?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed include simple.h from nb_kernel_avx_128_fma_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_128_fma_single / nb_kernel_ElecEw_VdwLJEw_GeomW4P1_avx_128_fma_single.c
blob162b9ad27130df3eda9b30a6774edc2d91a0d5c3
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015, 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.
8 *
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.
13 *
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.
18 *
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.
23 *
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.
31 *
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.
34 */
35 /*
36 * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
37 */
38 #include "gmxpre.h"
39
40 #include "config.h"
41
42 #include <math.h>
43
44 #include "../nb_kernel.h"
45 #include "gromacs/math/vec.h"
46 #include "gromacs/legacyheaders/nrnb.h"
47
48 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
49 #include "kernelutil_x86_avx_128_fma_single.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_avx_128_fma_single
53 * Electrostatics interaction: Ewald
54 * VdW interaction: LJEwald
55 * Geometry: Water4-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_avx_128_fma_single
60 (t_nblist * gmx_restrict nlist,
61 rvec * gmx_restrict xx,
62 rvec * gmx_restrict ff,
63 t_forcerec * gmx_restrict fr,
64 t_mdatoms * gmx_restrict mdatoms,
65 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
66 t_nrnb * gmx_restrict nrnb)
67 {
68 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
69 * just 0 for non-waters.
70 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
71 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 */
73 int i_shift_offset,i_coord_offset,outeriter,inneriter;
74 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int jnrA,jnrB,jnrC,jnrD;
76 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real rcutoff_scalar;
80 real *shiftvec,*fshift,*x,*f;
81 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 real scratch[4*DIM];
83 __m128 fscal,rcutoff,rcutoff2,jidxall;
84 int vdwioffset0;
85 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 int vdwioffset1;
87 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 int vdwioffset2;
89 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 int vdwioffset3;
91 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
92 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
93 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
94 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
95 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
96 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
97 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
98 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
99 real *charge;
100 int nvdwtype;
101 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
102 int *vdwtype;
103 real *vdwparam;
104 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
105 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
106 __m128 c6grid_00;
107 __m128 c6grid_10;
108 __m128 c6grid_20;
109 __m128 c6grid_30;
110 real *vdwgridparam;
111 __m128 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
112 __m128 one_half = _mm_set1_ps(0.5);
113 __m128 minus_one = _mm_set1_ps(-1.0);
114 __m128i ewitab;
115 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
116 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
117 real *ewtab;
118 __m128 dummy_mask,cutoff_mask;
119 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
120 __m128 one = _mm_set1_ps(1.0);
121 __m128 two = _mm_set1_ps(2.0);
122 x = xx[0];
123 f = ff[0];
124
125 nri = nlist->nri;
126 iinr = nlist->iinr;
127 jindex = nlist->jindex;
128 jjnr = nlist->jjnr;
129 shiftidx = nlist->shift;
130 gid = nlist->gid;
131 shiftvec = fr->shift_vec[0];
132 fshift = fr->fshift[0];
133 facel = _mm_set1_ps(fr->epsfac);
134 charge = mdatoms->chargeA;
135 nvdwtype = fr->ntype;
136 vdwparam = fr->nbfp;
137 vdwtype = mdatoms->typeA;
138 vdwgridparam = fr->ljpme_c6grid;
139 sh_lj_ewald = _mm_set1_ps(fr->ic->sh_lj_ewald);
140 ewclj = _mm_set1_ps(fr->ewaldcoeff_lj);
141 ewclj2 = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
142
143 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
144 beta = _mm_set1_ps(fr->ic->ewaldcoeff_q);
145 beta2 = _mm_mul_ps(beta,beta);
146 beta3 = _mm_mul_ps(beta,beta2);
147 ewtab = fr->ic->tabq_coul_FDV0;
148 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
149 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
150
151 /* Setup water-specific parameters */
152 inr = nlist->iinr[0];
153 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
154 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
155 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
156 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
157
158 /* Avoid stupid compiler warnings */
159 jnrA = jnrB = jnrC = jnrD = 0;
160 j_coord_offsetA = 0;
161 j_coord_offsetB = 0;
162 j_coord_offsetC = 0;
163 j_coord_offsetD = 0;
164
165 outeriter = 0;
166 inneriter = 0;
167
168 for(iidx=0;iidx<4*DIM;iidx++)
169 {
170 scratch[iidx] = 0.0;
171 }
172
173 /* Start outer loop over neighborlists */
174 for(iidx=0; iidx<nri; iidx++)
175 {
176 /* Load shift vector for this list */
177 i_shift_offset = DIM*shiftidx[iidx];
178
179 /* Load limits for loop over neighbors */
180 j_index_start = jindex[iidx];
181 j_index_end = jindex[iidx+1];
182
183 /* Get outer coordinate index */
184 inr = iinr[iidx];
185 i_coord_offset = DIM*inr;
186
187 /* Load i particle coords and add shift vector */
188 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
189 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
190
191 fix0 = _mm_setzero_ps();
192 fiy0 = _mm_setzero_ps();
193 fiz0 = _mm_setzero_ps();
194 fix1 = _mm_setzero_ps();
195 fiy1 = _mm_setzero_ps();
196 fiz1 = _mm_setzero_ps();
197 fix2 = _mm_setzero_ps();
198 fiy2 = _mm_setzero_ps();
199 fiz2 = _mm_setzero_ps();
200 fix3 = _mm_setzero_ps();
201 fiy3 = _mm_setzero_ps();
202 fiz3 = _mm_setzero_ps();
203
204 /* Reset potential sums */
205 velecsum = _mm_setzero_ps();
206 vvdwsum = _mm_setzero_ps();
207
208 /* Start inner kernel loop */
209 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
210 {
211
212 /* Get j neighbor index, and coordinate index */
213 jnrA = jjnr[jidx];
214 jnrB = jjnr[jidx+1];
215 jnrC = jjnr[jidx+2];
216 jnrD = jjnr[jidx+3];
217 j_coord_offsetA = DIM*jnrA;
218 j_coord_offsetB = DIM*jnrB;
219 j_coord_offsetC = DIM*jnrC;
220 j_coord_offsetD = DIM*jnrD;
221
222 /* load j atom coordinates */
223 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
224 x+j_coord_offsetC,x+j_coord_offsetD,
225 &jx0,&jy0,&jz0);
226
227 /* Calculate displacement vector */
228 dx00 = _mm_sub_ps(ix0,jx0);
229 dy00 = _mm_sub_ps(iy0,jy0);
230 dz00 = _mm_sub_ps(iz0,jz0);
231 dx10 = _mm_sub_ps(ix1,jx0);
232 dy10 = _mm_sub_ps(iy1,jy0);
233 dz10 = _mm_sub_ps(iz1,jz0);
234 dx20 = _mm_sub_ps(ix2,jx0);
235 dy20 = _mm_sub_ps(iy2,jy0);
236 dz20 = _mm_sub_ps(iz2,jz0);
237 dx30 = _mm_sub_ps(ix3,jx0);
238 dy30 = _mm_sub_ps(iy3,jy0);
239 dz30 = _mm_sub_ps(iz3,jz0);
240
241 /* Calculate squared distance and things based on it */
242 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
243 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
244 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
245 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
246
247 rinv00 = gmx_mm_invsqrt_ps(rsq00);
248 rinv10 = gmx_mm_invsqrt_ps(rsq10);
249 rinv20 = gmx_mm_invsqrt_ps(rsq20);
250 rinv30 = gmx_mm_invsqrt_ps(rsq30);
251
252 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
253 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
254 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
255 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
256
257 /* Load parameters for j particles */
258 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
259 charge+jnrC+0,charge+jnrD+0);
260 vdwjidx0A = 2*vdwtype[jnrA+0];
261 vdwjidx0B = 2*vdwtype[jnrB+0];
262 vdwjidx0C = 2*vdwtype[jnrC+0];
263 vdwjidx0D = 2*vdwtype[jnrD+0];
264
265 fjx0 = _mm_setzero_ps();
266 fjy0 = _mm_setzero_ps();
267 fjz0 = _mm_setzero_ps();
268
269 /**************************
270 * CALCULATE INTERACTIONS *
271 **************************/
272
273 r00 = _mm_mul_ps(rsq00,rinv00);
274
275 /* Compute parameters for interactions between i and j atoms */
276 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
277 vdwparam+vdwioffset0+vdwjidx0B,
278 vdwparam+vdwioffset0+vdwjidx0C,
279 vdwparam+vdwioffset0+vdwjidx0D,
280 &c6_00,&c12_00);
281
282 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
283 vdwgridparam+vdwioffset0+vdwjidx0B,
284 vdwgridparam+vdwioffset0+vdwjidx0C,
285 vdwgridparam+vdwioffset0+vdwjidx0D);
286
287 /* Analytical LJ-PME */
288 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
289 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
290 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
291 exponent = gmx_simd_exp_r(ewcljrsq);
292 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
293 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
294 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
295 vvdw6 = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
296 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
297 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
298 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
299 fvdw = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
300
301 /* Update potential sum for this i atom from the interaction with this j atom. */
302 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
303
304 fscal = fvdw;
305
306 /* Update vectorial force */
307 fix0 = _mm_macc_ps(dx00,fscal,fix0);
308 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
309 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
310
311 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
312 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
313 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
314
315 /**************************
316 * CALCULATE INTERACTIONS *
317 **************************/
318
319 r10 = _mm_mul_ps(rsq10,rinv10);
320
321 /* Compute parameters for interactions between i and j atoms */
322 qq10 = _mm_mul_ps(iq1,jq0);
323
324 /* EWALD ELECTROSTATICS */
325
326 /* Analytical PME correction */
327 zeta2 = _mm_mul_ps(beta2,rsq10);
328 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
329 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
330 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
331 felec = _mm_mul_ps(qq10,felec);
332 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
333 velec = _mm_nmacc_ps(pmecorrV,beta,rinv10);
334 velec = _mm_mul_ps(qq10,velec);
335
336 /* Update potential sum for this i atom from the interaction with this j atom. */
337 velecsum = _mm_add_ps(velecsum,velec);
338
339 fscal = felec;
340
341 /* Update vectorial force */
342 fix1 = _mm_macc_ps(dx10,fscal,fix1);
343 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
344 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
345
346 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
347 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
348 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
349
350 /**************************
351 * CALCULATE INTERACTIONS *
352 **************************/
353
354 r20 = _mm_mul_ps(rsq20,rinv20);
355
356 /* Compute parameters for interactions between i and j atoms */
357 qq20 = _mm_mul_ps(iq2,jq0);
358
359 /* EWALD ELECTROSTATICS */
360
361 /* Analytical PME correction */
362 zeta2 = _mm_mul_ps(beta2,rsq20);
363 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
364 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
365 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
366 felec = _mm_mul_ps(qq20,felec);
367 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
368 velec = _mm_nmacc_ps(pmecorrV,beta,rinv20);
369 velec = _mm_mul_ps(qq20,velec);
370
371 /* Update potential sum for this i atom from the interaction with this j atom. */
372 velecsum = _mm_add_ps(velecsum,velec);
373
374 fscal = felec;
375
376 /* Update vectorial force */
377 fix2 = _mm_macc_ps(dx20,fscal,fix2);
378 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
379 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
380
381 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
382 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
383 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
384
385 /**************************
386 * CALCULATE INTERACTIONS *
387 **************************/
388
389 r30 = _mm_mul_ps(rsq30,rinv30);
390
391 /* Compute parameters for interactions between i and j atoms */
392 qq30 = _mm_mul_ps(iq3,jq0);
393
394 /* EWALD ELECTROSTATICS */
395
396 /* Analytical PME correction */
397 zeta2 = _mm_mul_ps(beta2,rsq30);
398 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
399 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
400 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
401 felec = _mm_mul_ps(qq30,felec);
402 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
403 velec = _mm_nmacc_ps(pmecorrV,beta,rinv30);
404 velec = _mm_mul_ps(qq30,velec);
405
406 /* Update potential sum for this i atom from the interaction with this j atom. */
407 velecsum = _mm_add_ps(velecsum,velec);
408
409 fscal = felec;
410
411 /* Update vectorial force */
412 fix3 = _mm_macc_ps(dx30,fscal,fix3);
413 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
414 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
415
416 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
417 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
418 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
419
420 fjptrA = f+j_coord_offsetA;
421 fjptrB = f+j_coord_offsetB;
422 fjptrC = f+j_coord_offsetC;
423 fjptrD = f+j_coord_offsetD;
424
425 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
426
427 /* Inner loop uses 137 flops */
428 }
429
430 if(jidx<j_index_end)
431 {
432
433 /* Get j neighbor index, and coordinate index */
434 jnrlistA = jjnr[jidx];
435 jnrlistB = jjnr[jidx+1];
436 jnrlistC = jjnr[jidx+2];
437 jnrlistD = jjnr[jidx+3];
438 /* Sign of each element will be negative for non-real atoms.
439 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
440 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
441 */
442 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
443 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
444 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
445 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
446 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
447 j_coord_offsetA = DIM*jnrA;
448 j_coord_offsetB = DIM*jnrB;
449 j_coord_offsetC = DIM*jnrC;
450 j_coord_offsetD = DIM*jnrD;
451
452 /* load j atom coordinates */
453 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
454 x+j_coord_offsetC,x+j_coord_offsetD,
455 &jx0,&jy0,&jz0);
456
457 /* Calculate displacement vector */
458 dx00 = _mm_sub_ps(ix0,jx0);
459 dy00 = _mm_sub_ps(iy0,jy0);
460 dz00 = _mm_sub_ps(iz0,jz0);
461 dx10 = _mm_sub_ps(ix1,jx0);
462 dy10 = _mm_sub_ps(iy1,jy0);
463 dz10 = _mm_sub_ps(iz1,jz0);
464 dx20 = _mm_sub_ps(ix2,jx0);
465 dy20 = _mm_sub_ps(iy2,jy0);
466 dz20 = _mm_sub_ps(iz2,jz0);
467 dx30 = _mm_sub_ps(ix3,jx0);
468 dy30 = _mm_sub_ps(iy3,jy0);
469 dz30 = _mm_sub_ps(iz3,jz0);
470
471 /* Calculate squared distance and things based on it */
472 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
473 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
474 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
475 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
476
477 rinv00 = gmx_mm_invsqrt_ps(rsq00);
478 rinv10 = gmx_mm_invsqrt_ps(rsq10);
479 rinv20 = gmx_mm_invsqrt_ps(rsq20);
480 rinv30 = gmx_mm_invsqrt_ps(rsq30);
481
482 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
483 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
484 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
485 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
486
487 /* Load parameters for j particles */
488 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
489 charge+jnrC+0,charge+jnrD+0);
490 vdwjidx0A = 2*vdwtype[jnrA+0];
491 vdwjidx0B = 2*vdwtype[jnrB+0];
492 vdwjidx0C = 2*vdwtype[jnrC+0];
493 vdwjidx0D = 2*vdwtype[jnrD+0];
494
495 fjx0 = _mm_setzero_ps();
496 fjy0 = _mm_setzero_ps();
497 fjz0 = _mm_setzero_ps();
498
499 /**************************
500 * CALCULATE INTERACTIONS *
501 **************************/
502
503 r00 = _mm_mul_ps(rsq00,rinv00);
504 r00 = _mm_andnot_ps(dummy_mask,r00);
505
506 /* Compute parameters for interactions between i and j atoms */
507 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
508 vdwparam+vdwioffset0+vdwjidx0B,
509 vdwparam+vdwioffset0+vdwjidx0C,
510 vdwparam+vdwioffset0+vdwjidx0D,
511 &c6_00,&c12_00);
512
513 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
514 vdwgridparam+vdwioffset0+vdwjidx0B,
515 vdwgridparam+vdwioffset0+vdwjidx0C,
516 vdwgridparam+vdwioffset0+vdwjidx0D);
517
518 /* Analytical LJ-PME */
519 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
520 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
521 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
522 exponent = gmx_simd_exp_r(ewcljrsq);
523 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
524 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
525 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
526 vvdw6 = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
527 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
528 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
529 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
530 fvdw = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
531
532 /* Update potential sum for this i atom from the interaction with this j atom. */
533 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
534 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
535
536 fscal = fvdw;
537
538 fscal = _mm_andnot_ps(dummy_mask,fscal);
539
540 /* Update vectorial force */
541 fix0 = _mm_macc_ps(dx00,fscal,fix0);
542 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
543 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
544
545 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
546 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
547 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
548
549 /**************************
550 * CALCULATE INTERACTIONS *
551 **************************/
552
553 r10 = _mm_mul_ps(rsq10,rinv10);
554 r10 = _mm_andnot_ps(dummy_mask,r10);
555
556 /* Compute parameters for interactions between i and j atoms */
557 qq10 = _mm_mul_ps(iq1,jq0);
558
559 /* EWALD ELECTROSTATICS */
560
561 /* Analytical PME correction */
562 zeta2 = _mm_mul_ps(beta2,rsq10);
563 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
564 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
565 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
566 felec = _mm_mul_ps(qq10,felec);
567 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
568 velec = _mm_nmacc_ps(pmecorrV,beta,rinv10);
569 velec = _mm_mul_ps(qq10,velec);
570
571 /* Update potential sum for this i atom from the interaction with this j atom. */
572 velec = _mm_andnot_ps(dummy_mask,velec);
573 velecsum = _mm_add_ps(velecsum,velec);
574
575 fscal = felec;
576
577 fscal = _mm_andnot_ps(dummy_mask,fscal);
578
579 /* Update vectorial force */
580 fix1 = _mm_macc_ps(dx10,fscal,fix1);
581 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
582 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
583
584 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
585 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
586 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
587
588 /**************************
589 * CALCULATE INTERACTIONS *
590 **************************/
591
592 r20 = _mm_mul_ps(rsq20,rinv20);
593 r20 = _mm_andnot_ps(dummy_mask,r20);
594
595 /* Compute parameters for interactions between i and j atoms */
596 qq20 = _mm_mul_ps(iq2,jq0);
597
598 /* EWALD ELECTROSTATICS */
599
600 /* Analytical PME correction */
601 zeta2 = _mm_mul_ps(beta2,rsq20);
602 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
603 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
604 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
605 felec = _mm_mul_ps(qq20,felec);
606 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
607 velec = _mm_nmacc_ps(pmecorrV,beta,rinv20);
608 velec = _mm_mul_ps(qq20,velec);
609
610 /* Update potential sum for this i atom from the interaction with this j atom. */
611 velec = _mm_andnot_ps(dummy_mask,velec);
612 velecsum = _mm_add_ps(velecsum,velec);
613
614 fscal = felec;
615
616 fscal = _mm_andnot_ps(dummy_mask,fscal);
617
618 /* Update vectorial force */
619 fix2 = _mm_macc_ps(dx20,fscal,fix2);
620 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
621 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
622
623 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
624 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
625 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
626
627 /**************************
628 * CALCULATE INTERACTIONS *
629 **************************/
630
631 r30 = _mm_mul_ps(rsq30,rinv30);
632 r30 = _mm_andnot_ps(dummy_mask,r30);
633
634 /* Compute parameters for interactions between i and j atoms */
635 qq30 = _mm_mul_ps(iq3,jq0);
636
637 /* EWALD ELECTROSTATICS */
638
639 /* Analytical PME correction */
640 zeta2 = _mm_mul_ps(beta2,rsq30);
641 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
642 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
643 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
644 felec = _mm_mul_ps(qq30,felec);
645 pmecorrV = gmx_mm_pmecorrV_ps(zeta2);
646 velec = _mm_nmacc_ps(pmecorrV,beta,rinv30);
647 velec = _mm_mul_ps(qq30,velec);
648
649 /* Update potential sum for this i atom from the interaction with this j atom. */
650 velec = _mm_andnot_ps(dummy_mask,velec);
651 velecsum = _mm_add_ps(velecsum,velec);
652
653 fscal = felec;
654
655 fscal = _mm_andnot_ps(dummy_mask,fscal);
656
657 /* Update vectorial force */
658 fix3 = _mm_macc_ps(dx30,fscal,fix3);
659 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
660 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
661
662 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
663 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
664 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
665
666 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
667 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
668 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
669 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
670
671 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
672
673 /* Inner loop uses 141 flops */
674 }
675
676 /* End of innermost loop */
677
678 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
679 f+i_coord_offset,fshift+i_shift_offset);
680
681 ggid = gid[iidx];
682 /* Update potential energies */
683 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
684 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
685
686 /* Increment number of inner iterations */
687 inneriter += j_index_end - j_index_start;
688
689 /* Outer loop uses 26 flops */
690 }
691
692 /* Increment number of outer iterations */
693 outeriter += nri;
694
695 /* Update outer/inner flops */
696
697 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
698 }
699 /*
700 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_single
701 * Electrostatics interaction: Ewald
702 * VdW interaction: LJEwald
703 * Geometry: Water4-Particle
704 * Calculate force/pot: Force
705 */
706 void
707 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_single
708 (t_nblist * gmx_restrict nlist,
709 rvec * gmx_restrict xx,
710 rvec * gmx_restrict ff,
711 t_forcerec * gmx_restrict fr,
712 t_mdatoms * gmx_restrict mdatoms,
713 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
714 t_nrnb * gmx_restrict nrnb)
715 {
716 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
717 * just 0 for non-waters.
718 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
719 * jnr indices corresponding to data put in the four positions in the SIMD register.
720 */
721 int i_shift_offset,i_coord_offset,outeriter,inneriter;
722 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
723 int jnrA,jnrB,jnrC,jnrD;
724 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
725 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
726 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
727 real rcutoff_scalar;
728 real *shiftvec,*fshift,*x,*f;
729 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
730 real scratch[4*DIM];
731 __m128 fscal,rcutoff,rcutoff2,jidxall;
732 int vdwioffset0;
733 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
734 int vdwioffset1;
735 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
736 int vdwioffset2;
737 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
738 int vdwioffset3;
739 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
740 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
741 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
742 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
743 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
744 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
745 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
746 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
747 real *charge;
748 int nvdwtype;
749 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
750 int *vdwtype;
751 real *vdwparam;
752 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
753 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
754 __m128 c6grid_00;
755 __m128 c6grid_10;
756 __m128 c6grid_20;
757 __m128 c6grid_30;
758 real *vdwgridparam;
759 __m128 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
760 __m128 one_half = _mm_set1_ps(0.5);
761 __m128 minus_one = _mm_set1_ps(-1.0);
762 __m128i ewitab;
763 __m128 ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
764 __m128 beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
765 real *ewtab;
766 __m128 dummy_mask,cutoff_mask;
767 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
768 __m128 one = _mm_set1_ps(1.0);
769 __m128 two = _mm_set1_ps(2.0);
770 x = xx[0];
771 f = ff[0];
772
773 nri = nlist->nri;
774 iinr = nlist->iinr;
775 jindex = nlist->jindex;
776 jjnr = nlist->jjnr;
777 shiftidx = nlist->shift;
778 gid = nlist->gid;
779 shiftvec = fr->shift_vec[0];
780 fshift = fr->fshift[0];
781 facel = _mm_set1_ps(fr->epsfac);
782 charge = mdatoms->chargeA;
783 nvdwtype = fr->ntype;
784 vdwparam = fr->nbfp;
785 vdwtype = mdatoms->typeA;
786 vdwgridparam = fr->ljpme_c6grid;
787 sh_lj_ewald = _mm_set1_ps(fr->ic->sh_lj_ewald);
788 ewclj = _mm_set1_ps(fr->ewaldcoeff_lj);
789 ewclj2 = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
790
791 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
792 beta = _mm_set1_ps(fr->ic->ewaldcoeff_q);
793 beta2 = _mm_mul_ps(beta,beta);
794 beta3 = _mm_mul_ps(beta,beta2);
795 ewtab = fr->ic->tabq_coul_F;
796 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
797 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
798
799 /* Setup water-specific parameters */
800 inr = nlist->iinr[0];
801 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
802 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
803 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
804 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
805
806 /* Avoid stupid compiler warnings */
807 jnrA = jnrB = jnrC = jnrD = 0;
808 j_coord_offsetA = 0;
809 j_coord_offsetB = 0;
810 j_coord_offsetC = 0;
811 j_coord_offsetD = 0;
812
813 outeriter = 0;
814 inneriter = 0;
815
816 for(iidx=0;iidx<4*DIM;iidx++)
817 {
818 scratch[iidx] = 0.0;
819 }
820
821 /* Start outer loop over neighborlists */
822 for(iidx=0; iidx<nri; iidx++)
823 {
824 /* Load shift vector for this list */
825 i_shift_offset = DIM*shiftidx[iidx];
826
827 /* Load limits for loop over neighbors */
828 j_index_start = jindex[iidx];
829 j_index_end = jindex[iidx+1];
830
831 /* Get outer coordinate index */
832 inr = iinr[iidx];
833 i_coord_offset = DIM*inr;
834
835 /* Load i particle coords and add shift vector */
836 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
837 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
838
839 fix0 = _mm_setzero_ps();
840 fiy0 = _mm_setzero_ps();
841 fiz0 = _mm_setzero_ps();
842 fix1 = _mm_setzero_ps();
843 fiy1 = _mm_setzero_ps();
844 fiz1 = _mm_setzero_ps();
845 fix2 = _mm_setzero_ps();
846 fiy2 = _mm_setzero_ps();
847 fiz2 = _mm_setzero_ps();
848 fix3 = _mm_setzero_ps();
849 fiy3 = _mm_setzero_ps();
850 fiz3 = _mm_setzero_ps();
851
852 /* Start inner kernel loop */
853 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
854 {
855
856 /* Get j neighbor index, and coordinate index */
857 jnrA = jjnr[jidx];
858 jnrB = jjnr[jidx+1];
859 jnrC = jjnr[jidx+2];
860 jnrD = jjnr[jidx+3];
861 j_coord_offsetA = DIM*jnrA;
862 j_coord_offsetB = DIM*jnrB;
863 j_coord_offsetC = DIM*jnrC;
864 j_coord_offsetD = DIM*jnrD;
865
866 /* load j atom coordinates */
867 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
868 x+j_coord_offsetC,x+j_coord_offsetD,
869 &jx0,&jy0,&jz0);
870
871 /* Calculate displacement vector */
872 dx00 = _mm_sub_ps(ix0,jx0);
873 dy00 = _mm_sub_ps(iy0,jy0);
874 dz00 = _mm_sub_ps(iz0,jz0);
875 dx10 = _mm_sub_ps(ix1,jx0);
876 dy10 = _mm_sub_ps(iy1,jy0);
877 dz10 = _mm_sub_ps(iz1,jz0);
878 dx20 = _mm_sub_ps(ix2,jx0);
879 dy20 = _mm_sub_ps(iy2,jy0);
880 dz20 = _mm_sub_ps(iz2,jz0);
881 dx30 = _mm_sub_ps(ix3,jx0);
882 dy30 = _mm_sub_ps(iy3,jy0);
883 dz30 = _mm_sub_ps(iz3,jz0);
884
885 /* Calculate squared distance and things based on it */
886 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
887 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
888 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
889 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
890
891 rinv00 = gmx_mm_invsqrt_ps(rsq00);
892 rinv10 = gmx_mm_invsqrt_ps(rsq10);
893 rinv20 = gmx_mm_invsqrt_ps(rsq20);
894 rinv30 = gmx_mm_invsqrt_ps(rsq30);
895
896 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
897 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
898 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
899 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
900
901 /* Load parameters for j particles */
902 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
903 charge+jnrC+0,charge+jnrD+0);
904 vdwjidx0A = 2*vdwtype[jnrA+0];
905 vdwjidx0B = 2*vdwtype[jnrB+0];
906 vdwjidx0C = 2*vdwtype[jnrC+0];
907 vdwjidx0D = 2*vdwtype[jnrD+0];
908
909 fjx0 = _mm_setzero_ps();
910 fjy0 = _mm_setzero_ps();
911 fjz0 = _mm_setzero_ps();
912
913 /**************************
914 * CALCULATE INTERACTIONS *
915 **************************/
916
917 r00 = _mm_mul_ps(rsq00,rinv00);
918
919 /* Compute parameters for interactions between i and j atoms */
920 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
921 vdwparam+vdwioffset0+vdwjidx0B,
922 vdwparam+vdwioffset0+vdwjidx0C,
923 vdwparam+vdwioffset0+vdwjidx0D,
924 &c6_00,&c12_00);
925
926 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
927 vdwgridparam+vdwioffset0+vdwjidx0B,
928 vdwgridparam+vdwioffset0+vdwjidx0C,
929 vdwgridparam+vdwioffset0+vdwjidx0D);
930
931 /* Analytical LJ-PME */
932 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
933 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
934 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
935 exponent = gmx_simd_exp_r(ewcljrsq);
936 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
937 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
938 /* f6A = 6 * C6grid * (1 - poly) */
939 f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
940 /* f6B = C6grid * exponent * beta^6 */
941 f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
942 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
943 fvdw = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
944
945 fscal = fvdw;
946
947 /* Update vectorial force */
948 fix0 = _mm_macc_ps(dx00,fscal,fix0);
949 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
950 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
951
952 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
953 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
954 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
955
956 /**************************
957 * CALCULATE INTERACTIONS *
958 **************************/
959
960 r10 = _mm_mul_ps(rsq10,rinv10);
961
962 /* Compute parameters for interactions between i and j atoms */
963 qq10 = _mm_mul_ps(iq1,jq0);
964
965 /* EWALD ELECTROSTATICS */
966
967 /* Analytical PME correction */
968 zeta2 = _mm_mul_ps(beta2,rsq10);
969 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
970 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
971 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
972 felec = _mm_mul_ps(qq10,felec);
973
974 fscal = felec;
975
976 /* Update vectorial force */
977 fix1 = _mm_macc_ps(dx10,fscal,fix1);
978 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
979 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
980
981 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
982 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
983 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
984
985 /**************************
986 * CALCULATE INTERACTIONS *
987 **************************/
988
989 r20 = _mm_mul_ps(rsq20,rinv20);
990
991 /* Compute parameters for interactions between i and j atoms */
992 qq20 = _mm_mul_ps(iq2,jq0);
993
994 /* EWALD ELECTROSTATICS */
995
996 /* Analytical PME correction */
997 zeta2 = _mm_mul_ps(beta2,rsq20);
998 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
999 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1000 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1001 felec = _mm_mul_ps(qq20,felec);
1002
1003 fscal = felec;
1004
1005 /* Update vectorial force */
1006 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1007 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1008 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1009
1010 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1011 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1012 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1013
1014 /**************************
1015 * CALCULATE INTERACTIONS *
1016 **************************/
1017
1018 r30 = _mm_mul_ps(rsq30,rinv30);
1019
1020 /* Compute parameters for interactions between i and j atoms */
1021 qq30 = _mm_mul_ps(iq3,jq0);
1022
1023 /* EWALD ELECTROSTATICS */
1024
1025 /* Analytical PME correction */
1026 zeta2 = _mm_mul_ps(beta2,rsq30);
1027 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
1028 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1029 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1030 felec = _mm_mul_ps(qq30,felec);
1031
1032 fscal = felec;
1033
1034 /* Update vectorial force */
1035 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1036 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1037 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1038
1039 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1040 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1041 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1042
1043 fjptrA = f+j_coord_offsetA;
1044 fjptrB = f+j_coord_offsetB;
1045 fjptrC = f+j_coord_offsetC;
1046 fjptrD = f+j_coord_offsetD;
1047
1048 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1049
1050 /* Inner loop uses 131 flops */
1051 }
1052
1053 if(jidx<j_index_end)
1054 {
1055
1056 /* Get j neighbor index, and coordinate index */
1057 jnrlistA = jjnr[jidx];
1058 jnrlistB = jjnr[jidx+1];
1059 jnrlistC = jjnr[jidx+2];
1060 jnrlistD = jjnr[jidx+3];
1061 /* Sign of each element will be negative for non-real atoms.
1062 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1063 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1064 */
1065 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1066 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1067 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1068 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1069 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1070 j_coord_offsetA = DIM*jnrA;
1071 j_coord_offsetB = DIM*jnrB;
1072 j_coord_offsetC = DIM*jnrC;
1073 j_coord_offsetD = DIM*jnrD;
1074
1075 /* load j atom coordinates */
1076 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1077 x+j_coord_offsetC,x+j_coord_offsetD,
1078 &jx0,&jy0,&jz0);
1079
1080 /* Calculate displacement vector */
1081 dx00 = _mm_sub_ps(ix0,jx0);
1082 dy00 = _mm_sub_ps(iy0,jy0);
1083 dz00 = _mm_sub_ps(iz0,jz0);
1084 dx10 = _mm_sub_ps(ix1,jx0);
1085 dy10 = _mm_sub_ps(iy1,jy0);
1086 dz10 = _mm_sub_ps(iz1,jz0);
1087 dx20 = _mm_sub_ps(ix2,jx0);
1088 dy20 = _mm_sub_ps(iy2,jy0);
1089 dz20 = _mm_sub_ps(iz2,jz0);
1090 dx30 = _mm_sub_ps(ix3,jx0);
1091 dy30 = _mm_sub_ps(iy3,jy0);
1092 dz30 = _mm_sub_ps(iz3,jz0);
1093
1094 /* Calculate squared distance and things based on it */
1095 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1096 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1097 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1098 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1099
1100 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1101 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1102 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1103 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1104
1105 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1106 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1107 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1108 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1109
1110 /* Load parameters for j particles */
1111 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1112 charge+jnrC+0,charge+jnrD+0);
1113 vdwjidx0A = 2*vdwtype[jnrA+0];
1114 vdwjidx0B = 2*vdwtype[jnrB+0];
1115 vdwjidx0C = 2*vdwtype[jnrC+0];
1116 vdwjidx0D = 2*vdwtype[jnrD+0];
1117
1118 fjx0 = _mm_setzero_ps();
1119 fjy0 = _mm_setzero_ps();
1120 fjz0 = _mm_setzero_ps();
1121
1122 /**************************
1123 * CALCULATE INTERACTIONS *
1124 **************************/
1125
1126 r00 = _mm_mul_ps(rsq00,rinv00);
1127 r00 = _mm_andnot_ps(dummy_mask,r00);
1128
1129 /* Compute parameters for interactions between i and j atoms */
1130 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1131 vdwparam+vdwioffset0+vdwjidx0B,
1132 vdwparam+vdwioffset0+vdwjidx0C,
1133 vdwparam+vdwioffset0+vdwjidx0D,
1134 &c6_00,&c12_00);
1135
1136 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1137 vdwgridparam+vdwioffset0+vdwjidx0B,
1138 vdwgridparam+vdwioffset0+vdwjidx0C,
1139 vdwgridparam+vdwioffset0+vdwjidx0D);
1140
1141 /* Analytical LJ-PME */
1142 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1143 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
1144 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1145 exponent = gmx_simd_exp_r(ewcljrsq);
1146 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1147 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1148 /* f6A = 6 * C6grid * (1 - poly) */
1149 f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1150 /* f6B = C6grid * exponent * beta^6 */
1151 f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1152 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1153 fvdw = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1154
1155 fscal = fvdw;
1156
1157 fscal = _mm_andnot_ps(dummy_mask,fscal);
1158
1159 /* Update vectorial force */
1160 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1161 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1162 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1163
1164 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1165 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1166 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1167
1168 /**************************
1169 * CALCULATE INTERACTIONS *
1170 **************************/
1171
1172 r10 = _mm_mul_ps(rsq10,rinv10);
1173 r10 = _mm_andnot_ps(dummy_mask,r10);
1174
1175 /* Compute parameters for interactions between i and j atoms */
1176 qq10 = _mm_mul_ps(iq1,jq0);
1177
1178 /* EWALD ELECTROSTATICS */
1179
1180 /* Analytical PME correction */
1181 zeta2 = _mm_mul_ps(beta2,rsq10);
1182 rinv3 = _mm_mul_ps(rinvsq10,rinv10);
1183 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1184 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1185 felec = _mm_mul_ps(qq10,felec);
1186
1187 fscal = felec;
1188
1189 fscal = _mm_andnot_ps(dummy_mask,fscal);
1190
1191 /* Update vectorial force */
1192 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1193 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1194 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1195
1196 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1197 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1198 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1199
1200 /**************************
1201 * CALCULATE INTERACTIONS *
1202 **************************/
1203
1204 r20 = _mm_mul_ps(rsq20,rinv20);
1205 r20 = _mm_andnot_ps(dummy_mask,r20);
1206
1207 /* Compute parameters for interactions between i and j atoms */
1208 qq20 = _mm_mul_ps(iq2,jq0);
1209
1210 /* EWALD ELECTROSTATICS */
1211
1212 /* Analytical PME correction */
1213 zeta2 = _mm_mul_ps(beta2,rsq20);
1214 rinv3 = _mm_mul_ps(rinvsq20,rinv20);
1215 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1216 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1217 felec = _mm_mul_ps(qq20,felec);
1218
1219 fscal = felec;
1220
1221 fscal = _mm_andnot_ps(dummy_mask,fscal);
1222
1223 /* Update vectorial force */
1224 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1225 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1226 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1227
1228 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1229 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1230 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1231
1232 /**************************
1233 * CALCULATE INTERACTIONS *
1234 **************************/
1235
1236 r30 = _mm_mul_ps(rsq30,rinv30);
1237 r30 = _mm_andnot_ps(dummy_mask,r30);
1238
1239 /* Compute parameters for interactions between i and j atoms */
1240 qq30 = _mm_mul_ps(iq3,jq0);
1241
1242 /* EWALD ELECTROSTATICS */
1243
1244 /* Analytical PME correction */
1245 zeta2 = _mm_mul_ps(beta2,rsq30);
1246 rinv3 = _mm_mul_ps(rinvsq30,rinv30);
1247 pmecorrF = gmx_mm_pmecorrF_ps(zeta2);
1248 felec = _mm_macc_ps(pmecorrF,beta3,rinv3);
1249 felec = _mm_mul_ps(qq30,felec);
1250
1251 fscal = felec;
1252
1253 fscal = _mm_andnot_ps(dummy_mask,fscal);
1254
1255 /* Update vectorial force */
1256 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1257 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1258 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1259
1260 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1261 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1262 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1263
1264 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1265 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1266 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1267 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1268
1269 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1270
1271 /* Inner loop uses 135 flops */
1272 }
1273
1274 /* End of innermost loop */
1275
1276 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1277 f+i_coord_offset,fshift+i_shift_offset);
1278
1279 /* Increment number of inner iterations */
1280 inneriter += j_index_end - j_index_start;
1281
1282 /* Outer loop uses 24 flops */
1283 }
1284
1285 /* Increment number of outer iterations */
1286 outeriter += nri;
1287
1288 /* Update outer/inner flops */
1289
1290 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*135);
1291 }
The ultimate molecular dynamics simulation package