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.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW4P1_VF_avx_128_fma_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: None
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_VF_avx_128_fma_double
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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
83 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
85 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
86 int vdwjidx0A,vdwjidx0B;
87 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
88 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
89 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
90 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
91 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
94 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
96 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
97 real rswitch_scalar,d_scalar;
98 __m128d dummy_mask,cutoff_mask;
99 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
100 __m128d one = _mm_set1_pd(1.0);
101 __m128d two = _mm_set1_pd(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_pd(fr->epsfac);
114 charge = mdatoms->chargeA;
116 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
117 ewtab = fr->ic->tabq_coul_FDV0;
118 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
119 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
121 /* Setup water-specific parameters */
122 inr = nlist->iinr[0];
123 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
124 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
125 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
127 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
128 rcutoff_scalar = fr->rcoulomb;
129 rcutoff = _mm_set1_pd(rcutoff_scalar);
130 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
132 rswitch_scalar = fr->rcoulomb_switch;
133 rswitch = _mm_set1_pd(rswitch_scalar);
134 /* Setup switch parameters */
135 d_scalar = rcutoff_scalar-rswitch_scalar;
136 d = _mm_set1_pd(d_scalar);
137 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
138 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
139 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
140 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
141 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
142 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
144 /* Avoid stupid compiler warnings */
152 /* Start outer loop over neighborlists */
153 for(iidx=0; iidx<nri; iidx++)
155 /* Load shift vector for this list */
156 i_shift_offset = DIM*shiftidx[iidx];
158 /* Load limits for loop over neighbors */
159 j_index_start = jindex[iidx];
160 j_index_end = jindex[iidx+1];
162 /* Get outer coordinate index */
164 i_coord_offset = DIM*inr;
166 /* Load i particle coords and add shift vector */
167 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
168 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
170 fix1 = _mm_setzero_pd();
171 fiy1 = _mm_setzero_pd();
172 fiz1 = _mm_setzero_pd();
173 fix2 = _mm_setzero_pd();
174 fiy2 = _mm_setzero_pd();
175 fiz2 = _mm_setzero_pd();
176 fix3 = _mm_setzero_pd();
177 fiy3 = _mm_setzero_pd();
178 fiz3 = _mm_setzero_pd();
180 /* Reset potential sums */
181 velecsum = _mm_setzero_pd();
183 /* Start inner kernel loop */
184 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
187 /* Get j neighbor index, and coordinate index */
190 j_coord_offsetA = DIM*jnrA;
191 j_coord_offsetB = DIM*jnrB;
193 /* load j atom coordinates */
194 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
197 /* Calculate displacement vector */
198 dx10 = _mm_sub_pd(ix1,jx0);
199 dy10 = _mm_sub_pd(iy1,jy0);
200 dz10 = _mm_sub_pd(iz1,jz0);
201 dx20 = _mm_sub_pd(ix2,jx0);
202 dy20 = _mm_sub_pd(iy2,jy0);
203 dz20 = _mm_sub_pd(iz2,jz0);
204 dx30 = _mm_sub_pd(ix3,jx0);
205 dy30 = _mm_sub_pd(iy3,jy0);
206 dz30 = _mm_sub_pd(iz3,jz0);
208 /* Calculate squared distance and things based on it */
209 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
210 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
211 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
213 rinv10 = gmx_mm_invsqrt_pd(rsq10);
214 rinv20 = gmx_mm_invsqrt_pd(rsq20);
215 rinv30 = gmx_mm_invsqrt_pd(rsq30);
217 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
218 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
219 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
221 /* Load parameters for j particles */
222 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
224 fjx0 = _mm_setzero_pd();
225 fjy0 = _mm_setzero_pd();
226 fjz0 = _mm_setzero_pd();
228 /**************************
229 * CALCULATE INTERACTIONS *
230 **************************/
232 if (gmx_mm_any_lt(rsq10,rcutoff2))
235 r10 = _mm_mul_pd(rsq10,rinv10);
237 /* Compute parameters for interactions between i and j atoms */
238 qq10 = _mm_mul_pd(iq1,jq0);
240 /* EWALD ELECTROSTATICS */
242 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
243 ewrt = _mm_mul_pd(r10,ewtabscale);
244 ewitab = _mm_cvttpd_epi32(ewrt);
246 eweps = _mm_frcz_pd(ewrt);
248 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
250 twoeweps = _mm_add_pd(eweps,eweps);
251 ewitab = _mm_slli_epi32(ewitab,2);
252 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
253 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
254 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
255 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
256 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
257 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
258 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
259 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
260 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
261 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
263 d = _mm_sub_pd(r10,rswitch);
264 d = _mm_max_pd(d,_mm_setzero_pd());
265 d2 = _mm_mul_pd(d,d);
266 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
268 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
270 /* Evaluate switch function */
271 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
272 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
273 velec = _mm_mul_pd(velec,sw);
274 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
276 /* Update potential sum for this i atom from the interaction with this j atom. */
277 velec = _mm_and_pd(velec,cutoff_mask);
278 velecsum = _mm_add_pd(velecsum,velec);
282 fscal = _mm_and_pd(fscal,cutoff_mask);
284 /* Update vectorial force */
285 fix1 = _mm_macc_pd(dx10,fscal,fix1);
286 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
287 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
289 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
290 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
291 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
295 /**************************
296 * CALCULATE INTERACTIONS *
297 **************************/
299 if (gmx_mm_any_lt(rsq20,rcutoff2))
302 r20 = _mm_mul_pd(rsq20,rinv20);
304 /* Compute parameters for interactions between i and j atoms */
305 qq20 = _mm_mul_pd(iq2,jq0);
307 /* EWALD ELECTROSTATICS */
309 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
310 ewrt = _mm_mul_pd(r20,ewtabscale);
311 ewitab = _mm_cvttpd_epi32(ewrt);
313 eweps = _mm_frcz_pd(ewrt);
315 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
317 twoeweps = _mm_add_pd(eweps,eweps);
318 ewitab = _mm_slli_epi32(ewitab,2);
319 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
320 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
321 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
322 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
323 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
324 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
325 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
326 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
327 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
328 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
330 d = _mm_sub_pd(r20,rswitch);
331 d = _mm_max_pd(d,_mm_setzero_pd());
332 d2 = _mm_mul_pd(d,d);
333 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
335 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
337 /* Evaluate switch function */
338 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
339 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
340 velec = _mm_mul_pd(velec,sw);
341 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
343 /* Update potential sum for this i atom from the interaction with this j atom. */
344 velec = _mm_and_pd(velec,cutoff_mask);
345 velecsum = _mm_add_pd(velecsum,velec);
349 fscal = _mm_and_pd(fscal,cutoff_mask);
351 /* Update vectorial force */
352 fix2 = _mm_macc_pd(dx20,fscal,fix2);
353 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
354 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
356 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
357 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
358 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
362 /**************************
363 * CALCULATE INTERACTIONS *
364 **************************/
366 if (gmx_mm_any_lt(rsq30,rcutoff2))
369 r30 = _mm_mul_pd(rsq30,rinv30);
371 /* Compute parameters for interactions between i and j atoms */
372 qq30 = _mm_mul_pd(iq3,jq0);
374 /* EWALD ELECTROSTATICS */
376 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
377 ewrt = _mm_mul_pd(r30,ewtabscale);
378 ewitab = _mm_cvttpd_epi32(ewrt);
380 eweps = _mm_frcz_pd(ewrt);
382 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
384 twoeweps = _mm_add_pd(eweps,eweps);
385 ewitab = _mm_slli_epi32(ewitab,2);
386 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
387 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
388 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
389 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
390 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
391 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
392 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
393 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
394 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
395 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
397 d = _mm_sub_pd(r30,rswitch);
398 d = _mm_max_pd(d,_mm_setzero_pd());
399 d2 = _mm_mul_pd(d,d);
400 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
402 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
404 /* Evaluate switch function */
405 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
406 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
407 velec = _mm_mul_pd(velec,sw);
408 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
410 /* Update potential sum for this i atom from the interaction with this j atom. */
411 velec = _mm_and_pd(velec,cutoff_mask);
412 velecsum = _mm_add_pd(velecsum,velec);
416 fscal = _mm_and_pd(fscal,cutoff_mask);
418 /* Update vectorial force */
419 fix3 = _mm_macc_pd(dx30,fscal,fix3);
420 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
421 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
423 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
424 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
425 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
429 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
431 /* Inner loop uses 207 flops */
438 j_coord_offsetA = DIM*jnrA;
440 /* load j atom coordinates */
441 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
444 /* Calculate displacement vector */
445 dx10 = _mm_sub_pd(ix1,jx0);
446 dy10 = _mm_sub_pd(iy1,jy0);
447 dz10 = _mm_sub_pd(iz1,jz0);
448 dx20 = _mm_sub_pd(ix2,jx0);
449 dy20 = _mm_sub_pd(iy2,jy0);
450 dz20 = _mm_sub_pd(iz2,jz0);
451 dx30 = _mm_sub_pd(ix3,jx0);
452 dy30 = _mm_sub_pd(iy3,jy0);
453 dz30 = _mm_sub_pd(iz3,jz0);
455 /* Calculate squared distance and things based on it */
456 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
457 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
458 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
460 rinv10 = gmx_mm_invsqrt_pd(rsq10);
461 rinv20 = gmx_mm_invsqrt_pd(rsq20);
462 rinv30 = gmx_mm_invsqrt_pd(rsq30);
464 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
465 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
466 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
468 /* Load parameters for j particles */
469 jq0 = _mm_load_sd(charge+jnrA+0);
471 fjx0 = _mm_setzero_pd();
472 fjy0 = _mm_setzero_pd();
473 fjz0 = _mm_setzero_pd();
475 /**************************
476 * CALCULATE INTERACTIONS *
477 **************************/
479 if (gmx_mm_any_lt(rsq10,rcutoff2))
482 r10 = _mm_mul_pd(rsq10,rinv10);
484 /* Compute parameters for interactions between i and j atoms */
485 qq10 = _mm_mul_pd(iq1,jq0);
487 /* EWALD ELECTROSTATICS */
489 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
490 ewrt = _mm_mul_pd(r10,ewtabscale);
491 ewitab = _mm_cvttpd_epi32(ewrt);
493 eweps = _mm_frcz_pd(ewrt);
495 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
497 twoeweps = _mm_add_pd(eweps,eweps);
498 ewitab = _mm_slli_epi32(ewitab,2);
499 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
500 ewtabD = _mm_setzero_pd();
501 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
502 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
503 ewtabFn = _mm_setzero_pd();
504 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
505 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
506 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
507 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
508 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
510 d = _mm_sub_pd(r10,rswitch);
511 d = _mm_max_pd(d,_mm_setzero_pd());
512 d2 = _mm_mul_pd(d,d);
513 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
515 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
517 /* Evaluate switch function */
518 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
519 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
520 velec = _mm_mul_pd(velec,sw);
521 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
523 /* Update potential sum for this i atom from the interaction with this j atom. */
524 velec = _mm_and_pd(velec,cutoff_mask);
525 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
526 velecsum = _mm_add_pd(velecsum,velec);
530 fscal = _mm_and_pd(fscal,cutoff_mask);
532 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
534 /* Update vectorial force */
535 fix1 = _mm_macc_pd(dx10,fscal,fix1);
536 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
537 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
539 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
540 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
541 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
545 /**************************
546 * CALCULATE INTERACTIONS *
547 **************************/
549 if (gmx_mm_any_lt(rsq20,rcutoff2))
552 r20 = _mm_mul_pd(rsq20,rinv20);
554 /* Compute parameters for interactions between i and j atoms */
555 qq20 = _mm_mul_pd(iq2,jq0);
557 /* EWALD ELECTROSTATICS */
559 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
560 ewrt = _mm_mul_pd(r20,ewtabscale);
561 ewitab = _mm_cvttpd_epi32(ewrt);
563 eweps = _mm_frcz_pd(ewrt);
565 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
567 twoeweps = _mm_add_pd(eweps,eweps);
568 ewitab = _mm_slli_epi32(ewitab,2);
569 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
570 ewtabD = _mm_setzero_pd();
571 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
572 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
573 ewtabFn = _mm_setzero_pd();
574 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
575 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
576 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
577 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
578 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
580 d = _mm_sub_pd(r20,rswitch);
581 d = _mm_max_pd(d,_mm_setzero_pd());
582 d2 = _mm_mul_pd(d,d);
583 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
585 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
587 /* Evaluate switch function */
588 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
589 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
590 velec = _mm_mul_pd(velec,sw);
591 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
593 /* Update potential sum for this i atom from the interaction with this j atom. */
594 velec = _mm_and_pd(velec,cutoff_mask);
595 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
596 velecsum = _mm_add_pd(velecsum,velec);
600 fscal = _mm_and_pd(fscal,cutoff_mask);
602 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
604 /* Update vectorial force */
605 fix2 = _mm_macc_pd(dx20,fscal,fix2);
606 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
607 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
609 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
610 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
611 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
615 /**************************
616 * CALCULATE INTERACTIONS *
617 **************************/
619 if (gmx_mm_any_lt(rsq30,rcutoff2))
622 r30 = _mm_mul_pd(rsq30,rinv30);
624 /* Compute parameters for interactions between i and j atoms */
625 qq30 = _mm_mul_pd(iq3,jq0);
627 /* EWALD ELECTROSTATICS */
629 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
630 ewrt = _mm_mul_pd(r30,ewtabscale);
631 ewitab = _mm_cvttpd_epi32(ewrt);
633 eweps = _mm_frcz_pd(ewrt);
635 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
637 twoeweps = _mm_add_pd(eweps,eweps);
638 ewitab = _mm_slli_epi32(ewitab,2);
639 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
640 ewtabD = _mm_setzero_pd();
641 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
642 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
643 ewtabFn = _mm_setzero_pd();
644 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
645 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
646 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
647 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
648 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
650 d = _mm_sub_pd(r30,rswitch);
651 d = _mm_max_pd(d,_mm_setzero_pd());
652 d2 = _mm_mul_pd(d,d);
653 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
655 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
657 /* Evaluate switch function */
658 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
659 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
660 velec = _mm_mul_pd(velec,sw);
661 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
663 /* Update potential sum for this i atom from the interaction with this j atom. */
664 velec = _mm_and_pd(velec,cutoff_mask);
665 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
666 velecsum = _mm_add_pd(velecsum,velec);
670 fscal = _mm_and_pd(fscal,cutoff_mask);
672 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
674 /* Update vectorial force */
675 fix3 = _mm_macc_pd(dx30,fscal,fix3);
676 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
677 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
679 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
680 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
681 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
685 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
687 /* Inner loop uses 207 flops */
690 /* End of innermost loop */
692 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
693 f+i_coord_offset+DIM,fshift+i_shift_offset);
696 /* Update potential energies */
697 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
699 /* Increment number of inner iterations */
700 inneriter += j_index_end - j_index_start;
702 /* Outer loop uses 19 flops */
705 /* Increment number of outer iterations */
708 /* Update outer/inner flops */
710 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*207);
713 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_avx_128_fma_double
714 * Electrostatics interaction: Ewald
715 * VdW interaction: None
716 * Geometry: Water4-Particle
717 * Calculate force/pot: Force
720 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_avx_128_fma_double
721 (t_nblist * gmx_restrict nlist,
722 rvec * gmx_restrict xx,
723 rvec * gmx_restrict ff,
724 t_forcerec * gmx_restrict fr,
725 t_mdatoms * gmx_restrict mdatoms,
726 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
727 t_nrnb * gmx_restrict nrnb)
729 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
730 * just 0 for non-waters.
731 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
732 * jnr indices corresponding to data put in the four positions in the SIMD register.
734 int i_shift_offset,i_coord_offset,outeriter,inneriter;
735 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
737 int j_coord_offsetA,j_coord_offsetB;
738 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
740 real *shiftvec,*fshift,*x,*f;
741 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
743 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
745 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
747 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
748 int vdwjidx0A,vdwjidx0B;
749 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
750 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
751 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
752 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
753 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
756 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
758 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
759 real rswitch_scalar,d_scalar;
760 __m128d dummy_mask,cutoff_mask;
761 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
762 __m128d one = _mm_set1_pd(1.0);
763 __m128d two = _mm_set1_pd(2.0);
769 jindex = nlist->jindex;
771 shiftidx = nlist->shift;
773 shiftvec = fr->shift_vec[0];
774 fshift = fr->fshift[0];
775 facel = _mm_set1_pd(fr->epsfac);
776 charge = mdatoms->chargeA;
778 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
779 ewtab = fr->ic->tabq_coul_FDV0;
780 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
781 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
783 /* Setup water-specific parameters */
784 inr = nlist->iinr[0];
785 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
786 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
787 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
789 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
790 rcutoff_scalar = fr->rcoulomb;
791 rcutoff = _mm_set1_pd(rcutoff_scalar);
792 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
794 rswitch_scalar = fr->rcoulomb_switch;
795 rswitch = _mm_set1_pd(rswitch_scalar);
796 /* Setup switch parameters */
797 d_scalar = rcutoff_scalar-rswitch_scalar;
798 d = _mm_set1_pd(d_scalar);
799 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
800 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
801 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
802 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
803 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
804 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
806 /* Avoid stupid compiler warnings */
814 /* Start outer loop over neighborlists */
815 for(iidx=0; iidx<nri; iidx++)
817 /* Load shift vector for this list */
818 i_shift_offset = DIM*shiftidx[iidx];
820 /* Load limits for loop over neighbors */
821 j_index_start = jindex[iidx];
822 j_index_end = jindex[iidx+1];
824 /* Get outer coordinate index */
826 i_coord_offset = DIM*inr;
828 /* Load i particle coords and add shift vector */
829 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
830 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
832 fix1 = _mm_setzero_pd();
833 fiy1 = _mm_setzero_pd();
834 fiz1 = _mm_setzero_pd();
835 fix2 = _mm_setzero_pd();
836 fiy2 = _mm_setzero_pd();
837 fiz2 = _mm_setzero_pd();
838 fix3 = _mm_setzero_pd();
839 fiy3 = _mm_setzero_pd();
840 fiz3 = _mm_setzero_pd();
842 /* Start inner kernel loop */
843 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
846 /* Get j neighbor index, and coordinate index */
849 j_coord_offsetA = DIM*jnrA;
850 j_coord_offsetB = DIM*jnrB;
852 /* load j atom coordinates */
853 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
856 /* Calculate displacement vector */
857 dx10 = _mm_sub_pd(ix1,jx0);
858 dy10 = _mm_sub_pd(iy1,jy0);
859 dz10 = _mm_sub_pd(iz1,jz0);
860 dx20 = _mm_sub_pd(ix2,jx0);
861 dy20 = _mm_sub_pd(iy2,jy0);
862 dz20 = _mm_sub_pd(iz2,jz0);
863 dx30 = _mm_sub_pd(ix3,jx0);
864 dy30 = _mm_sub_pd(iy3,jy0);
865 dz30 = _mm_sub_pd(iz3,jz0);
867 /* Calculate squared distance and things based on it */
868 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
869 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
870 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
872 rinv10 = gmx_mm_invsqrt_pd(rsq10);
873 rinv20 = gmx_mm_invsqrt_pd(rsq20);
874 rinv30 = gmx_mm_invsqrt_pd(rsq30);
876 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
877 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
878 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
880 /* Load parameters for j particles */
881 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
883 fjx0 = _mm_setzero_pd();
884 fjy0 = _mm_setzero_pd();
885 fjz0 = _mm_setzero_pd();
887 /**************************
888 * CALCULATE INTERACTIONS *
889 **************************/
891 if (gmx_mm_any_lt(rsq10,rcutoff2))
894 r10 = _mm_mul_pd(rsq10,rinv10);
896 /* Compute parameters for interactions between i and j atoms */
897 qq10 = _mm_mul_pd(iq1,jq0);
899 /* EWALD ELECTROSTATICS */
901 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
902 ewrt = _mm_mul_pd(r10,ewtabscale);
903 ewitab = _mm_cvttpd_epi32(ewrt);
905 eweps = _mm_frcz_pd(ewrt);
907 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
909 twoeweps = _mm_add_pd(eweps,eweps);
910 ewitab = _mm_slli_epi32(ewitab,2);
911 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
912 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
913 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
914 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
915 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
916 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
917 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
918 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
919 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
920 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
922 d = _mm_sub_pd(r10,rswitch);
923 d = _mm_max_pd(d,_mm_setzero_pd());
924 d2 = _mm_mul_pd(d,d);
925 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
927 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
929 /* Evaluate switch function */
930 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
931 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
932 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
936 fscal = _mm_and_pd(fscal,cutoff_mask);
938 /* Update vectorial force */
939 fix1 = _mm_macc_pd(dx10,fscal,fix1);
940 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
941 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
943 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
944 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
945 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
949 /**************************
950 * CALCULATE INTERACTIONS *
951 **************************/
953 if (gmx_mm_any_lt(rsq20,rcutoff2))
956 r20 = _mm_mul_pd(rsq20,rinv20);
958 /* Compute parameters for interactions between i and j atoms */
959 qq20 = _mm_mul_pd(iq2,jq0);
961 /* EWALD ELECTROSTATICS */
963 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
964 ewrt = _mm_mul_pd(r20,ewtabscale);
965 ewitab = _mm_cvttpd_epi32(ewrt);
967 eweps = _mm_frcz_pd(ewrt);
969 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
971 twoeweps = _mm_add_pd(eweps,eweps);
972 ewitab = _mm_slli_epi32(ewitab,2);
973 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
974 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
975 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
976 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
977 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
978 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
979 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
980 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
981 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
982 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
984 d = _mm_sub_pd(r20,rswitch);
985 d = _mm_max_pd(d,_mm_setzero_pd());
986 d2 = _mm_mul_pd(d,d);
987 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
989 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
991 /* Evaluate switch function */
992 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
993 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
994 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
998 fscal = _mm_and_pd(fscal,cutoff_mask);
1000 /* Update vectorial force */
1001 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1002 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1003 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1005 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1006 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1007 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1011 /**************************
1012 * CALCULATE INTERACTIONS *
1013 **************************/
1015 if (gmx_mm_any_lt(rsq30,rcutoff2))
1018 r30 = _mm_mul_pd(rsq30,rinv30);
1020 /* Compute parameters for interactions between i and j atoms */
1021 qq30 = _mm_mul_pd(iq3,jq0);
1023 /* EWALD ELECTROSTATICS */
1025 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1026 ewrt = _mm_mul_pd(r30,ewtabscale);
1027 ewitab = _mm_cvttpd_epi32(ewrt);
1029 eweps = _mm_frcz_pd(ewrt);
1031 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1033 twoeweps = _mm_add_pd(eweps,eweps);
1034 ewitab = _mm_slli_epi32(ewitab,2);
1035 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1036 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1037 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1038 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1039 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1040 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1041 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1042 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1043 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1044 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1046 d = _mm_sub_pd(r30,rswitch);
1047 d = _mm_max_pd(d,_mm_setzero_pd());
1048 d2 = _mm_mul_pd(d,d);
1049 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1051 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1053 /* Evaluate switch function */
1054 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1055 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1056 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
1060 fscal = _mm_and_pd(fscal,cutoff_mask);
1062 /* Update vectorial force */
1063 fix3 = _mm_macc_pd(dx30,fscal,fix3);
1064 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
1065 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
1067 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
1068 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
1069 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
1073 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1075 /* Inner loop uses 198 flops */
1078 if(jidx<j_index_end)
1082 j_coord_offsetA = DIM*jnrA;
1084 /* load j atom coordinates */
1085 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1088 /* Calculate displacement vector */
1089 dx10 = _mm_sub_pd(ix1,jx0);
1090 dy10 = _mm_sub_pd(iy1,jy0);
1091 dz10 = _mm_sub_pd(iz1,jz0);
1092 dx20 = _mm_sub_pd(ix2,jx0);
1093 dy20 = _mm_sub_pd(iy2,jy0);
1094 dz20 = _mm_sub_pd(iz2,jz0);
1095 dx30 = _mm_sub_pd(ix3,jx0);
1096 dy30 = _mm_sub_pd(iy3,jy0);
1097 dz30 = _mm_sub_pd(iz3,jz0);
1099 /* Calculate squared distance and things based on it */
1100 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1101 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1102 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1104 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1105 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1106 rinv30 = gmx_mm_invsqrt_pd(rsq30);
1108 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1109 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1110 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
1112 /* Load parameters for j particles */
1113 jq0 = _mm_load_sd(charge+jnrA+0);
1115 fjx0 = _mm_setzero_pd();
1116 fjy0 = _mm_setzero_pd();
1117 fjz0 = _mm_setzero_pd();
1119 /**************************
1120 * CALCULATE INTERACTIONS *
1121 **************************/
1123 if (gmx_mm_any_lt(rsq10,rcutoff2))
1126 r10 = _mm_mul_pd(rsq10,rinv10);
1128 /* Compute parameters for interactions between i and j atoms */
1129 qq10 = _mm_mul_pd(iq1,jq0);
1131 /* EWALD ELECTROSTATICS */
1133 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1134 ewrt = _mm_mul_pd(r10,ewtabscale);
1135 ewitab = _mm_cvttpd_epi32(ewrt);
1137 eweps = _mm_frcz_pd(ewrt);
1139 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1141 twoeweps = _mm_add_pd(eweps,eweps);
1142 ewitab = _mm_slli_epi32(ewitab,2);
1143 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1144 ewtabD = _mm_setzero_pd();
1145 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1146 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1147 ewtabFn = _mm_setzero_pd();
1148 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1149 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1150 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1151 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1152 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1154 d = _mm_sub_pd(r10,rswitch);
1155 d = _mm_max_pd(d,_mm_setzero_pd());
1156 d2 = _mm_mul_pd(d,d);
1157 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1159 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1161 /* Evaluate switch function */
1162 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1163 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1164 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1168 fscal = _mm_and_pd(fscal,cutoff_mask);
1170 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1172 /* Update vectorial force */
1173 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1174 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1175 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1177 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1178 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1179 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1183 /**************************
1184 * CALCULATE INTERACTIONS *
1185 **************************/
1187 if (gmx_mm_any_lt(rsq20,rcutoff2))
1190 r20 = _mm_mul_pd(rsq20,rinv20);
1192 /* Compute parameters for interactions between i and j atoms */
1193 qq20 = _mm_mul_pd(iq2,jq0);
1195 /* EWALD ELECTROSTATICS */
1197 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1198 ewrt = _mm_mul_pd(r20,ewtabscale);
1199 ewitab = _mm_cvttpd_epi32(ewrt);
1201 eweps = _mm_frcz_pd(ewrt);
1203 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1205 twoeweps = _mm_add_pd(eweps,eweps);
1206 ewitab = _mm_slli_epi32(ewitab,2);
1207 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1208 ewtabD = _mm_setzero_pd();
1209 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1210 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1211 ewtabFn = _mm_setzero_pd();
1212 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1213 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1214 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1215 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1216 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1218 d = _mm_sub_pd(r20,rswitch);
1219 d = _mm_max_pd(d,_mm_setzero_pd());
1220 d2 = _mm_mul_pd(d,d);
1221 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1223 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1225 /* Evaluate switch function */
1226 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1227 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1228 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1232 fscal = _mm_and_pd(fscal,cutoff_mask);
1234 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1236 /* Update vectorial force */
1237 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1238 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1239 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1241 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1242 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1243 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1247 /**************************
1248 * CALCULATE INTERACTIONS *
1249 **************************/
1251 if (gmx_mm_any_lt(rsq30,rcutoff2))
1254 r30 = _mm_mul_pd(rsq30,rinv30);
1256 /* Compute parameters for interactions between i and j atoms */
1257 qq30 = _mm_mul_pd(iq3,jq0);
1259 /* EWALD ELECTROSTATICS */
1261 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1262 ewrt = _mm_mul_pd(r30,ewtabscale);
1263 ewitab = _mm_cvttpd_epi32(ewrt);
1265 eweps = _mm_frcz_pd(ewrt);
1267 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1269 twoeweps = _mm_add_pd(eweps,eweps);
1270 ewitab = _mm_slli_epi32(ewitab,2);
1271 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1272 ewtabD = _mm_setzero_pd();
1273 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1274 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1275 ewtabFn = _mm_setzero_pd();
1276 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1277 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1278 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1279 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1280 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1282 d = _mm_sub_pd(r30,rswitch);
1283 d = _mm_max_pd(d,_mm_setzero_pd());
1284 d2 = _mm_mul_pd(d,d);
1285 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1287 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1289 /* Evaluate switch function */
1290 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1291 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1292 cutoff_mask = _mm_cmplt_pd(rsq30,rcutoff2);
1296 fscal = _mm_and_pd(fscal,cutoff_mask);
1298 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1300 /* Update vectorial force */
1301 fix3 = _mm_macc_pd(dx30,fscal,fix3);
1302 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
1303 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
1305 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
1306 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
1307 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
1311 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1313 /* Inner loop uses 198 flops */
1316 /* End of innermost loop */
1318 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1319 f+i_coord_offset+DIM,fshift+i_shift_offset);
1321 /* Increment number of inner iterations */
1322 inneriter += j_index_end - j_index_start;
1324 /* Outer loop uses 18 flops */
1327 /* Increment number of outer iterations */
1330 /* Update outer/inner flops */
1332 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*198);