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 "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.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_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
52 * Electrostatics interaction: CubicSplineTable
53 * VdW interaction: CubicSplineTable
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCSTab_VdwCSTab_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 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
88 int vdwjidx0A,vdwjidx0B;
89 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
91 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
92 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
93 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
94 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
97 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
100 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
101 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
103 __m128i ifour = _mm_set1_epi32(4);
104 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
106 __m128d dummy_mask,cutoff_mask;
107 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
108 __m128d one = _mm_set1_pd(1.0);
109 __m128d two = _mm_set1_pd(2.0);
115 jindex = nlist->jindex;
117 shiftidx = nlist->shift;
119 shiftvec = fr->shift_vec[0];
120 fshift = fr->fshift[0];
121 facel = _mm_set1_pd(fr->epsfac);
122 charge = mdatoms->chargeA;
123 nvdwtype = fr->ntype;
125 vdwtype = mdatoms->typeA;
127 vftab = kernel_data->table_elec_vdw->data;
128 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
130 /* Setup water-specific parameters */
131 inr = nlist->iinr[0];
132 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
133 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
134 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
135 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
137 /* Avoid stupid compiler warnings */
145 /* Start outer loop over neighborlists */
146 for(iidx=0; iidx<nri; iidx++)
148 /* Load shift vector for this list */
149 i_shift_offset = DIM*shiftidx[iidx];
151 /* Load limits for loop over neighbors */
152 j_index_start = jindex[iidx];
153 j_index_end = jindex[iidx+1];
155 /* Get outer coordinate index */
157 i_coord_offset = DIM*inr;
159 /* Load i particle coords and add shift vector */
160 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
161 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
163 fix0 = _mm_setzero_pd();
164 fiy0 = _mm_setzero_pd();
165 fiz0 = _mm_setzero_pd();
166 fix1 = _mm_setzero_pd();
167 fiy1 = _mm_setzero_pd();
168 fiz1 = _mm_setzero_pd();
169 fix2 = _mm_setzero_pd();
170 fiy2 = _mm_setzero_pd();
171 fiz2 = _mm_setzero_pd();
172 fix3 = _mm_setzero_pd();
173 fiy3 = _mm_setzero_pd();
174 fiz3 = _mm_setzero_pd();
176 /* Reset potential sums */
177 velecsum = _mm_setzero_pd();
178 vvdwsum = _mm_setzero_pd();
180 /* Start inner kernel loop */
181 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
184 /* Get j neighbor index, and coordinate index */
187 j_coord_offsetA = DIM*jnrA;
188 j_coord_offsetB = DIM*jnrB;
190 /* load j atom coordinates */
191 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
194 /* Calculate displacement vector */
195 dx00 = _mm_sub_pd(ix0,jx0);
196 dy00 = _mm_sub_pd(iy0,jy0);
197 dz00 = _mm_sub_pd(iz0,jz0);
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 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
210 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
211 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
212 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
214 rinv00 = gmx_mm_invsqrt_pd(rsq00);
215 rinv10 = gmx_mm_invsqrt_pd(rsq10);
216 rinv20 = gmx_mm_invsqrt_pd(rsq20);
217 rinv30 = gmx_mm_invsqrt_pd(rsq30);
219 /* Load parameters for j particles */
220 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
221 vdwjidx0A = 2*vdwtype[jnrA+0];
222 vdwjidx0B = 2*vdwtype[jnrB+0];
224 fjx0 = _mm_setzero_pd();
225 fjy0 = _mm_setzero_pd();
226 fjz0 = _mm_setzero_pd();
228 /**************************
229 * CALCULATE INTERACTIONS *
230 **************************/
232 r00 = _mm_mul_pd(rsq00,rinv00);
234 /* Compute parameters for interactions between i and j atoms */
235 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
236 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
238 /* Calculate table index by multiplying r with table scale and truncate to integer */
239 rt = _mm_mul_pd(r00,vftabscale);
240 vfitab = _mm_cvttpd_epi32(rt);
242 vfeps = _mm_frcz_pd(rt);
244 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
246 twovfeps = _mm_add_pd(vfeps,vfeps);
247 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
249 /* CUBIC SPLINE TABLE DISPERSION */
250 vfitab = _mm_add_epi32(vfitab,ifour);
251 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
252 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
253 GMX_MM_TRANSPOSE2_PD(Y,F);
254 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
255 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
256 GMX_MM_TRANSPOSE2_PD(G,H);
257 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
258 VV = _mm_macc_pd(vfeps,Fp,Y);
259 vvdw6 = _mm_mul_pd(c6_00,VV);
260 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
261 fvdw6 = _mm_mul_pd(c6_00,FF);
263 /* CUBIC SPLINE TABLE REPULSION */
264 vfitab = _mm_add_epi32(vfitab,ifour);
265 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
266 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
267 GMX_MM_TRANSPOSE2_PD(Y,F);
268 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
269 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
270 GMX_MM_TRANSPOSE2_PD(G,H);
271 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
272 VV = _mm_macc_pd(vfeps,Fp,Y);
273 vvdw12 = _mm_mul_pd(c12_00,VV);
274 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
275 fvdw12 = _mm_mul_pd(c12_00,FF);
276 vvdw = _mm_add_pd(vvdw12,vvdw6);
277 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
279 /* Update potential sum for this i atom from the interaction with this j atom. */
280 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
284 /* Update vectorial force */
285 fix0 = _mm_macc_pd(dx00,fscal,fix0);
286 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
287 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
289 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
290 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
291 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
293 /**************************
294 * CALCULATE INTERACTIONS *
295 **************************/
297 r10 = _mm_mul_pd(rsq10,rinv10);
299 /* Compute parameters for interactions between i and j atoms */
300 qq10 = _mm_mul_pd(iq1,jq0);
302 /* Calculate table index by multiplying r with table scale and truncate to integer */
303 rt = _mm_mul_pd(r10,vftabscale);
304 vfitab = _mm_cvttpd_epi32(rt);
306 vfeps = _mm_frcz_pd(rt);
308 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
310 twovfeps = _mm_add_pd(vfeps,vfeps);
311 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
313 /* CUBIC SPLINE TABLE ELECTROSTATICS */
314 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
315 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
316 GMX_MM_TRANSPOSE2_PD(Y,F);
317 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
318 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
319 GMX_MM_TRANSPOSE2_PD(G,H);
320 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
321 VV = _mm_macc_pd(vfeps,Fp,Y);
322 velec = _mm_mul_pd(qq10,VV);
323 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
324 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
326 /* Update potential sum for this i atom from the interaction with this j atom. */
327 velecsum = _mm_add_pd(velecsum,velec);
331 /* Update vectorial force */
332 fix1 = _mm_macc_pd(dx10,fscal,fix1);
333 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
334 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
336 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
337 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
338 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
340 /**************************
341 * CALCULATE INTERACTIONS *
342 **************************/
344 r20 = _mm_mul_pd(rsq20,rinv20);
346 /* Compute parameters for interactions between i and j atoms */
347 qq20 = _mm_mul_pd(iq2,jq0);
349 /* Calculate table index by multiplying r with table scale and truncate to integer */
350 rt = _mm_mul_pd(r20,vftabscale);
351 vfitab = _mm_cvttpd_epi32(rt);
353 vfeps = _mm_frcz_pd(rt);
355 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
357 twovfeps = _mm_add_pd(vfeps,vfeps);
358 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
360 /* CUBIC SPLINE TABLE ELECTROSTATICS */
361 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
362 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
363 GMX_MM_TRANSPOSE2_PD(Y,F);
364 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
365 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
366 GMX_MM_TRANSPOSE2_PD(G,H);
367 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
368 VV = _mm_macc_pd(vfeps,Fp,Y);
369 velec = _mm_mul_pd(qq20,VV);
370 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
371 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
373 /* Update potential sum for this i atom from the interaction with this j atom. */
374 velecsum = _mm_add_pd(velecsum,velec);
378 /* Update vectorial force */
379 fix2 = _mm_macc_pd(dx20,fscal,fix2);
380 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
381 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
383 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
384 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
385 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
387 /**************************
388 * CALCULATE INTERACTIONS *
389 **************************/
391 r30 = _mm_mul_pd(rsq30,rinv30);
393 /* Compute parameters for interactions between i and j atoms */
394 qq30 = _mm_mul_pd(iq3,jq0);
396 /* Calculate table index by multiplying r with table scale and truncate to integer */
397 rt = _mm_mul_pd(r30,vftabscale);
398 vfitab = _mm_cvttpd_epi32(rt);
400 vfeps = _mm_frcz_pd(rt);
402 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
404 twovfeps = _mm_add_pd(vfeps,vfeps);
405 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
407 /* CUBIC SPLINE TABLE ELECTROSTATICS */
408 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
409 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
410 GMX_MM_TRANSPOSE2_PD(Y,F);
411 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
412 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
413 GMX_MM_TRANSPOSE2_PD(G,H);
414 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
415 VV = _mm_macc_pd(vfeps,Fp,Y);
416 velec = _mm_mul_pd(qq30,VV);
417 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
418 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
420 /* Update potential sum for this i atom from the interaction with this j atom. */
421 velecsum = _mm_add_pd(velecsum,velec);
425 /* Update vectorial force */
426 fix3 = _mm_macc_pd(dx30,fscal,fix3);
427 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
428 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
430 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
431 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
432 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
434 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
436 /* Inner loop uses 200 flops */
443 j_coord_offsetA = DIM*jnrA;
445 /* load j atom coordinates */
446 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
449 /* Calculate displacement vector */
450 dx00 = _mm_sub_pd(ix0,jx0);
451 dy00 = _mm_sub_pd(iy0,jy0);
452 dz00 = _mm_sub_pd(iz0,jz0);
453 dx10 = _mm_sub_pd(ix1,jx0);
454 dy10 = _mm_sub_pd(iy1,jy0);
455 dz10 = _mm_sub_pd(iz1,jz0);
456 dx20 = _mm_sub_pd(ix2,jx0);
457 dy20 = _mm_sub_pd(iy2,jy0);
458 dz20 = _mm_sub_pd(iz2,jz0);
459 dx30 = _mm_sub_pd(ix3,jx0);
460 dy30 = _mm_sub_pd(iy3,jy0);
461 dz30 = _mm_sub_pd(iz3,jz0);
463 /* Calculate squared distance and things based on it */
464 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
465 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
466 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
467 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
469 rinv00 = gmx_mm_invsqrt_pd(rsq00);
470 rinv10 = gmx_mm_invsqrt_pd(rsq10);
471 rinv20 = gmx_mm_invsqrt_pd(rsq20);
472 rinv30 = gmx_mm_invsqrt_pd(rsq30);
474 /* Load parameters for j particles */
475 jq0 = _mm_load_sd(charge+jnrA+0);
476 vdwjidx0A = 2*vdwtype[jnrA+0];
478 fjx0 = _mm_setzero_pd();
479 fjy0 = _mm_setzero_pd();
480 fjz0 = _mm_setzero_pd();
482 /**************************
483 * CALCULATE INTERACTIONS *
484 **************************/
486 r00 = _mm_mul_pd(rsq00,rinv00);
488 /* Compute parameters for interactions between i and j atoms */
489 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
491 /* Calculate table index by multiplying r with table scale and truncate to integer */
492 rt = _mm_mul_pd(r00,vftabscale);
493 vfitab = _mm_cvttpd_epi32(rt);
495 vfeps = _mm_frcz_pd(rt);
497 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
499 twovfeps = _mm_add_pd(vfeps,vfeps);
500 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
502 /* CUBIC SPLINE TABLE DISPERSION */
503 vfitab = _mm_add_epi32(vfitab,ifour);
504 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
505 F = _mm_setzero_pd();
506 GMX_MM_TRANSPOSE2_PD(Y,F);
507 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
508 H = _mm_setzero_pd();
509 GMX_MM_TRANSPOSE2_PD(G,H);
510 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
511 VV = _mm_macc_pd(vfeps,Fp,Y);
512 vvdw6 = _mm_mul_pd(c6_00,VV);
513 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
514 fvdw6 = _mm_mul_pd(c6_00,FF);
516 /* CUBIC SPLINE TABLE REPULSION */
517 vfitab = _mm_add_epi32(vfitab,ifour);
518 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
519 F = _mm_setzero_pd();
520 GMX_MM_TRANSPOSE2_PD(Y,F);
521 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
522 H = _mm_setzero_pd();
523 GMX_MM_TRANSPOSE2_PD(G,H);
524 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
525 VV = _mm_macc_pd(vfeps,Fp,Y);
526 vvdw12 = _mm_mul_pd(c12_00,VV);
527 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
528 fvdw12 = _mm_mul_pd(c12_00,FF);
529 vvdw = _mm_add_pd(vvdw12,vvdw6);
530 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
532 /* Update potential sum for this i atom from the interaction with this j atom. */
533 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
534 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
538 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
540 /* Update vectorial force */
541 fix0 = _mm_macc_pd(dx00,fscal,fix0);
542 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
543 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
545 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
546 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
547 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
549 /**************************
550 * CALCULATE INTERACTIONS *
551 **************************/
553 r10 = _mm_mul_pd(rsq10,rinv10);
555 /* Compute parameters for interactions between i and j atoms */
556 qq10 = _mm_mul_pd(iq1,jq0);
558 /* Calculate table index by multiplying r with table scale and truncate to integer */
559 rt = _mm_mul_pd(r10,vftabscale);
560 vfitab = _mm_cvttpd_epi32(rt);
562 vfeps = _mm_frcz_pd(rt);
564 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
566 twovfeps = _mm_add_pd(vfeps,vfeps);
567 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
569 /* CUBIC SPLINE TABLE ELECTROSTATICS */
570 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
571 F = _mm_setzero_pd();
572 GMX_MM_TRANSPOSE2_PD(Y,F);
573 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
574 H = _mm_setzero_pd();
575 GMX_MM_TRANSPOSE2_PD(G,H);
576 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
577 VV = _mm_macc_pd(vfeps,Fp,Y);
578 velec = _mm_mul_pd(qq10,VV);
579 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
580 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
582 /* Update potential sum for this i atom from the interaction with this j atom. */
583 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
584 velecsum = _mm_add_pd(velecsum,velec);
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);
599 /**************************
600 * CALCULATE INTERACTIONS *
601 **************************/
603 r20 = _mm_mul_pd(rsq20,rinv20);
605 /* Compute parameters for interactions between i and j atoms */
606 qq20 = _mm_mul_pd(iq2,jq0);
608 /* Calculate table index by multiplying r with table scale and truncate to integer */
609 rt = _mm_mul_pd(r20,vftabscale);
610 vfitab = _mm_cvttpd_epi32(rt);
612 vfeps = _mm_frcz_pd(rt);
614 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
616 twovfeps = _mm_add_pd(vfeps,vfeps);
617 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
619 /* CUBIC SPLINE TABLE ELECTROSTATICS */
620 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
621 F = _mm_setzero_pd();
622 GMX_MM_TRANSPOSE2_PD(Y,F);
623 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
624 H = _mm_setzero_pd();
625 GMX_MM_TRANSPOSE2_PD(G,H);
626 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
627 VV = _mm_macc_pd(vfeps,Fp,Y);
628 velec = _mm_mul_pd(qq20,VV);
629 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
630 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
632 /* Update potential sum for this i atom from the interaction with this j atom. */
633 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
634 velecsum = _mm_add_pd(velecsum,velec);
638 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
640 /* Update vectorial force */
641 fix2 = _mm_macc_pd(dx20,fscal,fix2);
642 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
643 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
645 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
646 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
647 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
649 /**************************
650 * CALCULATE INTERACTIONS *
651 **************************/
653 r30 = _mm_mul_pd(rsq30,rinv30);
655 /* Compute parameters for interactions between i and j atoms */
656 qq30 = _mm_mul_pd(iq3,jq0);
658 /* Calculate table index by multiplying r with table scale and truncate to integer */
659 rt = _mm_mul_pd(r30,vftabscale);
660 vfitab = _mm_cvttpd_epi32(rt);
662 vfeps = _mm_frcz_pd(rt);
664 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
666 twovfeps = _mm_add_pd(vfeps,vfeps);
667 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
669 /* CUBIC SPLINE TABLE ELECTROSTATICS */
670 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
671 F = _mm_setzero_pd();
672 GMX_MM_TRANSPOSE2_PD(Y,F);
673 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
674 H = _mm_setzero_pd();
675 GMX_MM_TRANSPOSE2_PD(G,H);
676 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
677 VV = _mm_macc_pd(vfeps,Fp,Y);
678 velec = _mm_mul_pd(qq30,VV);
679 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
680 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
682 /* Update potential sum for this i atom from the interaction with this j atom. */
683 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
684 velecsum = _mm_add_pd(velecsum,velec);
688 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
690 /* Update vectorial force */
691 fix3 = _mm_macc_pd(dx30,fscal,fix3);
692 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
693 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
695 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
696 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
697 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
699 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
701 /* Inner loop uses 200 flops */
704 /* End of innermost loop */
706 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
707 f+i_coord_offset,fshift+i_shift_offset);
710 /* Update potential energies */
711 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
712 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
714 /* Increment number of inner iterations */
715 inneriter += j_index_end - j_index_start;
717 /* Outer loop uses 26 flops */
720 /* Increment number of outer iterations */
723 /* Update outer/inner flops */
725 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*200);
728 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_double
729 * Electrostatics interaction: CubicSplineTable
730 * VdW interaction: CubicSplineTable
731 * Geometry: Water4-Particle
732 * Calculate force/pot: Force
735 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_double
736 (t_nblist * gmx_restrict nlist,
737 rvec * gmx_restrict xx,
738 rvec * gmx_restrict ff,
739 t_forcerec * gmx_restrict fr,
740 t_mdatoms * gmx_restrict mdatoms,
741 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
742 t_nrnb * gmx_restrict nrnb)
744 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
745 * just 0 for non-waters.
746 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
747 * jnr indices corresponding to data put in the four positions in the SIMD register.
749 int i_shift_offset,i_coord_offset,outeriter,inneriter;
750 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
752 int j_coord_offsetA,j_coord_offsetB;
753 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
755 real *shiftvec,*fshift,*x,*f;
756 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
758 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
760 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
762 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
764 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
765 int vdwjidx0A,vdwjidx0B;
766 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
767 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
768 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
769 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
770 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
771 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
774 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
777 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
778 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
780 __m128i ifour = _mm_set1_epi32(4);
781 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
783 __m128d dummy_mask,cutoff_mask;
784 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
785 __m128d one = _mm_set1_pd(1.0);
786 __m128d two = _mm_set1_pd(2.0);
792 jindex = nlist->jindex;
794 shiftidx = nlist->shift;
796 shiftvec = fr->shift_vec[0];
797 fshift = fr->fshift[0];
798 facel = _mm_set1_pd(fr->epsfac);
799 charge = mdatoms->chargeA;
800 nvdwtype = fr->ntype;
802 vdwtype = mdatoms->typeA;
804 vftab = kernel_data->table_elec_vdw->data;
805 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
807 /* Setup water-specific parameters */
808 inr = nlist->iinr[0];
809 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
810 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
811 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
812 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
814 /* Avoid stupid compiler warnings */
822 /* Start outer loop over neighborlists */
823 for(iidx=0; iidx<nri; iidx++)
825 /* Load shift vector for this list */
826 i_shift_offset = DIM*shiftidx[iidx];
828 /* Load limits for loop over neighbors */
829 j_index_start = jindex[iidx];
830 j_index_end = jindex[iidx+1];
832 /* Get outer coordinate index */
834 i_coord_offset = DIM*inr;
836 /* Load i particle coords and add shift vector */
837 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
838 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
840 fix0 = _mm_setzero_pd();
841 fiy0 = _mm_setzero_pd();
842 fiz0 = _mm_setzero_pd();
843 fix1 = _mm_setzero_pd();
844 fiy1 = _mm_setzero_pd();
845 fiz1 = _mm_setzero_pd();
846 fix2 = _mm_setzero_pd();
847 fiy2 = _mm_setzero_pd();
848 fiz2 = _mm_setzero_pd();
849 fix3 = _mm_setzero_pd();
850 fiy3 = _mm_setzero_pd();
851 fiz3 = _mm_setzero_pd();
853 /* Start inner kernel loop */
854 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
857 /* Get j neighbor index, and coordinate index */
860 j_coord_offsetA = DIM*jnrA;
861 j_coord_offsetB = DIM*jnrB;
863 /* load j atom coordinates */
864 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
867 /* Calculate displacement vector */
868 dx00 = _mm_sub_pd(ix0,jx0);
869 dy00 = _mm_sub_pd(iy0,jy0);
870 dz00 = _mm_sub_pd(iz0,jz0);
871 dx10 = _mm_sub_pd(ix1,jx0);
872 dy10 = _mm_sub_pd(iy1,jy0);
873 dz10 = _mm_sub_pd(iz1,jz0);
874 dx20 = _mm_sub_pd(ix2,jx0);
875 dy20 = _mm_sub_pd(iy2,jy0);
876 dz20 = _mm_sub_pd(iz2,jz0);
877 dx30 = _mm_sub_pd(ix3,jx0);
878 dy30 = _mm_sub_pd(iy3,jy0);
879 dz30 = _mm_sub_pd(iz3,jz0);
881 /* Calculate squared distance and things based on it */
882 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
883 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
884 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
885 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
887 rinv00 = gmx_mm_invsqrt_pd(rsq00);
888 rinv10 = gmx_mm_invsqrt_pd(rsq10);
889 rinv20 = gmx_mm_invsqrt_pd(rsq20);
890 rinv30 = gmx_mm_invsqrt_pd(rsq30);
892 /* Load parameters for j particles */
893 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
894 vdwjidx0A = 2*vdwtype[jnrA+0];
895 vdwjidx0B = 2*vdwtype[jnrB+0];
897 fjx0 = _mm_setzero_pd();
898 fjy0 = _mm_setzero_pd();
899 fjz0 = _mm_setzero_pd();
901 /**************************
902 * CALCULATE INTERACTIONS *
903 **************************/
905 r00 = _mm_mul_pd(rsq00,rinv00);
907 /* Compute parameters for interactions between i and j atoms */
908 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
909 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
911 /* Calculate table index by multiplying r with table scale and truncate to integer */
912 rt = _mm_mul_pd(r00,vftabscale);
913 vfitab = _mm_cvttpd_epi32(rt);
915 vfeps = _mm_frcz_pd(rt);
917 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
919 twovfeps = _mm_add_pd(vfeps,vfeps);
920 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
922 /* CUBIC SPLINE TABLE DISPERSION */
923 vfitab = _mm_add_epi32(vfitab,ifour);
924 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
925 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
926 GMX_MM_TRANSPOSE2_PD(Y,F);
927 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
928 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
929 GMX_MM_TRANSPOSE2_PD(G,H);
930 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
931 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
932 fvdw6 = _mm_mul_pd(c6_00,FF);
934 /* CUBIC SPLINE TABLE REPULSION */
935 vfitab = _mm_add_epi32(vfitab,ifour);
936 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
937 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
938 GMX_MM_TRANSPOSE2_PD(Y,F);
939 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
940 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
941 GMX_MM_TRANSPOSE2_PD(G,H);
942 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
943 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
944 fvdw12 = _mm_mul_pd(c12_00,FF);
945 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
949 /* Update vectorial force */
950 fix0 = _mm_macc_pd(dx00,fscal,fix0);
951 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
952 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
954 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
955 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
956 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
958 /**************************
959 * CALCULATE INTERACTIONS *
960 **************************/
962 r10 = _mm_mul_pd(rsq10,rinv10);
964 /* Compute parameters for interactions between i and j atoms */
965 qq10 = _mm_mul_pd(iq1,jq0);
967 /* Calculate table index by multiplying r with table scale and truncate to integer */
968 rt = _mm_mul_pd(r10,vftabscale);
969 vfitab = _mm_cvttpd_epi32(rt);
971 vfeps = _mm_frcz_pd(rt);
973 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
975 twovfeps = _mm_add_pd(vfeps,vfeps);
976 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
978 /* CUBIC SPLINE TABLE ELECTROSTATICS */
979 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
980 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
981 GMX_MM_TRANSPOSE2_PD(Y,F);
982 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
983 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
984 GMX_MM_TRANSPOSE2_PD(G,H);
985 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
986 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
987 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
991 /* Update vectorial force */
992 fix1 = _mm_macc_pd(dx10,fscal,fix1);
993 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
994 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
996 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
997 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
998 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1000 /**************************
1001 * CALCULATE INTERACTIONS *
1002 **************************/
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 /* Calculate table index by multiplying r with table scale and truncate to integer */
1010 rt = _mm_mul_pd(r20,vftabscale);
1011 vfitab = _mm_cvttpd_epi32(rt);
1013 vfeps = _mm_frcz_pd(rt);
1015 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1017 twovfeps = _mm_add_pd(vfeps,vfeps);
1018 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1020 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1021 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1022 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1023 GMX_MM_TRANSPOSE2_PD(Y,F);
1024 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1025 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1026 GMX_MM_TRANSPOSE2_PD(G,H);
1027 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1028 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1029 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1033 /* Update vectorial force */
1034 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1035 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1036 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1038 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1039 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1040 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1042 /**************************
1043 * CALCULATE INTERACTIONS *
1044 **************************/
1046 r30 = _mm_mul_pd(rsq30,rinv30);
1048 /* Compute parameters for interactions between i and j atoms */
1049 qq30 = _mm_mul_pd(iq3,jq0);
1051 /* Calculate table index by multiplying r with table scale and truncate to integer */
1052 rt = _mm_mul_pd(r30,vftabscale);
1053 vfitab = _mm_cvttpd_epi32(rt);
1055 vfeps = _mm_frcz_pd(rt);
1057 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1059 twovfeps = _mm_add_pd(vfeps,vfeps);
1060 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1062 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1063 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1064 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1065 GMX_MM_TRANSPOSE2_PD(Y,F);
1066 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1067 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1068 GMX_MM_TRANSPOSE2_PD(G,H);
1069 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1070 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1071 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1075 /* Update vectorial force */
1076 fix3 = _mm_macc_pd(dx30,fscal,fix3);
1077 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
1078 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
1080 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
1081 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
1082 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
1084 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1086 /* Inner loop uses 180 flops */
1089 if(jidx<j_index_end)
1093 j_coord_offsetA = DIM*jnrA;
1095 /* load j atom coordinates */
1096 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1099 /* Calculate displacement vector */
1100 dx00 = _mm_sub_pd(ix0,jx0);
1101 dy00 = _mm_sub_pd(iy0,jy0);
1102 dz00 = _mm_sub_pd(iz0,jz0);
1103 dx10 = _mm_sub_pd(ix1,jx0);
1104 dy10 = _mm_sub_pd(iy1,jy0);
1105 dz10 = _mm_sub_pd(iz1,jz0);
1106 dx20 = _mm_sub_pd(ix2,jx0);
1107 dy20 = _mm_sub_pd(iy2,jy0);
1108 dz20 = _mm_sub_pd(iz2,jz0);
1109 dx30 = _mm_sub_pd(ix3,jx0);
1110 dy30 = _mm_sub_pd(iy3,jy0);
1111 dz30 = _mm_sub_pd(iz3,jz0);
1113 /* Calculate squared distance and things based on it */
1114 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1115 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1116 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1117 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1119 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1120 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1121 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1122 rinv30 = gmx_mm_invsqrt_pd(rsq30);
1124 /* Load parameters for j particles */
1125 jq0 = _mm_load_sd(charge+jnrA+0);
1126 vdwjidx0A = 2*vdwtype[jnrA+0];
1128 fjx0 = _mm_setzero_pd();
1129 fjy0 = _mm_setzero_pd();
1130 fjz0 = _mm_setzero_pd();
1132 /**************************
1133 * CALCULATE INTERACTIONS *
1134 **************************/
1136 r00 = _mm_mul_pd(rsq00,rinv00);
1138 /* Compute parameters for interactions between i and j atoms */
1139 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1141 /* Calculate table index by multiplying r with table scale and truncate to integer */
1142 rt = _mm_mul_pd(r00,vftabscale);
1143 vfitab = _mm_cvttpd_epi32(rt);
1145 vfeps = _mm_frcz_pd(rt);
1147 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1149 twovfeps = _mm_add_pd(vfeps,vfeps);
1150 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1152 /* CUBIC SPLINE TABLE DISPERSION */
1153 vfitab = _mm_add_epi32(vfitab,ifour);
1154 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1155 F = _mm_setzero_pd();
1156 GMX_MM_TRANSPOSE2_PD(Y,F);
1157 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1158 H = _mm_setzero_pd();
1159 GMX_MM_TRANSPOSE2_PD(G,H);
1160 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1161 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1162 fvdw6 = _mm_mul_pd(c6_00,FF);
1164 /* CUBIC SPLINE TABLE REPULSION */
1165 vfitab = _mm_add_epi32(vfitab,ifour);
1166 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1167 F = _mm_setzero_pd();
1168 GMX_MM_TRANSPOSE2_PD(Y,F);
1169 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1170 H = _mm_setzero_pd();
1171 GMX_MM_TRANSPOSE2_PD(G,H);
1172 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1173 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1174 fvdw12 = _mm_mul_pd(c12_00,FF);
1175 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1179 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1181 /* Update vectorial force */
1182 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1183 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1184 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1186 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1187 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1188 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1190 /**************************
1191 * CALCULATE INTERACTIONS *
1192 **************************/
1194 r10 = _mm_mul_pd(rsq10,rinv10);
1196 /* Compute parameters for interactions between i and j atoms */
1197 qq10 = _mm_mul_pd(iq1,jq0);
1199 /* Calculate table index by multiplying r with table scale and truncate to integer */
1200 rt = _mm_mul_pd(r10,vftabscale);
1201 vfitab = _mm_cvttpd_epi32(rt);
1203 vfeps = _mm_frcz_pd(rt);
1205 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1207 twovfeps = _mm_add_pd(vfeps,vfeps);
1208 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1210 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1211 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1212 F = _mm_setzero_pd();
1213 GMX_MM_TRANSPOSE2_PD(Y,F);
1214 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1215 H = _mm_setzero_pd();
1216 GMX_MM_TRANSPOSE2_PD(G,H);
1217 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1218 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1219 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1223 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1225 /* Update vectorial force */
1226 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1227 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1228 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1230 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1231 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1232 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1234 /**************************
1235 * CALCULATE INTERACTIONS *
1236 **************************/
1238 r20 = _mm_mul_pd(rsq20,rinv20);
1240 /* Compute parameters for interactions between i and j atoms */
1241 qq20 = _mm_mul_pd(iq2,jq0);
1243 /* Calculate table index by multiplying r with table scale and truncate to integer */
1244 rt = _mm_mul_pd(r20,vftabscale);
1245 vfitab = _mm_cvttpd_epi32(rt);
1247 vfeps = _mm_frcz_pd(rt);
1249 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1251 twovfeps = _mm_add_pd(vfeps,vfeps);
1252 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1254 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1255 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1256 F = _mm_setzero_pd();
1257 GMX_MM_TRANSPOSE2_PD(Y,F);
1258 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1259 H = _mm_setzero_pd();
1260 GMX_MM_TRANSPOSE2_PD(G,H);
1261 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1262 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1263 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1267 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1269 /* Update vectorial force */
1270 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1271 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1272 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1274 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1275 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1276 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1278 /**************************
1279 * CALCULATE INTERACTIONS *
1280 **************************/
1282 r30 = _mm_mul_pd(rsq30,rinv30);
1284 /* Compute parameters for interactions between i and j atoms */
1285 qq30 = _mm_mul_pd(iq3,jq0);
1287 /* Calculate table index by multiplying r with table scale and truncate to integer */
1288 rt = _mm_mul_pd(r30,vftabscale);
1289 vfitab = _mm_cvttpd_epi32(rt);
1291 vfeps = _mm_frcz_pd(rt);
1293 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1295 twovfeps = _mm_add_pd(vfeps,vfeps);
1296 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1298 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1299 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1300 F = _mm_setzero_pd();
1301 GMX_MM_TRANSPOSE2_PD(Y,F);
1302 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1303 H = _mm_setzero_pd();
1304 GMX_MM_TRANSPOSE2_PD(G,H);
1305 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1306 FF = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1307 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1311 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1313 /* Update vectorial force */
1314 fix3 = _mm_macc_pd(dx30,fscal,fix3);
1315 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
1316 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
1318 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
1319 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
1320 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
1322 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1324 /* Inner loop uses 180 flops */
1327 /* End of innermost loop */
1329 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1330 f+i_coord_offset,fshift+i_shift_offset);
1332 /* Increment number of inner iterations */
1333 inneriter += j_index_end - j_index_start;
1335 /* Outer loop uses 24 flops */
1338 /* Increment number of outer iterations */
1341 /* Update outer/inner flops */
1343 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*180);