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 "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse2_single.h"
48 #include "kernelutil_x86_sse2_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEw_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_ElecEw_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 /* Avoid stupid compiler warnings */
129 jnrA = jnrB = jnrC = jnrD = 0;
138 for(iidx=0;iidx<4*DIM;iidx++)
143 /* Start outer loop over neighborlists */
144 for(iidx=0; iidx<nri; iidx++)
146 /* Load shift vector for this list */
147 i_shift_offset = DIM*shiftidx[iidx];
149 /* Load limits for loop over neighbors */
150 j_index_start = jindex[iidx];
151 j_index_end = jindex[iidx+1];
153 /* Get outer coordinate index */
155 i_coord_offset = DIM*inr;
157 /* Load i particle coords and add shift vector */
158 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
159 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
161 fix1 = _mm_setzero_ps();
162 fiy1 = _mm_setzero_ps();
163 fiz1 = _mm_setzero_ps();
164 fix2 = _mm_setzero_ps();
165 fiy2 = _mm_setzero_ps();
166 fiz2 = _mm_setzero_ps();
167 fix3 = _mm_setzero_ps();
168 fiy3 = _mm_setzero_ps();
169 fiz3 = _mm_setzero_ps();
171 /* Reset potential sums */
172 velecsum = _mm_setzero_ps();
174 /* Start inner kernel loop */
175 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
178 /* Get j neighbor index, and coordinate index */
183 j_coord_offsetA = DIM*jnrA;
184 j_coord_offsetB = DIM*jnrB;
185 j_coord_offsetC = DIM*jnrC;
186 j_coord_offsetD = DIM*jnrD;
188 /* load j atom coordinates */
189 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
190 x+j_coord_offsetC,x+j_coord_offsetD,
193 /* Calculate displacement vector */
194 dx10 = _mm_sub_ps(ix1,jx0);
195 dy10 = _mm_sub_ps(iy1,jy0);
196 dz10 = _mm_sub_ps(iz1,jz0);
197 dx20 = _mm_sub_ps(ix2,jx0);
198 dy20 = _mm_sub_ps(iy2,jy0);
199 dz20 = _mm_sub_ps(iz2,jz0);
200 dx30 = _mm_sub_ps(ix3,jx0);
201 dy30 = _mm_sub_ps(iy3,jy0);
202 dz30 = _mm_sub_ps(iz3,jz0);
204 /* Calculate squared distance and things based on it */
205 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
206 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
207 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
209 rinv10 = gmx_mm_invsqrt_ps(rsq10);
210 rinv20 = gmx_mm_invsqrt_ps(rsq20);
211 rinv30 = gmx_mm_invsqrt_ps(rsq30);
213 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
214 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
215 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
217 /* Load parameters for j particles */
218 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
219 charge+jnrC+0,charge+jnrD+0);
221 fjx0 = _mm_setzero_ps();
222 fjy0 = _mm_setzero_ps();
223 fjz0 = _mm_setzero_ps();
225 /**************************
226 * CALCULATE INTERACTIONS *
227 **************************/
229 r10 = _mm_mul_ps(rsq10,rinv10);
231 /* Compute parameters for interactions between i and j atoms */
232 qq10 = _mm_mul_ps(iq1,jq0);
234 /* EWALD ELECTROSTATICS */
236 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
237 ewrt = _mm_mul_ps(r10,ewtabscale);
238 ewitab = _mm_cvttps_epi32(ewrt);
239 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
240 ewitab = _mm_slli_epi32(ewitab,2);
241 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
242 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
243 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
244 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
245 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
246 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
247 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
248 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
249 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
251 /* Update potential sum for this i atom from the interaction with this j atom. */
252 velecsum = _mm_add_ps(velecsum,velec);
256 /* Calculate temporary vectorial force */
257 tx = _mm_mul_ps(fscal,dx10);
258 ty = _mm_mul_ps(fscal,dy10);
259 tz = _mm_mul_ps(fscal,dz10);
261 /* Update vectorial force */
262 fix1 = _mm_add_ps(fix1,tx);
263 fiy1 = _mm_add_ps(fiy1,ty);
264 fiz1 = _mm_add_ps(fiz1,tz);
266 fjx0 = _mm_add_ps(fjx0,tx);
267 fjy0 = _mm_add_ps(fjy0,ty);
268 fjz0 = _mm_add_ps(fjz0,tz);
270 /**************************
271 * CALCULATE INTERACTIONS *
272 **************************/
274 r20 = _mm_mul_ps(rsq20,rinv20);
276 /* Compute parameters for interactions between i and j atoms */
277 qq20 = _mm_mul_ps(iq2,jq0);
279 /* EWALD ELECTROSTATICS */
281 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
282 ewrt = _mm_mul_ps(r20,ewtabscale);
283 ewitab = _mm_cvttps_epi32(ewrt);
284 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
285 ewitab = _mm_slli_epi32(ewitab,2);
286 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
287 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
288 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
289 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
290 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
291 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
292 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
293 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
294 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
296 /* Update potential sum for this i atom from the interaction with this j atom. */
297 velecsum = _mm_add_ps(velecsum,velec);
301 /* Calculate temporary vectorial force */
302 tx = _mm_mul_ps(fscal,dx20);
303 ty = _mm_mul_ps(fscal,dy20);
304 tz = _mm_mul_ps(fscal,dz20);
306 /* Update vectorial force */
307 fix2 = _mm_add_ps(fix2,tx);
308 fiy2 = _mm_add_ps(fiy2,ty);
309 fiz2 = _mm_add_ps(fiz2,tz);
311 fjx0 = _mm_add_ps(fjx0,tx);
312 fjy0 = _mm_add_ps(fjy0,ty);
313 fjz0 = _mm_add_ps(fjz0,tz);
315 /**************************
316 * CALCULATE INTERACTIONS *
317 **************************/
319 r30 = _mm_mul_ps(rsq30,rinv30);
321 /* Compute parameters for interactions between i and j atoms */
322 qq30 = _mm_mul_ps(iq3,jq0);
324 /* EWALD ELECTROSTATICS */
326 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
327 ewrt = _mm_mul_ps(r30,ewtabscale);
328 ewitab = _mm_cvttps_epi32(ewrt);
329 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
330 ewitab = _mm_slli_epi32(ewitab,2);
331 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
332 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
333 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
334 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
335 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
336 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
337 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
338 velec = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
339 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
341 /* Update potential sum for this i atom from the interaction with this j atom. */
342 velecsum = _mm_add_ps(velecsum,velec);
346 /* Calculate temporary vectorial force */
347 tx = _mm_mul_ps(fscal,dx30);
348 ty = _mm_mul_ps(fscal,dy30);
349 tz = _mm_mul_ps(fscal,dz30);
351 /* Update vectorial force */
352 fix3 = _mm_add_ps(fix3,tx);
353 fiy3 = _mm_add_ps(fiy3,ty);
354 fiz3 = _mm_add_ps(fiz3,tz);
356 fjx0 = _mm_add_ps(fjx0,tx);
357 fjy0 = _mm_add_ps(fjy0,ty);
358 fjz0 = _mm_add_ps(fjz0,tz);
360 fjptrA = f+j_coord_offsetA;
361 fjptrB = f+j_coord_offsetB;
362 fjptrC = f+j_coord_offsetC;
363 fjptrD = f+j_coord_offsetD;
365 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
367 /* Inner loop uses 123 flops */
373 /* Get j neighbor index, and coordinate index */
374 jnrlistA = jjnr[jidx];
375 jnrlistB = jjnr[jidx+1];
376 jnrlistC = jjnr[jidx+2];
377 jnrlistD = jjnr[jidx+3];
378 /* Sign of each element will be negative for non-real atoms.
379 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
380 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
382 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
383 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
384 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
385 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
386 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
387 j_coord_offsetA = DIM*jnrA;
388 j_coord_offsetB = DIM*jnrB;
389 j_coord_offsetC = DIM*jnrC;
390 j_coord_offsetD = DIM*jnrD;
392 /* load j atom coordinates */
393 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
394 x+j_coord_offsetC,x+j_coord_offsetD,
397 /* Calculate displacement vector */
398 dx10 = _mm_sub_ps(ix1,jx0);
399 dy10 = _mm_sub_ps(iy1,jy0);
400 dz10 = _mm_sub_ps(iz1,jz0);
401 dx20 = _mm_sub_ps(ix2,jx0);
402 dy20 = _mm_sub_ps(iy2,jy0);
403 dz20 = _mm_sub_ps(iz2,jz0);
404 dx30 = _mm_sub_ps(ix3,jx0);
405 dy30 = _mm_sub_ps(iy3,jy0);
406 dz30 = _mm_sub_ps(iz3,jz0);
408 /* Calculate squared distance and things based on it */
409 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
410 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
411 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
413 rinv10 = gmx_mm_invsqrt_ps(rsq10);
414 rinv20 = gmx_mm_invsqrt_ps(rsq20);
415 rinv30 = gmx_mm_invsqrt_ps(rsq30);
417 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
418 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
419 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
421 /* Load parameters for j particles */
422 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
423 charge+jnrC+0,charge+jnrD+0);
425 fjx0 = _mm_setzero_ps();
426 fjy0 = _mm_setzero_ps();
427 fjz0 = _mm_setzero_ps();
429 /**************************
430 * CALCULATE INTERACTIONS *
431 **************************/
433 r10 = _mm_mul_ps(rsq10,rinv10);
434 r10 = _mm_andnot_ps(dummy_mask,r10);
436 /* Compute parameters for interactions between i and j atoms */
437 qq10 = _mm_mul_ps(iq1,jq0);
439 /* EWALD ELECTROSTATICS */
441 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
442 ewrt = _mm_mul_ps(r10,ewtabscale);
443 ewitab = _mm_cvttps_epi32(ewrt);
444 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
445 ewitab = _mm_slli_epi32(ewitab,2);
446 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
447 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
448 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
449 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
450 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
451 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
452 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
453 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
454 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
456 /* Update potential sum for this i atom from the interaction with this j atom. */
457 velec = _mm_andnot_ps(dummy_mask,velec);
458 velecsum = _mm_add_ps(velecsum,velec);
462 fscal = _mm_andnot_ps(dummy_mask,fscal);
464 /* Calculate temporary vectorial force */
465 tx = _mm_mul_ps(fscal,dx10);
466 ty = _mm_mul_ps(fscal,dy10);
467 tz = _mm_mul_ps(fscal,dz10);
469 /* Update vectorial force */
470 fix1 = _mm_add_ps(fix1,tx);
471 fiy1 = _mm_add_ps(fiy1,ty);
472 fiz1 = _mm_add_ps(fiz1,tz);
474 fjx0 = _mm_add_ps(fjx0,tx);
475 fjy0 = _mm_add_ps(fjy0,ty);
476 fjz0 = _mm_add_ps(fjz0,tz);
478 /**************************
479 * CALCULATE INTERACTIONS *
480 **************************/
482 r20 = _mm_mul_ps(rsq20,rinv20);
483 r20 = _mm_andnot_ps(dummy_mask,r20);
485 /* Compute parameters for interactions between i and j atoms */
486 qq20 = _mm_mul_ps(iq2,jq0);
488 /* EWALD ELECTROSTATICS */
490 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
491 ewrt = _mm_mul_ps(r20,ewtabscale);
492 ewitab = _mm_cvttps_epi32(ewrt);
493 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
494 ewitab = _mm_slli_epi32(ewitab,2);
495 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
496 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
497 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
498 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
499 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
500 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
501 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
502 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
503 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
505 /* Update potential sum for this i atom from the interaction with this j atom. */
506 velec = _mm_andnot_ps(dummy_mask,velec);
507 velecsum = _mm_add_ps(velecsum,velec);
511 fscal = _mm_andnot_ps(dummy_mask,fscal);
513 /* Calculate temporary vectorial force */
514 tx = _mm_mul_ps(fscal,dx20);
515 ty = _mm_mul_ps(fscal,dy20);
516 tz = _mm_mul_ps(fscal,dz20);
518 /* Update vectorial force */
519 fix2 = _mm_add_ps(fix2,tx);
520 fiy2 = _mm_add_ps(fiy2,ty);
521 fiz2 = _mm_add_ps(fiz2,tz);
523 fjx0 = _mm_add_ps(fjx0,tx);
524 fjy0 = _mm_add_ps(fjy0,ty);
525 fjz0 = _mm_add_ps(fjz0,tz);
527 /**************************
528 * CALCULATE INTERACTIONS *
529 **************************/
531 r30 = _mm_mul_ps(rsq30,rinv30);
532 r30 = _mm_andnot_ps(dummy_mask,r30);
534 /* Compute parameters for interactions between i and j atoms */
535 qq30 = _mm_mul_ps(iq3,jq0);
537 /* EWALD ELECTROSTATICS */
539 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
540 ewrt = _mm_mul_ps(r30,ewtabscale);
541 ewitab = _mm_cvttps_epi32(ewrt);
542 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
543 ewitab = _mm_slli_epi32(ewitab,2);
544 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
545 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
546 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
547 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
548 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
549 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
550 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
551 velec = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
552 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
554 /* Update potential sum for this i atom from the interaction with this j atom. */
555 velec = _mm_andnot_ps(dummy_mask,velec);
556 velecsum = _mm_add_ps(velecsum,velec);
560 fscal = _mm_andnot_ps(dummy_mask,fscal);
562 /* Calculate temporary vectorial force */
563 tx = _mm_mul_ps(fscal,dx30);
564 ty = _mm_mul_ps(fscal,dy30);
565 tz = _mm_mul_ps(fscal,dz30);
567 /* Update vectorial force */
568 fix3 = _mm_add_ps(fix3,tx);
569 fiy3 = _mm_add_ps(fiy3,ty);
570 fiz3 = _mm_add_ps(fiz3,tz);
572 fjx0 = _mm_add_ps(fjx0,tx);
573 fjy0 = _mm_add_ps(fjy0,ty);
574 fjz0 = _mm_add_ps(fjz0,tz);
576 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
577 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
578 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
579 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
581 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
583 /* Inner loop uses 126 flops */
586 /* End of innermost loop */
588 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
589 f+i_coord_offset+DIM,fshift+i_shift_offset);
592 /* Update potential energies */
593 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
595 /* Increment number of inner iterations */
596 inneriter += j_index_end - j_index_start;
598 /* Outer loop uses 19 flops */
601 /* Increment number of outer iterations */
604 /* Update outer/inner flops */
606 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*126);
609 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse2_single
610 * Electrostatics interaction: Ewald
611 * VdW interaction: None
612 * Geometry: Water4-Particle
613 * Calculate force/pot: Force
616 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse2_single
617 (t_nblist * gmx_restrict nlist,
618 rvec * gmx_restrict xx,
619 rvec * gmx_restrict ff,
620 t_forcerec * gmx_restrict fr,
621 t_mdatoms * gmx_restrict mdatoms,
622 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
623 t_nrnb * gmx_restrict nrnb)
625 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
626 * just 0 for non-waters.
627 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
628 * jnr indices corresponding to data put in the four positions in the SIMD register.
630 int i_shift_offset,i_coord_offset,outeriter,inneriter;
631 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
632 int jnrA,jnrB,jnrC,jnrD;
633 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
634 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
635 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
637 real *shiftvec,*fshift,*x,*f;
638 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
640 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
642 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
644 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
646 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
647 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
648 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
649 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
650 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
651 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
652 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
655 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
657 __m128 dummy_mask,cutoff_mask;
658 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
659 __m128 one = _mm_set1_ps(1.0);
660 __m128 two = _mm_set1_ps(2.0);
666 jindex = nlist->jindex;
668 shiftidx = nlist->shift;
670 shiftvec = fr->shift_vec[0];
671 fshift = fr->fshift[0];
672 facel = _mm_set1_ps(fr->epsfac);
673 charge = mdatoms->chargeA;
675 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
676 ewtab = fr->ic->tabq_coul_F;
677 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
678 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
680 /* Setup water-specific parameters */
681 inr = nlist->iinr[0];
682 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
683 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
684 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
686 /* Avoid stupid compiler warnings */
687 jnrA = jnrB = jnrC = jnrD = 0;
696 for(iidx=0;iidx<4*DIM;iidx++)
701 /* Start outer loop over neighborlists */
702 for(iidx=0; iidx<nri; iidx++)
704 /* Load shift vector for this list */
705 i_shift_offset = DIM*shiftidx[iidx];
707 /* Load limits for loop over neighbors */
708 j_index_start = jindex[iidx];
709 j_index_end = jindex[iidx+1];
711 /* Get outer coordinate index */
713 i_coord_offset = DIM*inr;
715 /* Load i particle coords and add shift vector */
716 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
717 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
719 fix1 = _mm_setzero_ps();
720 fiy1 = _mm_setzero_ps();
721 fiz1 = _mm_setzero_ps();
722 fix2 = _mm_setzero_ps();
723 fiy2 = _mm_setzero_ps();
724 fiz2 = _mm_setzero_ps();
725 fix3 = _mm_setzero_ps();
726 fiy3 = _mm_setzero_ps();
727 fiz3 = _mm_setzero_ps();
729 /* Start inner kernel loop */
730 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
733 /* Get j neighbor index, and coordinate index */
738 j_coord_offsetA = DIM*jnrA;
739 j_coord_offsetB = DIM*jnrB;
740 j_coord_offsetC = DIM*jnrC;
741 j_coord_offsetD = DIM*jnrD;
743 /* load j atom coordinates */
744 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
745 x+j_coord_offsetC,x+j_coord_offsetD,
748 /* Calculate displacement vector */
749 dx10 = _mm_sub_ps(ix1,jx0);
750 dy10 = _mm_sub_ps(iy1,jy0);
751 dz10 = _mm_sub_ps(iz1,jz0);
752 dx20 = _mm_sub_ps(ix2,jx0);
753 dy20 = _mm_sub_ps(iy2,jy0);
754 dz20 = _mm_sub_ps(iz2,jz0);
755 dx30 = _mm_sub_ps(ix3,jx0);
756 dy30 = _mm_sub_ps(iy3,jy0);
757 dz30 = _mm_sub_ps(iz3,jz0);
759 /* Calculate squared distance and things based on it */
760 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
761 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
762 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
764 rinv10 = gmx_mm_invsqrt_ps(rsq10);
765 rinv20 = gmx_mm_invsqrt_ps(rsq20);
766 rinv30 = gmx_mm_invsqrt_ps(rsq30);
768 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
769 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
770 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
772 /* Load parameters for j particles */
773 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
774 charge+jnrC+0,charge+jnrD+0);
776 fjx0 = _mm_setzero_ps();
777 fjy0 = _mm_setzero_ps();
778 fjz0 = _mm_setzero_ps();
780 /**************************
781 * CALCULATE INTERACTIONS *
782 **************************/
784 r10 = _mm_mul_ps(rsq10,rinv10);
786 /* Compute parameters for interactions between i and j atoms */
787 qq10 = _mm_mul_ps(iq1,jq0);
789 /* EWALD ELECTROSTATICS */
791 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
792 ewrt = _mm_mul_ps(r10,ewtabscale);
793 ewitab = _mm_cvttps_epi32(ewrt);
794 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
795 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
796 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
798 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
799 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
803 /* Calculate temporary vectorial force */
804 tx = _mm_mul_ps(fscal,dx10);
805 ty = _mm_mul_ps(fscal,dy10);
806 tz = _mm_mul_ps(fscal,dz10);
808 /* Update vectorial force */
809 fix1 = _mm_add_ps(fix1,tx);
810 fiy1 = _mm_add_ps(fiy1,ty);
811 fiz1 = _mm_add_ps(fiz1,tz);
813 fjx0 = _mm_add_ps(fjx0,tx);
814 fjy0 = _mm_add_ps(fjy0,ty);
815 fjz0 = _mm_add_ps(fjz0,tz);
817 /**************************
818 * CALCULATE INTERACTIONS *
819 **************************/
821 r20 = _mm_mul_ps(rsq20,rinv20);
823 /* Compute parameters for interactions between i and j atoms */
824 qq20 = _mm_mul_ps(iq2,jq0);
826 /* EWALD ELECTROSTATICS */
828 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
829 ewrt = _mm_mul_ps(r20,ewtabscale);
830 ewitab = _mm_cvttps_epi32(ewrt);
831 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
832 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
833 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
835 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
836 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
840 /* Calculate temporary vectorial force */
841 tx = _mm_mul_ps(fscal,dx20);
842 ty = _mm_mul_ps(fscal,dy20);
843 tz = _mm_mul_ps(fscal,dz20);
845 /* Update vectorial force */
846 fix2 = _mm_add_ps(fix2,tx);
847 fiy2 = _mm_add_ps(fiy2,ty);
848 fiz2 = _mm_add_ps(fiz2,tz);
850 fjx0 = _mm_add_ps(fjx0,tx);
851 fjy0 = _mm_add_ps(fjy0,ty);
852 fjz0 = _mm_add_ps(fjz0,tz);
854 /**************************
855 * CALCULATE INTERACTIONS *
856 **************************/
858 r30 = _mm_mul_ps(rsq30,rinv30);
860 /* Compute parameters for interactions between i and j atoms */
861 qq30 = _mm_mul_ps(iq3,jq0);
863 /* EWALD ELECTROSTATICS */
865 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
866 ewrt = _mm_mul_ps(r30,ewtabscale);
867 ewitab = _mm_cvttps_epi32(ewrt);
868 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
869 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
870 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
872 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
873 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
877 /* Calculate temporary vectorial force */
878 tx = _mm_mul_ps(fscal,dx30);
879 ty = _mm_mul_ps(fscal,dy30);
880 tz = _mm_mul_ps(fscal,dz30);
882 /* Update vectorial force */
883 fix3 = _mm_add_ps(fix3,tx);
884 fiy3 = _mm_add_ps(fiy3,ty);
885 fiz3 = _mm_add_ps(fiz3,tz);
887 fjx0 = _mm_add_ps(fjx0,tx);
888 fjy0 = _mm_add_ps(fjy0,ty);
889 fjz0 = _mm_add_ps(fjz0,tz);
891 fjptrA = f+j_coord_offsetA;
892 fjptrB = f+j_coord_offsetB;
893 fjptrC = f+j_coord_offsetC;
894 fjptrD = f+j_coord_offsetD;
896 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
898 /* Inner loop uses 108 flops */
904 /* Get j neighbor index, and coordinate index */
905 jnrlistA = jjnr[jidx];
906 jnrlistB = jjnr[jidx+1];
907 jnrlistC = jjnr[jidx+2];
908 jnrlistD = jjnr[jidx+3];
909 /* Sign of each element will be negative for non-real atoms.
910 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
911 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
913 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
914 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
915 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
916 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
917 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
918 j_coord_offsetA = DIM*jnrA;
919 j_coord_offsetB = DIM*jnrB;
920 j_coord_offsetC = DIM*jnrC;
921 j_coord_offsetD = DIM*jnrD;
923 /* load j atom coordinates */
924 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
925 x+j_coord_offsetC,x+j_coord_offsetD,
928 /* Calculate displacement vector */
929 dx10 = _mm_sub_ps(ix1,jx0);
930 dy10 = _mm_sub_ps(iy1,jy0);
931 dz10 = _mm_sub_ps(iz1,jz0);
932 dx20 = _mm_sub_ps(ix2,jx0);
933 dy20 = _mm_sub_ps(iy2,jy0);
934 dz20 = _mm_sub_ps(iz2,jz0);
935 dx30 = _mm_sub_ps(ix3,jx0);
936 dy30 = _mm_sub_ps(iy3,jy0);
937 dz30 = _mm_sub_ps(iz3,jz0);
939 /* Calculate squared distance and things based on it */
940 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
941 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
942 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
944 rinv10 = gmx_mm_invsqrt_ps(rsq10);
945 rinv20 = gmx_mm_invsqrt_ps(rsq20);
946 rinv30 = gmx_mm_invsqrt_ps(rsq30);
948 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
949 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
950 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
952 /* Load parameters for j particles */
953 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
954 charge+jnrC+0,charge+jnrD+0);
956 fjx0 = _mm_setzero_ps();
957 fjy0 = _mm_setzero_ps();
958 fjz0 = _mm_setzero_ps();
960 /**************************
961 * CALCULATE INTERACTIONS *
962 **************************/
964 r10 = _mm_mul_ps(rsq10,rinv10);
965 r10 = _mm_andnot_ps(dummy_mask,r10);
967 /* Compute parameters for interactions between i and j atoms */
968 qq10 = _mm_mul_ps(iq1,jq0);
970 /* EWALD ELECTROSTATICS */
972 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
973 ewrt = _mm_mul_ps(r10,ewtabscale);
974 ewitab = _mm_cvttps_epi32(ewrt);
975 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
976 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
977 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
979 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
980 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
984 fscal = _mm_andnot_ps(dummy_mask,fscal);
986 /* Calculate temporary vectorial force */
987 tx = _mm_mul_ps(fscal,dx10);
988 ty = _mm_mul_ps(fscal,dy10);
989 tz = _mm_mul_ps(fscal,dz10);
991 /* Update vectorial force */
992 fix1 = _mm_add_ps(fix1,tx);
993 fiy1 = _mm_add_ps(fiy1,ty);
994 fiz1 = _mm_add_ps(fiz1,tz);
996 fjx0 = _mm_add_ps(fjx0,tx);
997 fjy0 = _mm_add_ps(fjy0,ty);
998 fjz0 = _mm_add_ps(fjz0,tz);
1000 /**************************
1001 * CALCULATE INTERACTIONS *
1002 **************************/
1004 r20 = _mm_mul_ps(rsq20,rinv20);
1005 r20 = _mm_andnot_ps(dummy_mask,r20);
1007 /* Compute parameters for interactions between i and j atoms */
1008 qq20 = _mm_mul_ps(iq2,jq0);
1010 /* EWALD ELECTROSTATICS */
1012 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1013 ewrt = _mm_mul_ps(r20,ewtabscale);
1014 ewitab = _mm_cvttps_epi32(ewrt);
1015 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1016 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1017 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1019 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1020 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1024 fscal = _mm_andnot_ps(dummy_mask,fscal);
1026 /* Calculate temporary vectorial force */
1027 tx = _mm_mul_ps(fscal,dx20);
1028 ty = _mm_mul_ps(fscal,dy20);
1029 tz = _mm_mul_ps(fscal,dz20);
1031 /* Update vectorial force */
1032 fix2 = _mm_add_ps(fix2,tx);
1033 fiy2 = _mm_add_ps(fiy2,ty);
1034 fiz2 = _mm_add_ps(fiz2,tz);
1036 fjx0 = _mm_add_ps(fjx0,tx);
1037 fjy0 = _mm_add_ps(fjy0,ty);
1038 fjz0 = _mm_add_ps(fjz0,tz);
1040 /**************************
1041 * CALCULATE INTERACTIONS *
1042 **************************/
1044 r30 = _mm_mul_ps(rsq30,rinv30);
1045 r30 = _mm_andnot_ps(dummy_mask,r30);
1047 /* Compute parameters for interactions between i and j atoms */
1048 qq30 = _mm_mul_ps(iq3,jq0);
1050 /* EWALD ELECTROSTATICS */
1052 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1053 ewrt = _mm_mul_ps(r30,ewtabscale);
1054 ewitab = _mm_cvttps_epi32(ewrt);
1055 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1056 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1057 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1059 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1060 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1064 fscal = _mm_andnot_ps(dummy_mask,fscal);
1066 /* Calculate temporary vectorial force */
1067 tx = _mm_mul_ps(fscal,dx30);
1068 ty = _mm_mul_ps(fscal,dy30);
1069 tz = _mm_mul_ps(fscal,dz30);
1071 /* Update vectorial force */
1072 fix3 = _mm_add_ps(fix3,tx);
1073 fiy3 = _mm_add_ps(fiy3,ty);
1074 fiz3 = _mm_add_ps(fiz3,tz);
1076 fjx0 = _mm_add_ps(fjx0,tx);
1077 fjy0 = _mm_add_ps(fjy0,ty);
1078 fjz0 = _mm_add_ps(fjz0,tz);
1080 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1081 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1082 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1083 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1085 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1087 /* Inner loop uses 111 flops */
1090 /* End of innermost loop */
1092 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1093 f+i_coord_offset+DIM,fshift+i_shift_offset);
1095 /* Increment number of inner iterations */
1096 inneriter += j_index_end - j_index_start;
1098 /* Outer loop uses 18 flops */
1101 /* Increment number of outer iterations */
1104 /* Update outer/inner flops */
1106 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*111);