coord_offsetA = DIM*jnrA;
1016 j_coord_offsetB = DIM*jnrB;
1017 j_coord_offsetC = DIM*jnrC;
1018 j_coord_offsetD = DIM*jnrD;
1020 /* load j atom coordinates */
1021 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1022 x+j_coord_offsetC,x+j_coord_offsetD,
1025 /* Calculate displacement vector */
1026 dx10 = _mm_sub_ps(ix1,jx0);
1027 dy10 = _mm_sub_ps(iy1,jy0);
1028 dz10 = _mm_sub_ps(iz1,jz0);
1029 dx20 = _mm_sub_ps(ix2,jx0);
1030 dy20 = _mm_sub_ps(iy2,jy0);
1031 dz20 = _mm_sub_ps(iz2,jz0);
1032 dx30 = _mm_sub_ps(ix3,jx0);
1033 dy30 = _mm_sub_ps(iy3,jy0);
1034 dz30 = _mm_sub_ps(iz3,jz0);
1036 /* Calculate squared distance and things based on it */
1037 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1038 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1039 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1041 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1042 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1043 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1045 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1046 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1047 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1049 /* Load parameters for j particles */
1050 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1051 charge+jnrC+0,charge+jnrD+0);
1053 fjx0 = _mm_setzero_ps();
1054 fjy0 = _mm_setzero_ps();
1055 fjz0 = _mm_setzero_ps();
1057 /**************************
1058 * CALCULATE INTERACTIONS *
1059 **************************/
1061 if (gmx_mm_any_lt(rsq10,rcutoff2))
1064 r10 = _mm_mul_ps(rsq10,rinv10);
1065 r10 = _mm_andnot_ps(dummy_mask,r10);
1067 /* Compute parameters for interactions between i and j atoms */
1068 qq10 = _mm_mul_ps(iq1,jq0);
1070 /* EWALD ELECTROSTATICS */
1072 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1073 ewrt = _mm_mul_ps(r10,ewtabscale);
1074 ewitab = _mm_cvttps_epi32(ewrt);
1075 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1076 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1077 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1079 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1080 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1082 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1086 fscal = _mm_and_ps(fscal,cutoff_mask);
1088 fscal = _mm_andnot_ps(dummy_mask,fscal);
1090 /* Calculate temporary vectorial force */
1091 tx = _mm_mul_ps(fscal,dx10);
1092 ty = _mm_mul_ps(fscal,dy10);
1093 tz = _mm_mul_ps(fscal,dz10);
1095 /* Update vectorial force */
1096 fix1 = _mm_add_ps(fix1,tx);
1097 fiy1 = _mm_add_ps(fiy1,ty);
1098 fiz1 = _mm_add_ps(fiz1,tz);
1100 fjx0 = _mm_add_ps(fjx0,tx);
1101 fjy0 = _mm_add_ps(fjy0,ty);
1102 fjz0 = _mm_add_ps(fjz0,tz);
1106 /**************************
1107 * CALCULATE INTERACTIONS *
1108 **************************/
1110 if (gmx_mm_any_lt(rsq20,rcutoff2))
1113 r20 = _mm_mul_ps(rsq20,rinv20);
1114 r20 = _mm_andnot_ps(dummy_mask,r20);
1116 /* Compute parameters for interactions between i and j atoms */
1117 qq20 = _mm_mul_ps(iq2,jq0);
1119 /* EWALD ELECTROSTATICS */
1121 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1122 ewrt = _mm_mul_ps(r20,ewtabscale);
1123 ewitab = _mm_cvttps_epi32(ewrt);
1124 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1125 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1126 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1128 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1129 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1131 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1135 fscal = _mm_and_ps(fscal,cutoff_mask);
1137 fscal = _mm_andnot_ps(dummy_mask,fscal);
1139 /* Calculate temporary vectorial force */
1140 tx = _mm_mul_ps(fscal,dx20);
1141 ty = _mm_mul_ps(fscal,dy20);
1142 tz = _mm_mul_ps(fscal,dz20);
1144 /* Update vectorial force */
1145 fix2 = _mm_add_ps(fix2,tx);
1146 fiy2 = _mm_add_ps(fiy2,ty);
1147 fiz2 = _mm_add_ps(fiz2,tz);
1149 fjx0 = _mm_add_ps(fjx0,tx);
1150 fjy0 = _mm_add_ps(fjy0,ty);
1151 fjz0 = _mm_add_ps(fjz0,tz);
1155 /**************************
1156 * CALCULATE INTERACTIONS *
1157 **************************/
1159 if (gmx_mm_any_lt(rsq30,rcutoff2))
1162 r30 = _mm_mul_ps(rsq30,rinv30);
1163 r30 = _mm_andnot_ps(dummy_mask,r30);
1165 /* Compute parameters for interactions between i and j atoms */
1166 qq30 = _mm_mul_ps(iq3,jq0);
1168 /* EWALD ELECTROSTATICS */
1170 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1171 ewrt = _mm_mul_ps(r30,ewtabscale);
1172 ewitab = _mm_cvttps_epi32(ewrt);
1173 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1174 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1175 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1177 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1178 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1180 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1184 fscal = _mm_and_ps(fscal,cutoff_mask);
1186 fscal = _mm_andnot_ps(dummy_mask,fscal);
1188 /* Calculate temporary vectorial force */
1189 tx = _mm_mul_ps(fscal,dx30);
1190 ty = _mm_mul_ps(fscal,dy30);
1191 tz = _mm_mul_ps(fscal,dz30);
1193 /* Update vectorial force */
1194 fix3 = _mm_add_ps(fix3,tx);
1195 fiy3 = _mm_add_ps(fiy3,ty);
1196 fiz3 = _mm_add_ps(fiz3,tz);
1198 fjx0 = _mm_add_ps(fjx0,tx);
1199 fjy0 = _mm_add_ps(fjy0,ty);
1200 fjz0 = _mm_add_ps(fjz0,tz);
1204 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1205 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1206 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1207 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1209 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1211 /* Inner loop uses 120 flops */
1214 /* End of innermost loop */
1216 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1217 f+i_coord_offset+DIM,fshift+i_shift_offset);
1219 /* Increment number of inner iterations */
1220 inneriter += j_index_end - j_index_start;
1222 /* Outer loop uses 18 flops */
1225 /* Increment number of outer iterations */
1228 /* Update outer/inner flops */
1230 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);