2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013, 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.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gromacs/simd/math_x86_sse2_single.h"
50 #include "kernelutil_x86_sse2_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse2_single
54 * Electrostatics interaction: Ewald
55 * VdW interaction: LennardJones
56 * Geometry: Water3-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse2_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 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
102 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
103 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
105 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
107 __m128 dummy_mask,cutoff_mask;
108 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
109 __m128 one = _mm_set1_ps(1.0);
110 __m128 two = _mm_set1_ps(2.0);
116 jindex = nlist->jindex;
118 shiftidx = nlist->shift;
120 shiftvec = fr->shift_vec[0];
121 fshift = fr->fshift[0];
122 facel = _mm_set1_ps(fr->epsfac);
123 charge = mdatoms->chargeA;
124 nvdwtype = fr->ntype;
126 vdwtype = mdatoms->typeA;
128 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
129 ewtab = fr->ic->tabq_coul_FDV0;
130 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
131 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
133 /* Setup water-specific parameters */
134 inr = nlist->iinr[0];
135 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
136 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
137 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
138 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
140 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
141 rcutoff_scalar = fr->rcoulomb;
142 rcutoff = _mm_set1_ps(rcutoff_scalar);
143 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
145 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
146 rvdw = _mm_set1_ps(fr->rvdw);
148 /* Avoid stupid compiler warnings */
149 jnrA = jnrB = jnrC = jnrD = 0;
158 for(iidx=0;iidx<4*DIM;iidx++)
163 /* Start outer loop over neighborlists */
164 for(iidx=0; iidx<nri; iidx++)
166 /* Load shift vector for this list */
167 i_shift_offset = DIM*shiftidx[iidx];
169 /* Load limits for loop over neighbors */
170 j_index_start = jindex[iidx];
171 j_index_end = jindex[iidx+1];
173 /* Get outer coordinate index */
175 i_coord_offset = DIM*inr;
177 /* Load i particle coords and add shift vector */
178 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
179 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
181 fix0 = _mm_setzero_ps();
182 fiy0 = _mm_setzero_ps();
183 fiz0 = _mm_setzero_ps();
184 fix1 = _mm_setzero_ps();
185 fiy1 = _mm_setzero_ps();
186 fiz1 = _mm_setzero_ps();
187 fix2 = _mm_setzero_ps();
188 fiy2 = _mm_setzero_ps();
189 fiz2 = _mm_setzero_ps();
191 /* Reset potential sums */
192 velecsum = _mm_setzero_ps();
193 vvdwsum = _mm_setzero_ps();
195 /* Start inner kernel loop */
196 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
199 /* Get j neighbor index, and coordinate index */
204 j_coord_offsetA = DIM*jnrA;
205 j_coord_offsetB = DIM*jnrB;
206 j_coord_offsetC = DIM*jnrC;
207 j_coord_offsetD = DIM*jnrD;
209 /* load j atom coordinates */
210 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
211 x+j_coord_offsetC,x+j_coord_offsetD,
214 /* Calculate displacement vector */
215 dx00 = _mm_sub_ps(ix0,jx0);
216 dy00 = _mm_sub_ps(iy0,jy0);
217 dz00 = _mm_sub_ps(iz0,jz0);
218 dx10 = _mm_sub_ps(ix1,jx0);
219 dy10 = _mm_sub_ps(iy1,jy0);
220 dz10 = _mm_sub_ps(iz1,jz0);
221 dx20 = _mm_sub_ps(ix2,jx0);
222 dy20 = _mm_sub_ps(iy2,jy0);
223 dz20 = _mm_sub_ps(iz2,jz0);
225 /* Calculate squared distance and things based on it */
226 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
227 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
228 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
230 rinv00 = gmx_mm_invsqrt_ps(rsq00);
231 rinv10 = gmx_mm_invsqrt_ps(rsq10);
232 rinv20 = gmx_mm_invsqrt_ps(rsq20);
234 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
235 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
236 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
238 /* Load parameters for j particles */
239 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
240 charge+jnrC+0,charge+jnrD+0);
241 vdwjidx0A = 2*vdwtype[jnrA+0];
242 vdwjidx0B = 2*vdwtype[jnrB+0];
243 vdwjidx0C = 2*vdwtype[jnrC+0];
244 vdwjidx0D = 2*vdwtype[jnrD+0];
246 fjx0 = _mm_setzero_ps();
247 fjy0 = _mm_setzero_ps();
248 fjz0 = _mm_setzero_ps();
250 /**************************
251 * CALCULATE INTERACTIONS *
252 **************************/
254 if (gmx_mm_any_lt(rsq00,rcutoff2))
257 r00 = _mm_mul_ps(rsq00,rinv00);
259 /* Compute parameters for interactions between i and j atoms */
260 qq00 = _mm_mul_ps(iq0,jq0);
261 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
262 vdwparam+vdwioffset0+vdwjidx0B,
263 vdwparam+vdwioffset0+vdwjidx0C,
264 vdwparam+vdwioffset0+vdwjidx0D,
267 /* EWALD ELECTROSTATICS */
269 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
270 ewrt = _mm_mul_ps(r00,ewtabscale);
271 ewitab = _mm_cvttps_epi32(ewrt);
272 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
273 ewitab = _mm_slli_epi32(ewitab,2);
274 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
275 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
276 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
277 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
278 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
279 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
280 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
281 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
282 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
284 /* LENNARD-JONES DISPERSION/REPULSION */
286 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
287 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
288 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
289 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
290 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
291 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
293 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
295 /* Update potential sum for this i atom from the interaction with this j atom. */
296 velec = _mm_and_ps(velec,cutoff_mask);
297 velecsum = _mm_add_ps(velecsum,velec);
298 vvdw = _mm_and_ps(vvdw,cutoff_mask);
299 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
301 fscal = _mm_add_ps(felec,fvdw);
303 fscal = _mm_and_ps(fscal,cutoff_mask);
305 /* Calculate temporary vectorial force */
306 tx = _mm_mul_ps(fscal,dx00);
307 ty = _mm_mul_ps(fscal,dy00);
308 tz = _mm_mul_ps(fscal,dz00);
310 /* Update vectorial force */
311 fix0 = _mm_add_ps(fix0,tx);
312 fiy0 = _mm_add_ps(fiy0,ty);
313 fiz0 = _mm_add_ps(fiz0,tz);
315 fjx0 = _mm_add_ps(fjx0,tx);
316 fjy0 = _mm_add_ps(fjy0,ty);
317 fjz0 = _mm_add_ps(fjz0,tz);
321 /**************************
322 * CALCULATE INTERACTIONS *
323 **************************/
325 if (gmx_mm_any_lt(rsq10,rcutoff2))
328 r10 = _mm_mul_ps(rsq10,rinv10);
330 /* Compute parameters for interactions between i and j atoms */
331 qq10 = _mm_mul_ps(iq1,jq0);
333 /* EWALD ELECTROSTATICS */
335 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
336 ewrt = _mm_mul_ps(r10,ewtabscale);
337 ewitab = _mm_cvttps_epi32(ewrt);
338 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
339 ewitab = _mm_slli_epi32(ewitab,2);
340 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
341 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
342 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
343 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
344 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
345 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
346 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
347 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
348 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
350 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
352 /* Update potential sum for this i atom from the interaction with this j atom. */
353 velec = _mm_and_ps(velec,cutoff_mask);
354 velecsum = _mm_add_ps(velecsum,velec);
358 fscal = _mm_and_ps(fscal,cutoff_mask);
360 /* Calculate temporary vectorial force */
361 tx = _mm_mul_ps(fscal,dx10);
362 ty = _mm_mul_ps(fscal,dy10);
363 tz = _mm_mul_ps(fscal,dz10);
365 /* Update vectorial force */
366 fix1 = _mm_add_ps(fix1,tx);
367 fiy1 = _mm_add_ps(fiy1,ty);
368 fiz1 = _mm_add_ps(fiz1,tz);
370 fjx0 = _mm_add_ps(fjx0,tx);
371 fjy0 = _mm_add_ps(fjy0,ty);
372 fjz0 = _mm_add_ps(fjz0,tz);
376 /**************************
377 * CALCULATE INTERACTIONS *
378 **************************/
380 if (gmx_mm_any_lt(rsq20,rcutoff2))
383 r20 = _mm_mul_ps(rsq20,rinv20);
385 /* Compute parameters for interactions between i and j atoms */
386 qq20 = _mm_mul_ps(iq2,jq0);
388 /* EWALD ELECTROSTATICS */
390 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
391 ewrt = _mm_mul_ps(r20,ewtabscale);
392 ewitab = _mm_cvttps_epi32(ewrt);
393 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
394 ewitab = _mm_slli_epi32(ewitab,2);
395 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
396 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
397 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
398 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
399 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
400 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
401 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
402 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
403 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
405 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
407 /* Update potential sum for this i atom from the interaction with this j atom. */
408 velec = _mm_and_ps(velec,cutoff_mask);
409 velecsum = _mm_add_ps(velecsum,velec);
413 fscal = _mm_and_ps(fscal,cutoff_mask);
415 /* Calculate temporary vectorial force */
416 tx = _mm_mul_ps(fscal,dx20);
417 ty = _mm_mul_ps(fscal,dy20);
418 tz = _mm_mul_ps(fscal,dz20);
420 /* Update vectorial force */
421 fix2 = _mm_add_ps(fix2,tx);
422 fiy2 = _mm_add_ps(fiy2,ty);
423 fiz2 = _mm_add_ps(fiz2,tz);
425 fjx0 = _mm_add_ps(fjx0,tx);
426 fjy0 = _mm_add_ps(fjy0,ty);
427 fjz0 = _mm_add_ps(fjz0,tz);
431 fjptrA = f+j_coord_offsetA;
432 fjptrB = f+j_coord_offsetB;
433 fjptrC = f+j_coord_offsetC;
434 fjptrD = f+j_coord_offsetD;
436 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
438 /* Inner loop uses 156 flops */
444 /* Get j neighbor index, and coordinate index */
445 jnrlistA = jjnr[jidx];
446 jnrlistB = jjnr[jidx+1];
447 jnrlistC = jjnr[jidx+2];
448 jnrlistD = jjnr[jidx+3];
449 /* Sign of each element will be negative for non-real atoms.
450 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
451 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
453 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
454 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
455 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
456 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
457 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
458 j_coord_offsetA = DIM*jnrA;
459 j_coord_offsetB = DIM*jnrB;
460 j_coord_offsetC = DIM*jnrC;
461 j_coord_offsetD = DIM*jnrD;
463 /* load j atom coordinates */
464 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
465 x+j_coord_offsetC,x+j_coord_offsetD,
468 /* Calculate displacement vector */
469 dx00 = _mm_sub_ps(ix0,jx0);
470 dy00 = _mm_sub_ps(iy0,jy0);
471 dz00 = _mm_sub_ps(iz0,jz0);
472 dx10 = _mm_sub_ps(ix1,jx0);
473 dy10 = _mm_sub_ps(iy1,jy0);
474 dz10 = _mm_sub_ps(iz1,jz0);
475 dx20 = _mm_sub_ps(ix2,jx0);
476 dy20 = _mm_sub_ps(iy2,jy0);
477 dz20 = _mm_sub_ps(iz2,jz0);
479 /* Calculate squared distance and things based on it */
480 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
481 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
482 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
484 rinv00 = gmx_mm_invsqrt_ps(rsq00);
485 rinv10 = gmx_mm_invsqrt_ps(rsq10);
486 rinv20 = gmx_mm_invsqrt_ps(rsq20);
488 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
489 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
490 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
492 /* Load parameters for j particles */
493 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
494 charge+jnrC+0,charge+jnrD+0);
495 vdwjidx0A = 2*vdwtype[jnrA+0];
496 vdwjidx0B = 2*vdwtype[jnrB+0];
497 vdwjidx0C = 2*vdwtype[jnrC+0];
498 vdwjidx0D = 2*vdwtype[jnrD+0];
500 fjx0 = _mm_setzero_ps();
501 fjy0 = _mm_setzero_ps();
502 fjz0 = _mm_setzero_ps();
504 /**************************
505 * CALCULATE INTERACTIONS *
506 **************************/
508 if (gmx_mm_any_lt(rsq00,rcutoff2))
511 r00 = _mm_mul_ps(rsq00,rinv00);
512 r00 = _mm_andnot_ps(dummy_mask,r00);
514 /* Compute parameters for interactions between i and j atoms */
515 qq00 = _mm_mul_ps(iq0,jq0);
516 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
517 vdwparam+vdwioffset0+vdwjidx0B,
518 vdwparam+vdwioffset0+vdwjidx0C,
519 vdwparam+vdwioffset0+vdwjidx0D,
522 /* EWALD ELECTROSTATICS */
524 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
525 ewrt = _mm_mul_ps(r00,ewtabscale);
526 ewitab = _mm_cvttps_epi32(ewrt);
527 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
528 ewitab = _mm_slli_epi32(ewitab,2);
529 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
530 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
531 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
532 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
533 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
534 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
535 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
536 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
537 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
539 /* LENNARD-JONES DISPERSION/REPULSION */
541 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
542 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
543 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
544 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
545 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
546 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
548 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
550 /* Update potential sum for this i atom from the interaction with this j atom. */
551 velec = _mm_and_ps(velec,cutoff_mask);
552 velec = _mm_andnot_ps(dummy_mask,velec);
553 velecsum = _mm_add_ps(velecsum,velec);
554 vvdw = _mm_and_ps(vvdw,cutoff_mask);
555 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
556 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
558 fscal = _mm_add_ps(felec,fvdw);
560 fscal = _mm_and_ps(fscal,cutoff_mask);
562 fscal = _mm_andnot_ps(dummy_mask,fscal);
564 /* Calculate temporary vectorial force */
565 tx = _mm_mul_ps(fscal,dx00);
566 ty = _mm_mul_ps(fscal,dy00);
567 tz = _mm_mul_ps(fscal,dz00);
569 /* Update vectorial force */
570 fix0 = _mm_add_ps(fix0,tx);
571 fiy0 = _mm_add_ps(fiy0,ty);
572 fiz0 = _mm_add_ps(fiz0,tz);
574 fjx0 = _mm_add_ps(fjx0,tx);
575 fjy0 = _mm_add_ps(fjy0,ty);
576 fjz0 = _mm_add_ps(fjz0,tz);
580 /**************************
581 * CALCULATE INTERACTIONS *
582 **************************/
584 if (gmx_mm_any_lt(rsq10,rcutoff2))
587 r10 = _mm_mul_ps(rsq10,rinv10);
588 r10 = _mm_andnot_ps(dummy_mask,r10);
590 /* Compute parameters for interactions between i and j atoms */
591 qq10 = _mm_mul_ps(iq1,jq0);
593 /* EWALD ELECTROSTATICS */
595 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
596 ewrt = _mm_mul_ps(r10,ewtabscale);
597 ewitab = _mm_cvttps_epi32(ewrt);
598 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
599 ewitab = _mm_slli_epi32(ewitab,2);
600 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
601 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
602 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
603 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
604 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
605 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
606 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
607 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
608 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
610 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
612 /* Update potential sum for this i atom from the interaction with this j atom. */
613 velec = _mm_and_ps(velec,cutoff_mask);
614 velec = _mm_andnot_ps(dummy_mask,velec);
615 velecsum = _mm_add_ps(velecsum,velec);
619 fscal = _mm_and_ps(fscal,cutoff_mask);
621 fscal = _mm_andnot_ps(dummy_mask,fscal);
623 /* Calculate temporary vectorial force */
624 tx = _mm_mul_ps(fscal,dx10);
625 ty = _mm_mul_ps(fscal,dy10);
626 tz = _mm_mul_ps(fscal,dz10);
628 /* Update vectorial force */
629 fix1 = _mm_add_ps(fix1,tx);
630 fiy1 = _mm_add_ps(fiy1,ty);
631 fiz1 = _mm_add_ps(fiz1,tz);
633 fjx0 = _mm_add_ps(fjx0,tx);
634 fjy0 = _mm_add_ps(fjy0,ty);
635 fjz0 = _mm_add_ps(fjz0,tz);
639 /**************************
640 * CALCULATE INTERACTIONS *
641 **************************/
643 if (gmx_mm_any_lt(rsq20,rcutoff2))
646 r20 = _mm_mul_ps(rsq20,rinv20);
647 r20 = _mm_andnot_ps(dummy_mask,r20);
649 /* Compute parameters for interactions between i and j atoms */
650 qq20 = _mm_mul_ps(iq2,jq0);
652 /* EWALD ELECTROSTATICS */
654 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
655 ewrt = _mm_mul_ps(r20,ewtabscale);
656 ewitab = _mm_cvttps_epi32(ewrt);
657 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
658 ewitab = _mm_slli_epi32(ewitab,2);
659 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
660 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
661 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
662 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
663 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
664 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
665 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
666 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
667 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
669 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
671 /* Update potential sum for this i atom from the interaction with this j atom. */
672 velec = _mm_and_ps(velec,cutoff_mask);
673 velec = _mm_andnot_ps(dummy_mask,velec);
674 velecsum = _mm_add_ps(velecsum,velec);
678 fscal = _mm_and_ps(fscal,cutoff_mask);
680 fscal = _mm_andnot_ps(dummy_mask,fscal);
682 /* Calculate temporary vectorial force */
683 tx = _mm_mul_ps(fscal,dx20);
684 ty = _mm_mul_ps(fscal,dy20);
685 tz = _mm_mul_ps(fscal,dz20);
687 /* Update vectorial force */
688 fix2 = _mm_add_ps(fix2,tx);
689 fiy2 = _mm_add_ps(fiy2,ty);
690 fiz2 = _mm_add_ps(fiz2,tz);
692 fjx0 = _mm_add_ps(fjx0,tx);
693 fjy0 = _mm_add_ps(fjy0,ty);
694 fjz0 = _mm_add_ps(fjz0,tz);
698 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
699 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
700 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
701 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
703 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
705 /* Inner loop uses 159 flops */
708 /* End of innermost loop */
710 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
711 f+i_coord_offset,fshift+i_shift_offset);
714 /* Update potential energies */
715 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
716 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
718 /* Increment number of inner iterations */
719 inneriter += j_index_end - j_index_start;
721 /* Outer loop uses 20 flops */
724 /* Increment number of outer iterations */
727 /* Update outer/inner flops */
729 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
732 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse2_single
733 * Electrostatics interaction: Ewald
734 * VdW interaction: LennardJones
735 * Geometry: Water3-Particle
736 * Calculate force/pot: Force
739 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse2_single
740 (t_nblist * gmx_restrict nlist,
741 rvec * gmx_restrict xx,
742 rvec * gmx_restrict ff,
743 t_forcerec * gmx_restrict fr,
744 t_mdatoms * gmx_restrict mdatoms,
745 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
746 t_nrnb * gmx_restrict nrnb)
748 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
749 * just 0 for non-waters.
750 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
751 * jnr indices corresponding to data put in the four positions in the SIMD register.
753 int i_shift_offset,i_coord_offset,outeriter,inneriter;
754 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
755 int jnrA,jnrB,jnrC,jnrD;
756 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
757 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
758 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
760 real *shiftvec,*fshift,*x,*f;
761 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
763 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
765 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
767 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
769 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
770 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
771 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
772 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
773 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
774 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
775 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
778 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
781 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
782 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
784 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
786 __m128 dummy_mask,cutoff_mask;
787 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
788 __m128 one = _mm_set1_ps(1.0);
789 __m128 two = _mm_set1_ps(2.0);
795 jindex = nlist->jindex;
797 shiftidx = nlist->shift;
799 shiftvec = fr->shift_vec[0];
800 fshift = fr->fshift[0];
801 facel = _mm_set1_ps(fr->epsfac);
802 charge = mdatoms->chargeA;
803 nvdwtype = fr->ntype;
805 vdwtype = mdatoms->typeA;
807 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
808 ewtab = fr->ic->tabq_coul_F;
809 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
810 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
812 /* Setup water-specific parameters */
813 inr = nlist->iinr[0];
814 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
815 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
816 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
817 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
819 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
820 rcutoff_scalar = fr->rcoulomb;
821 rcutoff = _mm_set1_ps(rcutoff_scalar);
822 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
824 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
825 rvdw = _mm_set1_ps(fr->rvdw);
827 /* Avoid stupid compiler warnings */
828 jnrA = jnrB = jnrC = jnrD = 0;
837 for(iidx=0;iidx<4*DIM;iidx++)
842 /* Start outer loop over neighborlists */
843 for(iidx=0; iidx<nri; iidx++)
845 /* Load shift vector for this list */
846 i_shift_offset = DIM*shiftidx[iidx];
848 /* Load limits for loop over neighbors */
849 j_index_start = jindex[iidx];
850 j_index_end = jindex[iidx+1];
852 /* Get outer coordinate index */
854 i_coord_offset = DIM*inr;
856 /* Load i particle coords and add shift vector */
857 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
858 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
860 fix0 = _mm_setzero_ps();
861 fiy0 = _mm_setzero_ps();
862 fiz0 = _mm_setzero_ps();
863 fix1 = _mm_setzero_ps();
864 fiy1 = _mm_setzero_ps();
865 fiz1 = _mm_setzero_ps();
866 fix2 = _mm_setzero_ps();
867 fiy2 = _mm_setzero_ps();
868 fiz2 = _mm_setzero_ps();
870 /* Start inner kernel loop */
871 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
874 /* Get j neighbor index, and coordinate index */
879 j_coord_offsetA = DIM*jnrA;
880 j_coord_offsetB = DIM*jnrB;
881 j_coord_offsetC = DIM*jnrC;
882 j_coord_offsetD = DIM*jnrD;
884 /* load j atom coordinates */
885 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
886 x+j_coord_offsetC,x+j_coord_offsetD,
889 /* Calculate displacement vector */
890 dx00 = _mm_sub_ps(ix0,jx0);
891 dy00 = _mm_sub_ps(iy0,jy0);
892 dz00 = _mm_sub_ps(iz0,jz0);
893 dx10 = _mm_sub_ps(ix1,jx0);
894 dy10 = _mm_sub_ps(iy1,jy0);
895 dz10 = _mm_sub_ps(iz1,jz0);
896 dx20 = _mm_sub_ps(ix2,jx0);
897 dy20 = _mm_sub_ps(iy2,jy0);
898 dz20 = _mm_sub_ps(iz2,jz0);
900 /* Calculate squared distance and things based on it */
901 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
902 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
903 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
905 rinv00 = gmx_mm_invsqrt_ps(rsq00);
906 rinv10 = gmx_mm_invsqrt_ps(rsq10);
907 rinv20 = gmx_mm_invsqrt_ps(rsq20);
909 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
910 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
911 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
913 /* Load parameters for j particles */
914 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
915 charge+jnrC+0,charge+jnrD+0);
916 vdwjidx0A = 2*vdwtype[jnrA+0];
917 vdwjidx0B = 2*vdwtype[jnrB+0];
918 vdwjidx0C = 2*vdwtype[jnrC+0];
919 vdwjidx0D = 2*vdwtype[jnrD+0];
921 fjx0 = _mm_setzero_ps();
922 fjy0 = _mm_setzero_ps();
923 fjz0 = _mm_setzero_ps();
925 /**************************
926 * CALCULATE INTERACTIONS *
927 **************************/
929 if (gmx_mm_any_lt(rsq00,rcutoff2))
932 r00 = _mm_mul_ps(rsq00,rinv00);
934 /* Compute parameters for interactions between i and j atoms */
935 qq00 = _mm_mul_ps(iq0,jq0);
936 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
937 vdwparam+vdwioffset0+vdwjidx0B,
938 vdwparam+vdwioffset0+vdwjidx0C,
939 vdwparam+vdwioffset0+vdwjidx0D,
942 /* EWALD ELECTROSTATICS */
944 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
945 ewrt = _mm_mul_ps(r00,ewtabscale);
946 ewitab = _mm_cvttps_epi32(ewrt);
947 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
948 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
949 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
951 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
952 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
954 /* LENNARD-JONES DISPERSION/REPULSION */
956 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
957 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
959 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
961 fscal = _mm_add_ps(felec,fvdw);
963 fscal = _mm_and_ps(fscal,cutoff_mask);
965 /* Calculate temporary vectorial force */
966 tx = _mm_mul_ps(fscal,dx00);
967 ty = _mm_mul_ps(fscal,dy00);
968 tz = _mm_mul_ps(fscal,dz00);
970 /* Update vectorial force */
971 fix0 = _mm_add_ps(fix0,tx);
972 fiy0 = _mm_add_ps(fiy0,ty);
973 fiz0 = _mm_add_ps(fiz0,tz);
975 fjx0 = _mm_add_ps(fjx0,tx);
976 fjy0 = _mm_add_ps(fjy0,ty);
977 fjz0 = _mm_add_ps(fjz0,tz);
981 /**************************
982 * CALCULATE INTERACTIONS *
983 **************************/
985 if (gmx_mm_any_lt(rsq10,rcutoff2))
988 r10 = _mm_mul_ps(rsq10,rinv10);
990 /* Compute parameters for interactions between i and j atoms */
991 qq10 = _mm_mul_ps(iq1,jq0);
993 /* EWALD ELECTROSTATICS */
995 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
996 ewrt = _mm_mul_ps(r10,ewtabscale);
997 ewitab = _mm_cvttps_epi32(ewrt);
998 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
999 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1000 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1002 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1003 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1005 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1009 fscal = _mm_and_ps(fscal,cutoff_mask);
1011 /* Calculate temporary vectorial force */
1012 tx = _mm_mul_ps(fscal,dx10);
1013 ty = _mm_mul_ps(fscal,dy10);
1014 tz = _mm_mul_ps(fscal,dz10);
1016 /* Update vectorial force */
1017 fix1 = _mm_add_ps(fix1,tx);
1018 fiy1 = _mm_add_ps(fiy1,ty);
1019 fiz1 = _mm_add_ps(fiz1,tz);
1021 fjx0 = _mm_add_ps(fjx0,tx);
1022 fjy0 = _mm_add_ps(fjy0,ty);
1023 fjz0 = _mm_add_ps(fjz0,tz);
1027 /**************************
1028 * CALCULATE INTERACTIONS *
1029 **************************/
1031 if (gmx_mm_any_lt(rsq20,rcutoff2))
1034 r20 = _mm_mul_ps(rsq20,rinv20);
1036 /* Compute parameters for interactions between i and j atoms */
1037 qq20 = _mm_mul_ps(iq2,jq0);
1039 /* EWALD ELECTROSTATICS */
1041 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1042 ewrt = _mm_mul_ps(r20,ewtabscale);
1043 ewitab = _mm_cvttps_epi32(ewrt);
1044 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1045 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1046 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1048 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1049 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1051 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1055 fscal = _mm_and_ps(fscal,cutoff_mask);
1057 /* Calculate temporary vectorial force */
1058 tx = _mm_mul_ps(fscal,dx20);
1059 ty = _mm_mul_ps(fscal,dy20);
1060 tz = _mm_mul_ps(fscal,dz20);
1062 /* Update vectorial force */
1063 fix2 = _mm_add_ps(fix2,tx);
1064 fiy2 = _mm_add_ps(fiy2,ty);
1065 fiz2 = _mm_add_ps(fiz2,tz);
1067 fjx0 = _mm_add_ps(fjx0,tx);
1068 fjy0 = _mm_add_ps(fjy0,ty);
1069 fjz0 = _mm_add_ps(fjz0,tz);
1073 fjptrA = f+j_coord_offsetA;
1074 fjptrB = f+j_coord_offsetB;
1075 fjptrC = f+j_coord_offsetC;
1076 fjptrD = f+j_coord_offsetD;
1078 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1080 /* Inner loop uses 124 flops */
1083 if(jidx<j_index_end)
1086 /* Get j neighbor index, and coordinate index */
1087 jnrlistA = jjnr[jidx];
1088 jnrlistB = jjnr[jidx+1];
1089 jnrlistC = jjnr[jidx+2];
1090 jnrlistD = jjnr[jidx+3];
1091 /* Sign of each element will be negative for non-real atoms.
1092 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1093 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1095 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1096 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1097 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1098 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1099 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1100 j_coord_offsetA = DIM*jnrA;
1101 j_coord_offsetB = DIM*jnrB;
1102 j_coord_offsetC = DIM*jnrC;
1103 j_coord_offsetD = DIM*jnrD;
1105 /* load j atom coordinates */
1106 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1107 x+j_coord_offsetC,x+j_coord_offsetD,
1110 /* Calculate displacement vector */
1111 dx00 = _mm_sub_ps(ix0,jx0);
1112 dy00 = _mm_sub_ps(iy0,jy0);
1113 dz00 = _mm_sub_ps(iz0,jz0);
1114 dx10 = _mm_sub_ps(ix1,jx0);
1115 dy10 = _mm_sub_ps(iy1,jy0);
1116 dz10 = _mm_sub_ps(iz1,jz0);
1117 dx20 = _mm_sub_ps(ix2,jx0);
1118 dy20 = _mm_sub_ps(iy2,jy0);
1119 dz20 = _mm_sub_ps(iz2,jz0);
1121 /* Calculate squared distance and things based on it */
1122 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1123 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1124 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1126 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1127 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1128 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1130 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1131 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1132 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1134 /* Load parameters for j particles */
1135 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1136 charge+jnrC+0,charge+jnrD+0);
1137 vdwjidx0A = 2*vdwtype[jnrA+0];
1138 vdwjidx0B = 2*vdwtype[jnrB+0];
1139 vdwjidx0C = 2*vdwtype[jnrC+0];
1140 vdwjidx0D = 2*vdwtype[jnrD+0];
1142 fjx0 = _mm_setzero_ps();
1143 fjy0 = _mm_setzero_ps();
1144 fjz0 = _mm_setzero_ps();
1146 /**************************
1147 * CALCULATE INTERACTIONS *
1148 **************************/
1150 if (gmx_mm_any_lt(rsq00,rcutoff2))
1153 r00 = _mm_mul_ps(rsq00,rinv00);
1154 r00 = _mm_andnot_ps(dummy_mask,r00);
1156 /* Compute parameters for interactions between i and j atoms */
1157 qq00 = _mm_mul_ps(iq0,jq0);
1158 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1159 vdwparam+vdwioffset0+vdwjidx0B,
1160 vdwparam+vdwioffset0+vdwjidx0C,
1161 vdwparam+vdwioffset0+vdwjidx0D,
1164 /* EWALD ELECTROSTATICS */
1166 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1167 ewrt = _mm_mul_ps(r00,ewtabscale);
1168 ewitab = _mm_cvttps_epi32(ewrt);
1169 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1170 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1171 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1173 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1174 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1176 /* LENNARD-JONES DISPERSION/REPULSION */
1178 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1179 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1181 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1183 fscal = _mm_add_ps(felec,fvdw);
1185 fscal = _mm_and_ps(fscal,cutoff_mask);
1187 fscal = _mm_andnot_ps(dummy_mask,fscal);
1189 /* Calculate temporary vectorial force */
1190 tx = _mm_mul_ps(fscal,dx00);
1191 ty = _mm_mul_ps(fscal,dy00);
1192 tz = _mm_mul_ps(fscal,dz00);
1194 /* Update vectorial force */
1195 fix0 = _mm_add_ps(fix0,tx);
1196 fiy0 = _mm_add_ps(fiy0,ty);
1197 fiz0 = _mm_add_ps(fiz0,tz);
1199 fjx0 = _mm_add_ps(fjx0,tx);
1200 fjy0 = _mm_add_ps(fjy0,ty);
1201 fjz0 = _mm_add_ps(fjz0,tz);
1205 /**************************
1206 * CALCULATE INTERACTIONS *
1207 **************************/
1209 if (gmx_mm_any_lt(rsq10,rcutoff2))
1212 r10 = _mm_mul_ps(rsq10,rinv10);
1213 r10 = _mm_andnot_ps(dummy_mask,r10);
1215 /* Compute parameters for interactions between i and j atoms */
1216 qq10 = _mm_mul_ps(iq1,jq0);
1218 /* EWALD ELECTROSTATICS */
1220 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1221 ewrt = _mm_mul_ps(r10,ewtabscale);
1222 ewitab = _mm_cvttps_epi32(ewrt);
1223 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1224 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1225 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1227 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1228 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1230 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1234 fscal = _mm_and_ps(fscal,cutoff_mask);
1236 fscal = _mm_andnot_ps(dummy_mask,fscal);
1238 /* Calculate temporary vectorial force */
1239 tx = _mm_mul_ps(fscal,dx10);
1240 ty = _mm_mul_ps(fscal,dy10);
1241 tz = _mm_mul_ps(fscal,dz10);
1243 /* Update vectorial force */
1244 fix1 = _mm_add_ps(fix1,tx);
1245 fiy1 = _mm_add_ps(fiy1,ty);
1246 fiz1 = _mm_add_ps(fiz1,tz);
1248 fjx0 = _mm_add_ps(fjx0,tx);
1249 fjy0 = _mm_add_ps(fjy0,ty);
1250 fjz0 = _mm_add_ps(fjz0,tz);
1254 /**************************
1255 * CALCULATE INTERACTIONS *
1256 **************************/
1258 if (gmx_mm_any_lt(rsq20,rcutoff2))
1261 r20 = _mm_mul_ps(rsq20,rinv20);
1262 r20 = _mm_andnot_ps(dummy_mask,r20);
1264 /* Compute parameters for interactions between i and j atoms */
1265 qq20 = _mm_mul_ps(iq2,jq0);
1267 /* EWALD ELECTROSTATICS */
1269 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1270 ewrt = _mm_mul_ps(r20,ewtabscale);
1271 ewitab = _mm_cvttps_epi32(ewrt);
1272 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1273 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1274 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1276 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1277 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1279 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1283 fscal = _mm_and_ps(fscal,cutoff_mask);
1285 fscal = _mm_andnot_ps(dummy_mask,fscal);
1287 /* Calculate temporary vectorial force */
1288 tx = _mm_mul_ps(fscal,dx20);
1289 ty = _mm_mul_ps(fscal,dy20);
1290 tz = _mm_mul_ps(fscal,dz20);
1292 /* Update vectorial force */
1293 fix2 = _mm_add_ps(fix2,tx);
1294 fiy2 = _mm_add_ps(fiy2,ty);
1295 fiz2 = _mm_add_ps(fiz2,tz);
1297 fjx0 = _mm_add_ps(fjx0,tx);
1298 fjy0 = _mm_add_ps(fjy0,ty);
1299 fjz0 = _mm_add_ps(fjz0,tz);
1303 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1304 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1305 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1306 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1308 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1310 /* Inner loop uses 127 flops */
1313 /* End of innermost loop */
1315 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1316 f+i_coord_offset,fshift+i_shift_offset);
1318 /* Increment number of inner iterations */
1319 inneriter += j_index_end - j_index_start;
1321 /* Outer loop uses 18 flops */
1324 /* Increment number of outer iterations */
1327 /* Update outer/inner flops */
1329 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);