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_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
46 #include "gromacs/math/vec.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
50 #include "kernelutil_x86_avx_128_fma_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_128_fma_double
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_double
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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
84 int vdwjidx0A,vdwjidx0B;
85 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
86 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
87 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
90 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
92 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
93 real rswitch_scalar,d_scalar;
94 __m128d dummy_mask,cutoff_mask;
95 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
96 __m128d one = _mm_set1_pd(1.0);
97 __m128d two = _mm_set1_pd(2.0);
103 jindex = nlist->jindex;
105 shiftidx = nlist->shift;
107 shiftvec = fr->shift_vec[0];
108 fshift = fr->fshift[0];
109 facel = _mm_set1_pd(fr->epsfac);
110 charge = mdatoms->chargeA;
112 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
113 ewtab = fr->ic->tabq_coul_FDV0;
114 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
115 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
117 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
118 rcutoff_scalar = fr->rcoulomb;
119 rcutoff = _mm_set1_pd(rcutoff_scalar);
120 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
122 rswitch_scalar = fr->rcoulomb_switch;
123 rswitch = _mm_set1_pd(rswitch_scalar);
124 /* Setup switch parameters */
125 d_scalar = rcutoff_scalar-rswitch_scalar;
126 d = _mm_set1_pd(d_scalar);
127 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
128 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
129 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
130 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
131 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
132 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
134 /* Avoid stupid compiler warnings */
142 /* Start outer loop over neighborlists */
143 for(iidx=0; iidx<nri; iidx++)
145 /* Load shift vector for this list */
146 i_shift_offset = DIM*shiftidx[iidx];
148 /* Load limits for loop over neighbors */
149 j_index_start = jindex[iidx];
150 j_index_end = jindex[iidx+1];
152 /* Get outer coordinate index */
154 i_coord_offset = DIM*inr;
156 /* Load i particle coords and add shift vector */
157 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
159 fix0 = _mm_setzero_pd();
160 fiy0 = _mm_setzero_pd();
161 fiz0 = _mm_setzero_pd();
163 /* Load parameters for i particles */
164 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
166 /* Reset potential sums */
167 velecsum = _mm_setzero_pd();
169 /* Start inner kernel loop */
170 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
173 /* Get j neighbor index, and coordinate index */
176 j_coord_offsetA = DIM*jnrA;
177 j_coord_offsetB = DIM*jnrB;
179 /* load j atom coordinates */
180 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
183 /* Calculate displacement vector */
184 dx00 = _mm_sub_pd(ix0,jx0);
185 dy00 = _mm_sub_pd(iy0,jy0);
186 dz00 = _mm_sub_pd(iz0,jz0);
188 /* Calculate squared distance and things based on it */
189 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
191 rinv00 = gmx_mm_invsqrt_pd(rsq00);
193 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
195 /* Load parameters for j particles */
196 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
198 /**************************
199 * CALCULATE INTERACTIONS *
200 **************************/
202 if (gmx_mm_any_lt(rsq00,rcutoff2))
205 r00 = _mm_mul_pd(rsq00,rinv00);
207 /* Compute parameters for interactions between i and j atoms */
208 qq00 = _mm_mul_pd(iq0,jq0);
210 /* EWALD ELECTROSTATICS */
212 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
213 ewrt = _mm_mul_pd(r00,ewtabscale);
214 ewitab = _mm_cvttpd_epi32(ewrt);
216 eweps = _mm_frcz_pd(ewrt);
218 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
220 twoeweps = _mm_add_pd(eweps,eweps);
221 ewitab = _mm_slli_epi32(ewitab,2);
222 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
223 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
224 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
225 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
226 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
227 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
228 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
229 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
230 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
231 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
233 d = _mm_sub_pd(r00,rswitch);
234 d = _mm_max_pd(d,_mm_setzero_pd());
235 d2 = _mm_mul_pd(d,d);
236 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
238 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
240 /* Evaluate switch function */
241 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
242 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
243 velec = _mm_mul_pd(velec,sw);
244 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
246 /* Update potential sum for this i atom from the interaction with this j atom. */
247 velec = _mm_and_pd(velec,cutoff_mask);
248 velecsum = _mm_add_pd(velecsum,velec);
252 fscal = _mm_and_pd(fscal,cutoff_mask);
254 /* Update vectorial force */
255 fix0 = _mm_macc_pd(dx00,fscal,fix0);
256 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
257 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
259 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
260 _mm_mul_pd(dx00,fscal),
261 _mm_mul_pd(dy00,fscal),
262 _mm_mul_pd(dz00,fscal));
266 /* Inner loop uses 68 flops */
273 j_coord_offsetA = DIM*jnrA;
275 /* load j atom coordinates */
276 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
279 /* Calculate displacement vector */
280 dx00 = _mm_sub_pd(ix0,jx0);
281 dy00 = _mm_sub_pd(iy0,jy0);
282 dz00 = _mm_sub_pd(iz0,jz0);
284 /* Calculate squared distance and things based on it */
285 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
287 rinv00 = gmx_mm_invsqrt_pd(rsq00);
289 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
291 /* Load parameters for j particles */
292 jq0 = _mm_load_sd(charge+jnrA+0);
294 /**************************
295 * CALCULATE INTERACTIONS *
296 **************************/
298 if (gmx_mm_any_lt(rsq00,rcutoff2))
301 r00 = _mm_mul_pd(rsq00,rinv00);
303 /* Compute parameters for interactions between i and j atoms */
304 qq00 = _mm_mul_pd(iq0,jq0);
306 /* EWALD ELECTROSTATICS */
308 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
309 ewrt = _mm_mul_pd(r00,ewtabscale);
310 ewitab = _mm_cvttpd_epi32(ewrt);
312 eweps = _mm_frcz_pd(ewrt);
314 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
316 twoeweps = _mm_add_pd(eweps,eweps);
317 ewitab = _mm_slli_epi32(ewitab,2);
318 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
319 ewtabD = _mm_setzero_pd();
320 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
321 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
322 ewtabFn = _mm_setzero_pd();
323 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
324 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
325 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
326 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
327 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
329 d = _mm_sub_pd(r00,rswitch);
330 d = _mm_max_pd(d,_mm_setzero_pd());
331 d2 = _mm_mul_pd(d,d);
332 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
334 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
336 /* Evaluate switch function */
337 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
338 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
339 velec = _mm_mul_pd(velec,sw);
340 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
342 /* Update potential sum for this i atom from the interaction with this j atom. */
343 velec = _mm_and_pd(velec,cutoff_mask);
344 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
345 velecsum = _mm_add_pd(velecsum,velec);
349 fscal = _mm_and_pd(fscal,cutoff_mask);
351 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
353 /* Update vectorial force */
354 fix0 = _mm_macc_pd(dx00,fscal,fix0);
355 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
356 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
358 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
359 _mm_mul_pd(dx00,fscal),
360 _mm_mul_pd(dy00,fscal),
361 _mm_mul_pd(dz00,fscal));
365 /* Inner loop uses 68 flops */
368 /* End of innermost loop */
370 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
371 f+i_coord_offset,fshift+i_shift_offset);
374 /* Update potential energies */
375 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
377 /* Increment number of inner iterations */
378 inneriter += j_index_end - j_index_start;
380 /* Outer loop uses 8 flops */
383 /* Increment number of outer iterations */
386 /* Update outer/inner flops */
388 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*68);
391 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_128_fma_double
392 * Electrostatics interaction: Ewald
393 * VdW interaction: None
394 * Geometry: Particle-Particle
395 * Calculate force/pot: Force
398 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_128_fma_double
399 (t_nblist * gmx_restrict nlist,
400 rvec * gmx_restrict xx,
401 rvec * gmx_restrict ff,
402 t_forcerec * gmx_restrict fr,
403 t_mdatoms * gmx_restrict mdatoms,
404 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
405 t_nrnb * gmx_restrict nrnb)
407 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
408 * just 0 for non-waters.
409 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
410 * jnr indices corresponding to data put in the four positions in the SIMD register.
412 int i_shift_offset,i_coord_offset,outeriter,inneriter;
413 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
415 int j_coord_offsetA,j_coord_offsetB;
416 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
418 real *shiftvec,*fshift,*x,*f;
419 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
421 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
422 int vdwjidx0A,vdwjidx0B;
423 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
424 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
425 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
428 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
430 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
431 real rswitch_scalar,d_scalar;
432 __m128d dummy_mask,cutoff_mask;
433 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
434 __m128d one = _mm_set1_pd(1.0);
435 __m128d two = _mm_set1_pd(2.0);
441 jindex = nlist->jindex;
443 shiftidx = nlist->shift;
445 shiftvec = fr->shift_vec[0];
446 fshift = fr->fshift[0];
447 facel = _mm_set1_pd(fr->epsfac);
448 charge = mdatoms->chargeA;
450 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
451 ewtab = fr->ic->tabq_coul_FDV0;
452 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
453 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
455 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
456 rcutoff_scalar = fr->rcoulomb;
457 rcutoff = _mm_set1_pd(rcutoff_scalar);
458 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
460 rswitch_scalar = fr->rcoulomb_switch;
461 rswitch = _mm_set1_pd(rswitch_scalar);
462 /* Setup switch parameters */
463 d_scalar = rcutoff_scalar-rswitch_scalar;
464 d = _mm_set1_pd(d_scalar);
465 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
466 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
467 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
468 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
469 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
470 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
472 /* Avoid stupid compiler warnings */
480 /* Start outer loop over neighborlists */
481 for(iidx=0; iidx<nri; iidx++)
483 /* Load shift vector for this list */
484 i_shift_offset = DIM*shiftidx[iidx];
486 /* Load limits for loop over neighbors */
487 j_index_start = jindex[iidx];
488 j_index_end = jindex[iidx+1];
490 /* Get outer coordinate index */
492 i_coord_offset = DIM*inr;
494 /* Load i particle coords and add shift vector */
495 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
497 fix0 = _mm_setzero_pd();
498 fiy0 = _mm_setzero_pd();
499 fiz0 = _mm_setzero_pd();
501 /* Load parameters for i particles */
502 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
504 /* Start inner kernel loop */
505 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
508 /* Get j neighbor index, and coordinate index */
511 j_coord_offsetA = DIM*jnrA;
512 j_coord_offsetB = DIM*jnrB;
514 /* load j atom coordinates */
515 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
518 /* Calculate displacement vector */
519 dx00 = _mm_sub_pd(ix0,jx0);
520 dy00 = _mm_sub_pd(iy0,jy0);
521 dz00 = _mm_sub_pd(iz0,jz0);
523 /* Calculate squared distance and things based on it */
524 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
526 rinv00 = gmx_mm_invsqrt_pd(rsq00);
528 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
530 /* Load parameters for j particles */
531 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
533 /**************************
534 * CALCULATE INTERACTIONS *
535 **************************/
537 if (gmx_mm_any_lt(rsq00,rcutoff2))
540 r00 = _mm_mul_pd(rsq00,rinv00);
542 /* Compute parameters for interactions between i and j atoms */
543 qq00 = _mm_mul_pd(iq0,jq0);
545 /* EWALD ELECTROSTATICS */
547 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
548 ewrt = _mm_mul_pd(r00,ewtabscale);
549 ewitab = _mm_cvttpd_epi32(ewrt);
551 eweps = _mm_frcz_pd(ewrt);
553 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
555 twoeweps = _mm_add_pd(eweps,eweps);
556 ewitab = _mm_slli_epi32(ewitab,2);
557 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
558 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
559 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
560 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
561 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
562 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
563 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
564 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
565 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
566 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
568 d = _mm_sub_pd(r00,rswitch);
569 d = _mm_max_pd(d,_mm_setzero_pd());
570 d2 = _mm_mul_pd(d,d);
571 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
573 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
575 /* Evaluate switch function */
576 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
577 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
578 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
582 fscal = _mm_and_pd(fscal,cutoff_mask);
584 /* Update vectorial force */
585 fix0 = _mm_macc_pd(dx00,fscal,fix0);
586 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
587 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
589 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
590 _mm_mul_pd(dx00,fscal),
591 _mm_mul_pd(dy00,fscal),
592 _mm_mul_pd(dz00,fscal));
596 /* Inner loop uses 65 flops */
603 j_coord_offsetA = DIM*jnrA;
605 /* load j atom coordinates */
606 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
609 /* Calculate displacement vector */
610 dx00 = _mm_sub_pd(ix0,jx0);
611 dy00 = _mm_sub_pd(iy0,jy0);
612 dz00 = _mm_sub_pd(iz0,jz0);
614 /* Calculate squared distance and things based on it */
615 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
617 rinv00 = gmx_mm_invsqrt_pd(rsq00);
619 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
621 /* Load parameters for j particles */
622 jq0 = _mm_load_sd(charge+jnrA+0);
624 /**************************
625 * CALCULATE INTERACTIONS *
626 **************************/
628 if (gmx_mm_any_lt(rsq00,rcutoff2))
631 r00 = _mm_mul_pd(rsq00,rinv00);
633 /* Compute parameters for interactions between i and j atoms */
634 qq00 = _mm_mul_pd(iq0,jq0);
636 /* EWALD ELECTROSTATICS */
638 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
639 ewrt = _mm_mul_pd(r00,ewtabscale);
640 ewitab = _mm_cvttpd_epi32(ewrt);
642 eweps = _mm_frcz_pd(ewrt);
644 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
646 twoeweps = _mm_add_pd(eweps,eweps);
647 ewitab = _mm_slli_epi32(ewitab,2);
648 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
649 ewtabD = _mm_setzero_pd();
650 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
651 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
652 ewtabFn = _mm_setzero_pd();
653 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
654 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
655 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
656 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
657 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
659 d = _mm_sub_pd(r00,rswitch);
660 d = _mm_max_pd(d,_mm_setzero_pd());
661 d2 = _mm_mul_pd(d,d);
662 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
664 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
666 /* Evaluate switch function */
667 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
668 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
669 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
673 fscal = _mm_and_pd(fscal,cutoff_mask);
675 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
677 /* Update vectorial force */
678 fix0 = _mm_macc_pd(dx00,fscal,fix0);
679 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
680 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
682 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
683 _mm_mul_pd(dx00,fscal),
684 _mm_mul_pd(dy00,fscal),
685 _mm_mul_pd(dz00,fscal));
689 /* Inner loop uses 65 flops */
692 /* End of innermost loop */
694 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
695 f+i_coord_offset,fshift+i_shift_offset);
697 /* Increment number of inner iterations */
698 inneriter += j_index_end - j_index_start;
700 /* Outer loop uses 7 flops */
703 /* Increment number of outer iterations */
706 /* Update outer/inner flops */
708 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*65);