2 * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_avx_128_fma_double.h"
34 #include "kernelutil_x86_avx_128_fma_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_avx_128_fma_double
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_avx_128_fma_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
69 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
71 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
72 int vdwjidx0A,vdwjidx0B;
73 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
75 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
76 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
77 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
80 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
82 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
83 real rswitch_scalar,d_scalar;
84 __m128d dummy_mask,cutoff_mask;
85 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
86 __m128d one = _mm_set1_pd(1.0);
87 __m128d two = _mm_set1_pd(2.0);
93 jindex = nlist->jindex;
95 shiftidx = nlist->shift;
97 shiftvec = fr->shift_vec[0];
98 fshift = fr->fshift[0];
99 facel = _mm_set1_pd(fr->epsfac);
100 charge = mdatoms->chargeA;
102 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
103 ewtab = fr->ic->tabq_coul_FDV0;
104 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
105 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
107 /* Setup water-specific parameters */
108 inr = nlist->iinr[0];
109 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
110 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
111 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
113 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
114 rcutoff_scalar = fr->rcoulomb;
115 rcutoff = _mm_set1_pd(rcutoff_scalar);
116 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
118 rswitch_scalar = fr->rcoulomb_switch;
119 rswitch = _mm_set1_pd(rswitch_scalar);
120 /* Setup switch parameters */
121 d_scalar = rcutoff_scalar-rswitch_scalar;
122 d = _mm_set1_pd(d_scalar);
123 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
124 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
125 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
126 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
127 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
128 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
130 /* Avoid stupid compiler warnings */
138 /* Start outer loop over neighborlists */
139 for(iidx=0; iidx<nri; iidx++)
141 /* Load shift vector for this list */
142 i_shift_offset = DIM*shiftidx[iidx];
144 /* Load limits for loop over neighbors */
145 j_index_start = jindex[iidx];
146 j_index_end = jindex[iidx+1];
148 /* Get outer coordinate index */
150 i_coord_offset = DIM*inr;
152 /* Load i particle coords and add shift vector */
153 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
154 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
156 fix0 = _mm_setzero_pd();
157 fiy0 = _mm_setzero_pd();
158 fiz0 = _mm_setzero_pd();
159 fix1 = _mm_setzero_pd();
160 fiy1 = _mm_setzero_pd();
161 fiz1 = _mm_setzero_pd();
162 fix2 = _mm_setzero_pd();
163 fiy2 = _mm_setzero_pd();
164 fiz2 = _mm_setzero_pd();
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);
187 dx10 = _mm_sub_pd(ix1,jx0);
188 dy10 = _mm_sub_pd(iy1,jy0);
189 dz10 = _mm_sub_pd(iz1,jz0);
190 dx20 = _mm_sub_pd(ix2,jx0);
191 dy20 = _mm_sub_pd(iy2,jy0);
192 dz20 = _mm_sub_pd(iz2,jz0);
194 /* Calculate squared distance and things based on it */
195 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
196 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
197 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
199 rinv00 = gmx_mm_invsqrt_pd(rsq00);
200 rinv10 = gmx_mm_invsqrt_pd(rsq10);
201 rinv20 = gmx_mm_invsqrt_pd(rsq20);
203 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
204 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
205 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
207 /* Load parameters for j particles */
208 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
210 fjx0 = _mm_setzero_pd();
211 fjy0 = _mm_setzero_pd();
212 fjz0 = _mm_setzero_pd();
214 /**************************
215 * CALCULATE INTERACTIONS *
216 **************************/
218 if (gmx_mm_any_lt(rsq00,rcutoff2))
221 r00 = _mm_mul_pd(rsq00,rinv00);
223 /* Compute parameters for interactions between i and j atoms */
224 qq00 = _mm_mul_pd(iq0,jq0);
226 /* EWALD ELECTROSTATICS */
228 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
229 ewrt = _mm_mul_pd(r00,ewtabscale);
230 ewitab = _mm_cvttpd_epi32(ewrt);
232 eweps = _mm_frcz_pd(ewrt);
234 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
236 twoeweps = _mm_add_pd(eweps,eweps);
237 ewitab = _mm_slli_epi32(ewitab,2);
238 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
239 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
240 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
241 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
242 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
243 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
244 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
245 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
246 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
247 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
249 d = _mm_sub_pd(r00,rswitch);
250 d = _mm_max_pd(d,_mm_setzero_pd());
251 d2 = _mm_mul_pd(d,d);
252 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
254 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
256 /* Evaluate switch function */
257 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
258 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
259 velec = _mm_mul_pd(velec,sw);
260 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
262 /* Update potential sum for this i atom from the interaction with this j atom. */
263 velec = _mm_and_pd(velec,cutoff_mask);
264 velecsum = _mm_add_pd(velecsum,velec);
268 fscal = _mm_and_pd(fscal,cutoff_mask);
270 /* Update vectorial force */
271 fix0 = _mm_macc_pd(dx00,fscal,fix0);
272 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
273 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
275 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
276 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
277 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
281 /**************************
282 * CALCULATE INTERACTIONS *
283 **************************/
285 if (gmx_mm_any_lt(rsq10,rcutoff2))
288 r10 = _mm_mul_pd(rsq10,rinv10);
290 /* Compute parameters for interactions between i and j atoms */
291 qq10 = _mm_mul_pd(iq1,jq0);
293 /* EWALD ELECTROSTATICS */
295 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
296 ewrt = _mm_mul_pd(r10,ewtabscale);
297 ewitab = _mm_cvttpd_epi32(ewrt);
299 eweps = _mm_frcz_pd(ewrt);
301 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
303 twoeweps = _mm_add_pd(eweps,eweps);
304 ewitab = _mm_slli_epi32(ewitab,2);
305 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
306 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
307 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
308 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
309 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
310 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
311 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
312 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
313 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
314 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
316 d = _mm_sub_pd(r10,rswitch);
317 d = _mm_max_pd(d,_mm_setzero_pd());
318 d2 = _mm_mul_pd(d,d);
319 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
321 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
323 /* Evaluate switch function */
324 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
325 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
326 velec = _mm_mul_pd(velec,sw);
327 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
329 /* Update potential sum for this i atom from the interaction with this j atom. */
330 velec = _mm_and_pd(velec,cutoff_mask);
331 velecsum = _mm_add_pd(velecsum,velec);
335 fscal = _mm_and_pd(fscal,cutoff_mask);
337 /* Update vectorial force */
338 fix1 = _mm_macc_pd(dx10,fscal,fix1);
339 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
340 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
342 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
343 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
344 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
348 /**************************
349 * CALCULATE INTERACTIONS *
350 **************************/
352 if (gmx_mm_any_lt(rsq20,rcutoff2))
355 r20 = _mm_mul_pd(rsq20,rinv20);
357 /* Compute parameters for interactions between i and j atoms */
358 qq20 = _mm_mul_pd(iq2,jq0);
360 /* EWALD ELECTROSTATICS */
362 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
363 ewrt = _mm_mul_pd(r20,ewtabscale);
364 ewitab = _mm_cvttpd_epi32(ewrt);
366 eweps = _mm_frcz_pd(ewrt);
368 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
370 twoeweps = _mm_add_pd(eweps,eweps);
371 ewitab = _mm_slli_epi32(ewitab,2);
372 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
373 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
374 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
375 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
376 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
377 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
378 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
379 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
380 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
381 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
383 d = _mm_sub_pd(r20,rswitch);
384 d = _mm_max_pd(d,_mm_setzero_pd());
385 d2 = _mm_mul_pd(d,d);
386 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
388 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
390 /* Evaluate switch function */
391 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
392 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
393 velec = _mm_mul_pd(velec,sw);
394 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
396 /* Update potential sum for this i atom from the interaction with this j atom. */
397 velec = _mm_and_pd(velec,cutoff_mask);
398 velecsum = _mm_add_pd(velecsum,velec);
402 fscal = _mm_and_pd(fscal,cutoff_mask);
404 /* Update vectorial force */
405 fix2 = _mm_macc_pd(dx20,fscal,fix2);
406 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
407 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
409 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
410 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
411 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
415 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
417 /* Inner loop uses 207 flops */
424 j_coord_offsetA = DIM*jnrA;
426 /* load j atom coordinates */
427 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
430 /* Calculate displacement vector */
431 dx00 = _mm_sub_pd(ix0,jx0);
432 dy00 = _mm_sub_pd(iy0,jy0);
433 dz00 = _mm_sub_pd(iz0,jz0);
434 dx10 = _mm_sub_pd(ix1,jx0);
435 dy10 = _mm_sub_pd(iy1,jy0);
436 dz10 = _mm_sub_pd(iz1,jz0);
437 dx20 = _mm_sub_pd(ix2,jx0);
438 dy20 = _mm_sub_pd(iy2,jy0);
439 dz20 = _mm_sub_pd(iz2,jz0);
441 /* Calculate squared distance and things based on it */
442 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
443 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
444 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
446 rinv00 = gmx_mm_invsqrt_pd(rsq00);
447 rinv10 = gmx_mm_invsqrt_pd(rsq10);
448 rinv20 = gmx_mm_invsqrt_pd(rsq20);
450 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
451 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
452 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
454 /* Load parameters for j particles */
455 jq0 = _mm_load_sd(charge+jnrA+0);
457 fjx0 = _mm_setzero_pd();
458 fjy0 = _mm_setzero_pd();
459 fjz0 = _mm_setzero_pd();
461 /**************************
462 * CALCULATE INTERACTIONS *
463 **************************/
465 if (gmx_mm_any_lt(rsq00,rcutoff2))
468 r00 = _mm_mul_pd(rsq00,rinv00);
470 /* Compute parameters for interactions between i and j atoms */
471 qq00 = _mm_mul_pd(iq0,jq0);
473 /* EWALD ELECTROSTATICS */
475 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
476 ewrt = _mm_mul_pd(r00,ewtabscale);
477 ewitab = _mm_cvttpd_epi32(ewrt);
479 eweps = _mm_frcz_pd(ewrt);
481 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
483 twoeweps = _mm_add_pd(eweps,eweps);
484 ewitab = _mm_slli_epi32(ewitab,2);
485 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
486 ewtabD = _mm_setzero_pd();
487 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
488 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
489 ewtabFn = _mm_setzero_pd();
490 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
491 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
492 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
493 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
494 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
496 d = _mm_sub_pd(r00,rswitch);
497 d = _mm_max_pd(d,_mm_setzero_pd());
498 d2 = _mm_mul_pd(d,d);
499 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
501 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
503 /* Evaluate switch function */
504 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
505 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
506 velec = _mm_mul_pd(velec,sw);
507 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
509 /* Update potential sum for this i atom from the interaction with this j atom. */
510 velec = _mm_and_pd(velec,cutoff_mask);
511 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
512 velecsum = _mm_add_pd(velecsum,velec);
516 fscal = _mm_and_pd(fscal,cutoff_mask);
518 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
520 /* Update vectorial force */
521 fix0 = _mm_macc_pd(dx00,fscal,fix0);
522 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
523 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
525 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
526 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
527 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
531 /**************************
532 * CALCULATE INTERACTIONS *
533 **************************/
535 if (gmx_mm_any_lt(rsq10,rcutoff2))
538 r10 = _mm_mul_pd(rsq10,rinv10);
540 /* Compute parameters for interactions between i and j atoms */
541 qq10 = _mm_mul_pd(iq1,jq0);
543 /* EWALD ELECTROSTATICS */
545 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
546 ewrt = _mm_mul_pd(r10,ewtabscale);
547 ewitab = _mm_cvttpd_epi32(ewrt);
549 eweps = _mm_frcz_pd(ewrt);
551 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
553 twoeweps = _mm_add_pd(eweps,eweps);
554 ewitab = _mm_slli_epi32(ewitab,2);
555 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
556 ewtabD = _mm_setzero_pd();
557 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
558 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
559 ewtabFn = _mm_setzero_pd();
560 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
561 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
562 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
563 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
564 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
566 d = _mm_sub_pd(r10,rswitch);
567 d = _mm_max_pd(d,_mm_setzero_pd());
568 d2 = _mm_mul_pd(d,d);
569 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
571 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
573 /* Evaluate switch function */
574 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
575 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
576 velec = _mm_mul_pd(velec,sw);
577 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
579 /* Update potential sum for this i atom from the interaction with this j atom. */
580 velec = _mm_and_pd(velec,cutoff_mask);
581 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
582 velecsum = _mm_add_pd(velecsum,velec);
586 fscal = _mm_and_pd(fscal,cutoff_mask);
588 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
590 /* Update vectorial force */
591 fix1 = _mm_macc_pd(dx10,fscal,fix1);
592 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
593 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
595 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
596 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
597 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
601 /**************************
602 * CALCULATE INTERACTIONS *
603 **************************/
605 if (gmx_mm_any_lt(rsq20,rcutoff2))
608 r20 = _mm_mul_pd(rsq20,rinv20);
610 /* Compute parameters for interactions between i and j atoms */
611 qq20 = _mm_mul_pd(iq2,jq0);
613 /* EWALD ELECTROSTATICS */
615 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
616 ewrt = _mm_mul_pd(r20,ewtabscale);
617 ewitab = _mm_cvttpd_epi32(ewrt);
619 eweps = _mm_frcz_pd(ewrt);
621 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
623 twoeweps = _mm_add_pd(eweps,eweps);
624 ewitab = _mm_slli_epi32(ewitab,2);
625 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
626 ewtabD = _mm_setzero_pd();
627 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
628 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
629 ewtabFn = _mm_setzero_pd();
630 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
631 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
632 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
633 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
634 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
636 d = _mm_sub_pd(r20,rswitch);
637 d = _mm_max_pd(d,_mm_setzero_pd());
638 d2 = _mm_mul_pd(d,d);
639 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
641 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
643 /* Evaluate switch function */
644 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
645 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
646 velec = _mm_mul_pd(velec,sw);
647 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
649 /* Update potential sum for this i atom from the interaction with this j atom. */
650 velec = _mm_and_pd(velec,cutoff_mask);
651 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
652 velecsum = _mm_add_pd(velecsum,velec);
656 fscal = _mm_and_pd(fscal,cutoff_mask);
658 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
660 /* Update vectorial force */
661 fix2 = _mm_macc_pd(dx20,fscal,fix2);
662 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
663 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
665 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
666 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
667 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
671 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
673 /* Inner loop uses 207 flops */
676 /* End of innermost loop */
678 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
679 f+i_coord_offset,fshift+i_shift_offset);
682 /* Update potential energies */
683 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
685 /* Increment number of inner iterations */
686 inneriter += j_index_end - j_index_start;
688 /* Outer loop uses 19 flops */
691 /* Increment number of outer iterations */
694 /* Update outer/inner flops */
696 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*207);
699 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_128_fma_double
700 * Electrostatics interaction: Ewald
701 * VdW interaction: None
702 * Geometry: Water3-Particle
703 * Calculate force/pot: Force
706 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_128_fma_double
707 (t_nblist * gmx_restrict nlist,
708 rvec * gmx_restrict xx,
709 rvec * gmx_restrict ff,
710 t_forcerec * gmx_restrict fr,
711 t_mdatoms * gmx_restrict mdatoms,
712 nb_kernel_data_t * gmx_restrict kernel_data,
713 t_nrnb * gmx_restrict nrnb)
715 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
716 * just 0 for non-waters.
717 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
718 * jnr indices corresponding to data put in the four positions in the SIMD register.
720 int i_shift_offset,i_coord_offset,outeriter,inneriter;
721 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
723 int j_coord_offsetA,j_coord_offsetB;
724 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
726 real *shiftvec,*fshift,*x,*f;
727 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
729 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
731 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
733 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
734 int vdwjidx0A,vdwjidx0B;
735 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
736 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
737 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
738 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
739 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
742 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
744 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
745 real rswitch_scalar,d_scalar;
746 __m128d dummy_mask,cutoff_mask;
747 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
748 __m128d one = _mm_set1_pd(1.0);
749 __m128d two = _mm_set1_pd(2.0);
755 jindex = nlist->jindex;
757 shiftidx = nlist->shift;
759 shiftvec = fr->shift_vec[0];
760 fshift = fr->fshift[0];
761 facel = _mm_set1_pd(fr->epsfac);
762 charge = mdatoms->chargeA;
764 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
765 ewtab = fr->ic->tabq_coul_FDV0;
766 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
767 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
769 /* Setup water-specific parameters */
770 inr = nlist->iinr[0];
771 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
772 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
773 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
775 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
776 rcutoff_scalar = fr->rcoulomb;
777 rcutoff = _mm_set1_pd(rcutoff_scalar);
778 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
780 rswitch_scalar = fr->rcoulomb_switch;
781 rswitch = _mm_set1_pd(rswitch_scalar);
782 /* Setup switch parameters */
783 d_scalar = rcutoff_scalar-rswitch_scalar;
784 d = _mm_set1_pd(d_scalar);
785 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
786 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
787 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
788 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
789 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
790 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
792 /* Avoid stupid compiler warnings */
800 /* Start outer loop over neighborlists */
801 for(iidx=0; iidx<nri; iidx++)
803 /* Load shift vector for this list */
804 i_shift_offset = DIM*shiftidx[iidx];
806 /* Load limits for loop over neighbors */
807 j_index_start = jindex[iidx];
808 j_index_end = jindex[iidx+1];
810 /* Get outer coordinate index */
812 i_coord_offset = DIM*inr;
814 /* Load i particle coords and add shift vector */
815 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
816 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
818 fix0 = _mm_setzero_pd();
819 fiy0 = _mm_setzero_pd();
820 fiz0 = _mm_setzero_pd();
821 fix1 = _mm_setzero_pd();
822 fiy1 = _mm_setzero_pd();
823 fiz1 = _mm_setzero_pd();
824 fix2 = _mm_setzero_pd();
825 fiy2 = _mm_setzero_pd();
826 fiz2 = _mm_setzero_pd();
828 /* Start inner kernel loop */
829 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
832 /* Get j neighbor index, and coordinate index */
835 j_coord_offsetA = DIM*jnrA;
836 j_coord_offsetB = DIM*jnrB;
838 /* load j atom coordinates */
839 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
842 /* Calculate displacement vector */
843 dx00 = _mm_sub_pd(ix0,jx0);
844 dy00 = _mm_sub_pd(iy0,jy0);
845 dz00 = _mm_sub_pd(iz0,jz0);
846 dx10 = _mm_sub_pd(ix1,jx0);
847 dy10 = _mm_sub_pd(iy1,jy0);
848 dz10 = _mm_sub_pd(iz1,jz0);
849 dx20 = _mm_sub_pd(ix2,jx0);
850 dy20 = _mm_sub_pd(iy2,jy0);
851 dz20 = _mm_sub_pd(iz2,jz0);
853 /* Calculate squared distance and things based on it */
854 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
855 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
856 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
858 rinv00 = gmx_mm_invsqrt_pd(rsq00);
859 rinv10 = gmx_mm_invsqrt_pd(rsq10);
860 rinv20 = gmx_mm_invsqrt_pd(rsq20);
862 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
863 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
864 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
866 /* Load parameters for j particles */
867 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
869 fjx0 = _mm_setzero_pd();
870 fjy0 = _mm_setzero_pd();
871 fjz0 = _mm_setzero_pd();
873 /**************************
874 * CALCULATE INTERACTIONS *
875 **************************/
877 if (gmx_mm_any_lt(rsq00,rcutoff2))
880 r00 = _mm_mul_pd(rsq00,rinv00);
882 /* Compute parameters for interactions between i and j atoms */
883 qq00 = _mm_mul_pd(iq0,jq0);
885 /* EWALD ELECTROSTATICS */
887 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
888 ewrt = _mm_mul_pd(r00,ewtabscale);
889 ewitab = _mm_cvttpd_epi32(ewrt);
891 eweps = _mm_frcz_pd(ewrt);
893 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
895 twoeweps = _mm_add_pd(eweps,eweps);
896 ewitab = _mm_slli_epi32(ewitab,2);
897 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
898 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
899 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
900 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
901 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
902 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
903 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
904 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
905 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
906 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
908 d = _mm_sub_pd(r00,rswitch);
909 d = _mm_max_pd(d,_mm_setzero_pd());
910 d2 = _mm_mul_pd(d,d);
911 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
913 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
915 /* Evaluate switch function */
916 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
917 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
918 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
922 fscal = _mm_and_pd(fscal,cutoff_mask);
924 /* Update vectorial force */
925 fix0 = _mm_macc_pd(dx00,fscal,fix0);
926 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
927 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
929 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
930 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
931 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
935 /**************************
936 * CALCULATE INTERACTIONS *
937 **************************/
939 if (gmx_mm_any_lt(rsq10,rcutoff2))
942 r10 = _mm_mul_pd(rsq10,rinv10);
944 /* Compute parameters for interactions between i and j atoms */
945 qq10 = _mm_mul_pd(iq1,jq0);
947 /* EWALD ELECTROSTATICS */
949 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
950 ewrt = _mm_mul_pd(r10,ewtabscale);
951 ewitab = _mm_cvttpd_epi32(ewrt);
953 eweps = _mm_frcz_pd(ewrt);
955 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
957 twoeweps = _mm_add_pd(eweps,eweps);
958 ewitab = _mm_slli_epi32(ewitab,2);
959 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
960 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
961 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
962 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
963 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
964 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
965 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
966 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
967 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
968 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
970 d = _mm_sub_pd(r10,rswitch);
971 d = _mm_max_pd(d,_mm_setzero_pd());
972 d2 = _mm_mul_pd(d,d);
973 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
975 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
977 /* Evaluate switch function */
978 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
979 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
980 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
984 fscal = _mm_and_pd(fscal,cutoff_mask);
986 /* Update vectorial force */
987 fix1 = _mm_macc_pd(dx10,fscal,fix1);
988 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
989 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
991 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
992 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
993 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
997 /**************************
998 * CALCULATE INTERACTIONS *
999 **************************/
1001 if (gmx_mm_any_lt(rsq20,rcutoff2))
1004 r20 = _mm_mul_pd(rsq20,rinv20);
1006 /* Compute parameters for interactions between i and j atoms */
1007 qq20 = _mm_mul_pd(iq2,jq0);
1009 /* EWALD ELECTROSTATICS */
1011 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1012 ewrt = _mm_mul_pd(r20,ewtabscale);
1013 ewitab = _mm_cvttpd_epi32(ewrt);
1015 eweps = _mm_frcz_pd(ewrt);
1017 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1019 twoeweps = _mm_add_pd(eweps,eweps);
1020 ewitab = _mm_slli_epi32(ewitab,2);
1021 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1022 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1023 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1024 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1025 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1026 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1027 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1028 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1029 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1030 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1032 d = _mm_sub_pd(r20,rswitch);
1033 d = _mm_max_pd(d,_mm_setzero_pd());
1034 d2 = _mm_mul_pd(d,d);
1035 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1037 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1039 /* Evaluate switch function */
1040 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1041 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1042 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1046 fscal = _mm_and_pd(fscal,cutoff_mask);
1048 /* Update vectorial force */
1049 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1050 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1051 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1053 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1054 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1055 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1059 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1061 /* Inner loop uses 198 flops */
1064 if(jidx<j_index_end)
1068 j_coord_offsetA = DIM*jnrA;
1070 /* load j atom coordinates */
1071 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1074 /* Calculate displacement vector */
1075 dx00 = _mm_sub_pd(ix0,jx0);
1076 dy00 = _mm_sub_pd(iy0,jy0);
1077 dz00 = _mm_sub_pd(iz0,jz0);
1078 dx10 = _mm_sub_pd(ix1,jx0);
1079 dy10 = _mm_sub_pd(iy1,jy0);
1080 dz10 = _mm_sub_pd(iz1,jz0);
1081 dx20 = _mm_sub_pd(ix2,jx0);
1082 dy20 = _mm_sub_pd(iy2,jy0);
1083 dz20 = _mm_sub_pd(iz2,jz0);
1085 /* Calculate squared distance and things based on it */
1086 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1087 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1088 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1090 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1091 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1092 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1094 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1095 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1096 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1098 /* Load parameters for j particles */
1099 jq0 = _mm_load_sd(charge+jnrA+0);
1101 fjx0 = _mm_setzero_pd();
1102 fjy0 = _mm_setzero_pd();
1103 fjz0 = _mm_setzero_pd();
1105 /**************************
1106 * CALCULATE INTERACTIONS *
1107 **************************/
1109 if (gmx_mm_any_lt(rsq00,rcutoff2))
1112 r00 = _mm_mul_pd(rsq00,rinv00);
1114 /* Compute parameters for interactions between i and j atoms */
1115 qq00 = _mm_mul_pd(iq0,jq0);
1117 /* EWALD ELECTROSTATICS */
1119 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1120 ewrt = _mm_mul_pd(r00,ewtabscale);
1121 ewitab = _mm_cvttpd_epi32(ewrt);
1123 eweps = _mm_frcz_pd(ewrt);
1125 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1127 twoeweps = _mm_add_pd(eweps,eweps);
1128 ewitab = _mm_slli_epi32(ewitab,2);
1129 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1130 ewtabD = _mm_setzero_pd();
1131 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1132 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1133 ewtabFn = _mm_setzero_pd();
1134 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1135 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1136 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1137 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
1138 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1140 d = _mm_sub_pd(r00,rswitch);
1141 d = _mm_max_pd(d,_mm_setzero_pd());
1142 d2 = _mm_mul_pd(d,d);
1143 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1145 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1147 /* Evaluate switch function */
1148 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1149 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
1150 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1154 fscal = _mm_and_pd(fscal,cutoff_mask);
1156 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1158 /* Update vectorial force */
1159 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1160 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1161 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1163 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1164 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1165 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1169 /**************************
1170 * CALCULATE INTERACTIONS *
1171 **************************/
1173 if (gmx_mm_any_lt(rsq10,rcutoff2))
1176 r10 = _mm_mul_pd(rsq10,rinv10);
1178 /* Compute parameters for interactions between i and j atoms */
1179 qq10 = _mm_mul_pd(iq1,jq0);
1181 /* EWALD ELECTROSTATICS */
1183 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1184 ewrt = _mm_mul_pd(r10,ewtabscale);
1185 ewitab = _mm_cvttpd_epi32(ewrt);
1187 eweps = _mm_frcz_pd(ewrt);
1189 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1191 twoeweps = _mm_add_pd(eweps,eweps);
1192 ewitab = _mm_slli_epi32(ewitab,2);
1193 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1194 ewtabD = _mm_setzero_pd();
1195 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1196 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1197 ewtabFn = _mm_setzero_pd();
1198 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1199 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1200 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1201 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1202 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1204 d = _mm_sub_pd(r10,rswitch);
1205 d = _mm_max_pd(d,_mm_setzero_pd());
1206 d2 = _mm_mul_pd(d,d);
1207 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1209 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1211 /* Evaluate switch function */
1212 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1213 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1214 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1218 fscal = _mm_and_pd(fscal,cutoff_mask);
1220 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1222 /* Update vectorial force */
1223 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1224 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1225 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1227 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1228 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1229 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1233 /**************************
1234 * CALCULATE INTERACTIONS *
1235 **************************/
1237 if (gmx_mm_any_lt(rsq20,rcutoff2))
1240 r20 = _mm_mul_pd(rsq20,rinv20);
1242 /* Compute parameters for interactions between i and j atoms */
1243 qq20 = _mm_mul_pd(iq2,jq0);
1245 /* EWALD ELECTROSTATICS */
1247 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1248 ewrt = _mm_mul_pd(r20,ewtabscale);
1249 ewitab = _mm_cvttpd_epi32(ewrt);
1251 eweps = _mm_frcz_pd(ewrt);
1253 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1255 twoeweps = _mm_add_pd(eweps,eweps);
1256 ewitab = _mm_slli_epi32(ewitab,2);
1257 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1258 ewtabD = _mm_setzero_pd();
1259 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1260 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1261 ewtabFn = _mm_setzero_pd();
1262 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1263 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
1264 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1265 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1266 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1268 d = _mm_sub_pd(r20,rswitch);
1269 d = _mm_max_pd(d,_mm_setzero_pd());
1270 d2 = _mm_mul_pd(d,d);
1271 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1273 dsw = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1275 /* Evaluate switch function */
1276 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1277 felec = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1278 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1282 fscal = _mm_and_pd(fscal,cutoff_mask);
1284 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1286 /* Update vectorial force */
1287 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1288 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1289 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1291 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1292 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1293 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1297 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1299 /* Inner loop uses 198 flops */
1302 /* End of innermost loop */
1304 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1305 f+i_coord_offset,fshift+i_shift_offset);
1307 /* Increment number of inner iterations */
1308 inneriter += j_index_end - j_index_start;
1310 /* Outer loop uses 18 flops */
1313 /* Increment number of outer iterations */
1316 /* Update outer/inner flops */
1318 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*198);