2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
50 #include "kernelutil_x86_avx_128_fma_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_128_fma_single
54 * Electrostatics interaction: Ewald
55 * VdW interaction: None
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_128_fma_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 AVX_128, 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 fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
87 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
88 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
89 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
90 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
93 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
94 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
96 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
97 real rswitch_scalar,d_scalar;
98 __m128 dummy_mask,cutoff_mask;
99 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
100 __m128 one = _mm_set1_ps(1.0);
101 __m128 two = _mm_set1_ps(2.0);
107 jindex = nlist->jindex;
109 shiftidx = nlist->shift;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm_set1_ps(fr->epsfac);
114 charge = mdatoms->chargeA;
116 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
117 beta = _mm_set1_ps(fr->ic->ewaldcoeff_q);
118 beta2 = _mm_mul_ps(beta,beta);
119 beta3 = _mm_mul_ps(beta,beta2);
120 ewtab = fr->ic->tabq_coul_FDV0;
121 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
122 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
124 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
125 rcutoff_scalar = fr->rcoulomb;
126 rcutoff = _mm_set1_ps(rcutoff_scalar);
127 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
129 rswitch_scalar = fr->rcoulomb_switch;
130 rswitch = _mm_set1_ps(rswitch_scalar);
131 /* Setup switch parameters */
132 d_scalar = rcutoff_scalar-rswitch_scalar;
133 d = _mm_set1_ps(d_scalar);
134 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
135 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
136 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
137 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
138 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
139 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
141 /* Avoid stupid compiler warnings */
142 jnrA = jnrB = jnrC = jnrD = 0;
151 for(iidx=0;iidx<4*DIM;iidx++)
156 /* Start outer loop over neighborlists */
157 for(iidx=0; iidx<nri; iidx++)
159 /* Load shift vector for this list */
160 i_shift_offset = DIM*shiftidx[iidx];
162 /* Load limits for loop over neighbors */
163 j_index_start = jindex[iidx];
164 j_index_end = jindex[iidx+1];
166 /* Get outer coordinate index */
168 i_coord_offset = DIM*inr;
170 /* Load i particle coords and add shift vector */
171 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
173 fix0 = _mm_setzero_ps();
174 fiy0 = _mm_setzero_ps();
175 fiz0 = _mm_setzero_ps();
177 /* Load parameters for i particles */
178 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
180 /* Reset potential sums */
181 velecsum = _mm_setzero_ps();
183 /* Start inner kernel loop */
184 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
187 /* Get j neighbor index, and coordinate index */
192 j_coord_offsetA = DIM*jnrA;
193 j_coord_offsetB = DIM*jnrB;
194 j_coord_offsetC = DIM*jnrC;
195 j_coord_offsetD = DIM*jnrD;
197 /* load j atom coordinates */
198 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
199 x+j_coord_offsetC,x+j_coord_offsetD,
202 /* Calculate displacement vector */
203 dx00 = _mm_sub_ps(ix0,jx0);
204 dy00 = _mm_sub_ps(iy0,jy0);
205 dz00 = _mm_sub_ps(iz0,jz0);
207 /* Calculate squared distance and things based on it */
208 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
210 rinv00 = gmx_mm_invsqrt_ps(rsq00);
212 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
214 /* Load parameters for j particles */
215 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
216 charge+jnrC+0,charge+jnrD+0);
218 /**************************
219 * CALCULATE INTERACTIONS *
220 **************************/
222 if (gmx_mm_any_lt(rsq00,rcutoff2))
225 r00 = _mm_mul_ps(rsq00,rinv00);
227 /* Compute parameters for interactions between i and j atoms */
228 qq00 = _mm_mul_ps(iq0,jq0);
230 /* EWALD ELECTROSTATICS */
232 /* Analytical PME correction */
233 zeta2 = _mm_mul_ps(beta2,rsq00);
234 rinv3 = _mm_mul_ps(rinvsq00,rinv00);
235 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
236 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
237 felec = _mm_mul_ps(qq00,felec);
238 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
239 velec = _mm_nmacc_ps(pmecorrV,beta,rinv00);
240 velec = _mm_mul_ps(qq00,velec);
242 d = _mm_sub_ps(r00,rswitch);
243 d = _mm_max_ps(d,_mm_setzero_ps());
244 d2 = _mm_mul_ps(d,d);
245 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
247 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
249 /* Evaluate switch function */
250 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
251 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
252 velec = _mm_mul_ps(velec,sw);
253 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
255 /* Update potential sum for this i atom from the interaction with this j atom. */
256 velec = _mm_and_ps(velec,cutoff_mask);
257 velecsum = _mm_add_ps(velecsum,velec);
261 fscal = _mm_and_ps(fscal,cutoff_mask);
263 /* Update vectorial force */
264 fix0 = _mm_macc_ps(dx00,fscal,fix0);
265 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
266 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
268 fjptrA = f+j_coord_offsetA;
269 fjptrB = f+j_coord_offsetB;
270 fjptrC = f+j_coord_offsetC;
271 fjptrD = f+j_coord_offsetD;
272 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
273 _mm_mul_ps(dx00,fscal),
274 _mm_mul_ps(dy00,fscal),
275 _mm_mul_ps(dz00,fscal));
279 /* Inner loop uses 53 flops */
285 /* Get j neighbor index, and coordinate index */
286 jnrlistA = jjnr[jidx];
287 jnrlistB = jjnr[jidx+1];
288 jnrlistC = jjnr[jidx+2];
289 jnrlistD = jjnr[jidx+3];
290 /* Sign of each element will be negative for non-real atoms.
291 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
292 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
294 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
295 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
296 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
297 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
298 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
299 j_coord_offsetA = DIM*jnrA;
300 j_coord_offsetB = DIM*jnrB;
301 j_coord_offsetC = DIM*jnrC;
302 j_coord_offsetD = DIM*jnrD;
304 /* load j atom coordinates */
305 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
306 x+j_coord_offsetC,x+j_coord_offsetD,
309 /* Calculate displacement vector */
310 dx00 = _mm_sub_ps(ix0,jx0);
311 dy00 = _mm_sub_ps(iy0,jy0);
312 dz00 = _mm_sub_ps(iz0,jz0);
314 /* Calculate squared distance and things based on it */
315 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
317 rinv00 = gmx_mm_invsqrt_ps(rsq00);
319 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
321 /* Load parameters for j particles */
322 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
323 charge+jnrC+0,charge+jnrD+0);
325 /**************************
326 * CALCULATE INTERACTIONS *
327 **************************/
329 if (gmx_mm_any_lt(rsq00,rcutoff2))
332 r00 = _mm_mul_ps(rsq00,rinv00);
333 r00 = _mm_andnot_ps(dummy_mask,r00);
335 /* Compute parameters for interactions between i and j atoms */
336 qq00 = _mm_mul_ps(iq0,jq0);
338 /* EWALD ELECTROSTATICS */
340 /* Analytical PME correction */
341 zeta2 = _mm_mul_ps(beta2,rsq00);
342 rinv3 = _mm_mul_ps(rinvsq00,rinv00);
343 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
344 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
345 felec = _mm_mul_ps(qq00,felec);
346 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
347 velec = _mm_nmacc_ps(pmecorrV,beta,rinv00);
348 velec = _mm_mul_ps(qq00,velec);
350 d = _mm_sub_ps(r00,rswitch);
351 d = _mm_max_ps(d,_mm_setzero_ps());
352 d2 = _mm_mul_ps(d,d);
353 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
355 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
357 /* Evaluate switch function */
358 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
359 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
360 velec = _mm_mul_ps(velec,sw);
361 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
363 /* Update potential sum for this i atom from the interaction with this j atom. */
364 velec = _mm_and_ps(velec,cutoff_mask);
365 velec = _mm_andnot_ps(dummy_mask,velec);
366 velecsum = _mm_add_ps(velecsum,velec);
370 fscal = _mm_and_ps(fscal,cutoff_mask);
372 fscal = _mm_andnot_ps(dummy_mask,fscal);
374 /* Update vectorial force */
375 fix0 = _mm_macc_ps(dx00,fscal,fix0);
376 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
377 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
379 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
380 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
381 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
382 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
383 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
384 _mm_mul_ps(dx00,fscal),
385 _mm_mul_ps(dy00,fscal),
386 _mm_mul_ps(dz00,fscal));
390 /* Inner loop uses 54 flops */
393 /* End of innermost loop */
395 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
396 f+i_coord_offset,fshift+i_shift_offset);
399 /* Update potential energies */
400 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
402 /* Increment number of inner iterations */
403 inneriter += j_index_end - j_index_start;
405 /* Outer loop uses 8 flops */
408 /* Increment number of outer iterations */
411 /* Update outer/inner flops */
413 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*54);
416 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_128_fma_single
417 * Electrostatics interaction: Ewald
418 * VdW interaction: None
419 * Geometry: Particle-Particle
420 * Calculate force/pot: Force
423 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_128_fma_single
424 (t_nblist * gmx_restrict nlist,
425 rvec * gmx_restrict xx,
426 rvec * gmx_restrict ff,
427 t_forcerec * gmx_restrict fr,
428 t_mdatoms * gmx_restrict mdatoms,
429 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
430 t_nrnb * gmx_restrict nrnb)
432 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
433 * just 0 for non-waters.
434 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
435 * jnr indices corresponding to data put in the four positions in the SIMD register.
437 int i_shift_offset,i_coord_offset,outeriter,inneriter;
438 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
439 int jnrA,jnrB,jnrC,jnrD;
440 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
441 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
442 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
444 real *shiftvec,*fshift,*x,*f;
445 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
447 __m128 fscal,rcutoff,rcutoff2,jidxall;
449 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
450 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
451 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
452 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
453 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
456 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
457 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
459 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
460 real rswitch_scalar,d_scalar;
461 __m128 dummy_mask,cutoff_mask;
462 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
463 __m128 one = _mm_set1_ps(1.0);
464 __m128 two = _mm_set1_ps(2.0);
470 jindex = nlist->jindex;
472 shiftidx = nlist->shift;
474 shiftvec = fr->shift_vec[0];
475 fshift = fr->fshift[0];
476 facel = _mm_set1_ps(fr->epsfac);
477 charge = mdatoms->chargeA;
479 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
480 beta = _mm_set1_ps(fr->ic->ewaldcoeff_q);
481 beta2 = _mm_mul_ps(beta,beta);
482 beta3 = _mm_mul_ps(beta,beta2);
483 ewtab = fr->ic->tabq_coul_FDV0;
484 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
485 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
487 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
488 rcutoff_scalar = fr->rcoulomb;
489 rcutoff = _mm_set1_ps(rcutoff_scalar);
490 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
492 rswitch_scalar = fr->rcoulomb_switch;
493 rswitch = _mm_set1_ps(rswitch_scalar);
494 /* Setup switch parameters */
495 d_scalar = rcutoff_scalar-rswitch_scalar;
496 d = _mm_set1_ps(d_scalar);
497 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
498 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
499 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
500 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
501 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
502 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
504 /* Avoid stupid compiler warnings */
505 jnrA = jnrB = jnrC = jnrD = 0;
514 for(iidx=0;iidx<4*DIM;iidx++)
519 /* Start outer loop over neighborlists */
520 for(iidx=0; iidx<nri; iidx++)
522 /* Load shift vector for this list */
523 i_shift_offset = DIM*shiftidx[iidx];
525 /* Load limits for loop over neighbors */
526 j_index_start = jindex[iidx];
527 j_index_end = jindex[iidx+1];
529 /* Get outer coordinate index */
531 i_coord_offset = DIM*inr;
533 /* Load i particle coords and add shift vector */
534 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
536 fix0 = _mm_setzero_ps();
537 fiy0 = _mm_setzero_ps();
538 fiz0 = _mm_setzero_ps();
540 /* Load parameters for i particles */
541 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
543 /* Start inner kernel loop */
544 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
547 /* Get j neighbor index, and coordinate index */
552 j_coord_offsetA = DIM*jnrA;
553 j_coord_offsetB = DIM*jnrB;
554 j_coord_offsetC = DIM*jnrC;
555 j_coord_offsetD = DIM*jnrD;
557 /* load j atom coordinates */
558 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
559 x+j_coord_offsetC,x+j_coord_offsetD,
562 /* Calculate displacement vector */
563 dx00 = _mm_sub_ps(ix0,jx0);
564 dy00 = _mm_sub_ps(iy0,jy0);
565 dz00 = _mm_sub_ps(iz0,jz0);
567 /* Calculate squared distance and things based on it */
568 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
570 rinv00 = gmx_mm_invsqrt_ps(rsq00);
572 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
574 /* Load parameters for j particles */
575 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
576 charge+jnrC+0,charge+jnrD+0);
578 /**************************
579 * CALCULATE INTERACTIONS *
580 **************************/
582 if (gmx_mm_any_lt(rsq00,rcutoff2))
585 r00 = _mm_mul_ps(rsq00,rinv00);
587 /* Compute parameters for interactions between i and j atoms */
588 qq00 = _mm_mul_ps(iq0,jq0);
590 /* EWALD ELECTROSTATICS */
592 /* Analytical PME correction */
593 zeta2 = _mm_mul_ps(beta2,rsq00);
594 rinv3 = _mm_mul_ps(rinvsq00,rinv00);
595 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
596 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
597 felec = _mm_mul_ps(qq00,felec);
598 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
599 velec = _mm_nmacc_ps(pmecorrV,beta,rinv00);
600 velec = _mm_mul_ps(qq00,velec);
602 d = _mm_sub_ps(r00,rswitch);
603 d = _mm_max_ps(d,_mm_setzero_ps());
604 d2 = _mm_mul_ps(d,d);
605 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
607 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
609 /* Evaluate switch function */
610 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
611 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
612 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
616 fscal = _mm_and_ps(fscal,cutoff_mask);
618 /* Update vectorial force */
619 fix0 = _mm_macc_ps(dx00,fscal,fix0);
620 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
621 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
623 fjptrA = f+j_coord_offsetA;
624 fjptrB = f+j_coord_offsetB;
625 fjptrC = f+j_coord_offsetC;
626 fjptrD = f+j_coord_offsetD;
627 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
628 _mm_mul_ps(dx00,fscal),
629 _mm_mul_ps(dy00,fscal),
630 _mm_mul_ps(dz00,fscal));
634 /* Inner loop uses 50 flops */
640 /* Get j neighbor index, and coordinate index */
641 jnrlistA = jjnr[jidx];
642 jnrlistB = jjnr[jidx+1];
643 jnrlistC = jjnr[jidx+2];
644 jnrlistD = jjnr[jidx+3];
645 /* Sign of each element will be negative for non-real atoms.
646 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
647 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
649 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
650 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
651 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
652 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
653 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
654 j_coord_offsetA = DIM*jnrA;
655 j_coord_offsetB = DIM*jnrB;
656 j_coord_offsetC = DIM*jnrC;
657 j_coord_offsetD = DIM*jnrD;
659 /* load j atom coordinates */
660 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
661 x+j_coord_offsetC,x+j_coord_offsetD,
664 /* Calculate displacement vector */
665 dx00 = _mm_sub_ps(ix0,jx0);
666 dy00 = _mm_sub_ps(iy0,jy0);
667 dz00 = _mm_sub_ps(iz0,jz0);
669 /* Calculate squared distance and things based on it */
670 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
672 rinv00 = gmx_mm_invsqrt_ps(rsq00);
674 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
676 /* Load parameters for j particles */
677 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
678 charge+jnrC+0,charge+jnrD+0);
680 /**************************
681 * CALCULATE INTERACTIONS *
682 **************************/
684 if (gmx_mm_any_lt(rsq00,rcutoff2))
687 r00 = _mm_mul_ps(rsq00,rinv00);
688 r00 = _mm_andnot_ps(dummy_mask,r00);
690 /* Compute parameters for interactions between i and j atoms */
691 qq00 = _mm_mul_ps(iq0,jq0);
693 /* EWALD ELECTROSTATICS */
695 /* Analytical PME correction */
696 zeta2 = _mm_mul_ps(beta2,rsq00);
697 rinv3 = _mm_mul_ps(rinvsq00,rinv00);
698 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
699 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
700 felec = _mm_mul_ps(qq00,felec);
701 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
702 velec = _mm_nmacc_ps(pmecorrV,beta,rinv00);
703 velec = _mm_mul_ps(qq00,velec);
705 d = _mm_sub_ps(r00,rswitch);
706 d = _mm_max_ps(d,_mm_setzero_ps());
707 d2 = _mm_mul_ps(d,d);
708 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
710 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
712 /* Evaluate switch function */
713 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
714 felec = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
715 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
719 fscal = _mm_and_ps(fscal,cutoff_mask);
721 fscal = _mm_andnot_ps(dummy_mask,fscal);
723 /* Update vectorial force */
724 fix0 = _mm_macc_ps(dx00,fscal,fix0);
725 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
726 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
728 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
729 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
730 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
731 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
732 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
733 _mm_mul_ps(dx00,fscal),
734 _mm_mul_ps(dy00,fscal),
735 _mm_mul_ps(dz00,fscal));
739 /* Inner loop uses 51 flops */
742 /* End of innermost loop */
744 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
745 f+i_coord_offset,fshift+i_shift_offset);
747 /* Increment number of inner iterations */
748 inneriter += j_index_end - j_index_start;
750 /* Outer loop uses 7 flops */
753 /* Increment number of outer iterations */
756 /* Update outer/inner flops */
758 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*51);