?  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_ElecRFCut_VdwCSTab_GeomW4P1_avx_128_fma_single.c
blobae56bc365d213782b26cb40e67595401b7135de8
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_ElecRFCut_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
53 * Electrostatics interaction: ReactionField
54 * VdW interaction: CubicSplineTable
55 * Geometry: Water4-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecRFCut_VdwCSTab_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 __m128i vfitab;
107 __m128i ifour = _mm_set1_epi32(4);
108 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
109 real *vftab;
110 __m128 dummy_mask,cutoff_mask;
111 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
112 __m128 one = _mm_set1_ps(1.0);
113 __m128 two = _mm_set1_ps(2.0);
114 x = xx[0];
115 f = ff[0];
116
117 nri = nlist->nri;
118 iinr = nlist->iinr;
119 jindex = nlist->jindex;
120 jjnr = nlist->jjnr;
121 shiftidx = nlist->shift;
122 gid = nlist->gid;
123 shiftvec = fr->shift_vec[0];
124 fshift = fr->fshift[0];
125 facel = _mm_set1_ps(fr->epsfac);
126 charge = mdatoms->chargeA;
127 krf = _mm_set1_ps(fr->ic->k_rf);
128 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
129 crf = _mm_set1_ps(fr->ic->c_rf);
130 nvdwtype = fr->ntype;
131 vdwparam = fr->nbfp;
132 vdwtype = mdatoms->typeA;
133
134 vftab = kernel_data->table_vdw->data;
135 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
136
137 /* Setup water-specific parameters */
138 inr = nlist->iinr[0];
139 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
140 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
141 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
142 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
143
144 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
145 rcutoff_scalar = fr->rcoulomb;
146 rcutoff = _mm_set1_ps(rcutoff_scalar);
147 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
148
149 /* Avoid stupid compiler warnings */
150 jnrA = jnrB = jnrC = jnrD = 0;
151 j_coord_offsetA = 0;
152 j_coord_offsetB = 0;
153 j_coord_offsetC = 0;
154 j_coord_offsetD = 0;
155
156 outeriter = 0;
157 inneriter = 0;
158
159 for(iidx=0;iidx<4*DIM;iidx++)
160 {
161 scratch[iidx] = 0.0;
162 }
163
164 /* Start outer loop over neighborlists */
165 for(iidx=0; iidx<nri; iidx++)
166 {
167 /* Load shift vector for this list */
168 i_shift_offset = DIM*shiftidx[iidx];
169
170 /* Load limits for loop over neighbors */
171 j_index_start = jindex[iidx];
172 j_index_end = jindex[iidx+1];
173
174 /* Get outer coordinate index */
175 inr = iinr[iidx];
176 i_coord_offset = DIM*inr;
177
178 /* Load i particle coords and add shift vector */
179 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
180 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
181
182 fix0 = _mm_setzero_ps();
183 fiy0 = _mm_setzero_ps();
184 fiz0 = _mm_setzero_ps();
185 fix1 = _mm_setzero_ps();
186 fiy1 = _mm_setzero_ps();
187 fiz1 = _mm_setzero_ps();
188 fix2 = _mm_setzero_ps();
189 fiy2 = _mm_setzero_ps();
190 fiz2 = _mm_setzero_ps();
191 fix3 = _mm_setzero_ps();
192 fiy3 = _mm_setzero_ps();
193 fiz3 = _mm_setzero_ps();
194
195 /* Reset potential sums */
196 velecsum = _mm_setzero_ps();
197 vvdwsum = _mm_setzero_ps();
198
199 /* Start inner kernel loop */
200 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
201 {
202
203 /* Get j neighbor index, and coordinate index */
204 jnrA = jjnr[jidx];
205 jnrB = jjnr[jidx+1];
206 jnrC = jjnr[jidx+2];
207 jnrD = jjnr[jidx+3];
208 j_coord_offsetA = DIM*jnrA;
209 j_coord_offsetB = DIM*jnrB;
210 j_coord_offsetC = DIM*jnrC;
211 j_coord_offsetD = DIM*jnrD;
212
213 /* load j atom coordinates */
214 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
215 x+j_coord_offsetC,x+j_coord_offsetD,
216 &jx0,&jy0,&jz0);
217
218 /* Calculate displacement vector */
219 dx00 = _mm_sub_ps(ix0,jx0);
220 dy00 = _mm_sub_ps(iy0,jy0);
221 dz00 = _mm_sub_ps(iz0,jz0);
222 dx10 = _mm_sub_ps(ix1,jx0);
223 dy10 = _mm_sub_ps(iy1,jy0);
224 dz10 = _mm_sub_ps(iz1,jz0);
225 dx20 = _mm_sub_ps(ix2,jx0);
226 dy20 = _mm_sub_ps(iy2,jy0);
227 dz20 = _mm_sub_ps(iz2,jz0);
228 dx30 = _mm_sub_ps(ix3,jx0);
229 dy30 = _mm_sub_ps(iy3,jy0);
230 dz30 = _mm_sub_ps(iz3,jz0);
231
232 /* Calculate squared distance and things based on it */
233 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
234 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
235 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
236 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
237
238 rinv00 = gmx_mm_invsqrt_ps(rsq00);
239 rinv10 = gmx_mm_invsqrt_ps(rsq10);
240 rinv20 = gmx_mm_invsqrt_ps(rsq20);
241 rinv30 = gmx_mm_invsqrt_ps(rsq30);
242
243 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
244 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
245 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
246
247 /* Load parameters for j particles */
248 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
249 charge+jnrC+0,charge+jnrD+0);
250 vdwjidx0A = 2*vdwtype[jnrA+0];
251 vdwjidx0B = 2*vdwtype[jnrB+0];
252 vdwjidx0C = 2*vdwtype[jnrC+0];
253 vdwjidx0D = 2*vdwtype[jnrD+0];
254
255 fjx0 = _mm_setzero_ps();
256 fjy0 = _mm_setzero_ps();
257 fjz0 = _mm_setzero_ps();
258
259 /**************************
260 * CALCULATE INTERACTIONS *
261 **************************/
262
263 r00 = _mm_mul_ps(rsq00,rinv00);
264
265 /* Compute parameters for interactions between i and j atoms */
266 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
267 vdwparam+vdwioffset0+vdwjidx0B,
268 vdwparam+vdwioffset0+vdwjidx0C,
269 vdwparam+vdwioffset0+vdwjidx0D,
270 &c6_00,&c12_00);
271
272 /* Calculate table index by multiplying r with table scale and truncate to integer */
273 rt = _mm_mul_ps(r00,vftabscale);
274 vfitab = _mm_cvttps_epi32(rt);
275 #ifdef __XOP__
276 vfeps = _mm_frcz_ps(rt);
277 #else
278 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
279 #endif
280 twovfeps = _mm_add_ps(vfeps,vfeps);
281 vfitab = _mm_slli_epi32(vfitab,3);
282
283 /* CUBIC SPLINE TABLE DISPERSION */
284 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
285 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
286 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
287 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
288 _MM_TRANSPOSE4_PS(Y,F,G,H);
289 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
290 VV = _mm_macc_ps(vfeps,Fp,Y);
291 vvdw6 = _mm_mul_ps(c6_00,VV);
292 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
293 fvdw6 = _mm_mul_ps(c6_00,FF);
294
295 /* CUBIC SPLINE TABLE REPULSION */
296 vfitab = _mm_add_epi32(vfitab,ifour);
297 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
298 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
299 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
300 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
301 _MM_TRANSPOSE4_PS(Y,F,G,H);
302 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
303 VV = _mm_macc_ps(vfeps,Fp,Y);
304 vvdw12 = _mm_mul_ps(c12_00,VV);
305 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
306 fvdw12 = _mm_mul_ps(c12_00,FF);
307 vvdw = _mm_add_ps(vvdw12,vvdw6);
308 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
309
310 /* Update potential sum for this i atom from the interaction with this j atom. */
311 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
312
313 fscal = fvdw;
314
315 /* Update vectorial force */
316 fix0 = _mm_macc_ps(dx00,fscal,fix0);
317 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
318 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
319
320 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
321 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
322 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
323
324 /**************************
325 * CALCULATE INTERACTIONS *
326 **************************/
327
328 if (gmx_mm_any_lt(rsq10,rcutoff2))
329 {
330
331 /* Compute parameters for interactions between i and j atoms */
332 qq10 = _mm_mul_ps(iq1,jq0);
333
334 /* REACTION-FIELD ELECTROSTATICS */
335 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
336 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
337
338 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
339
340 /* Update potential sum for this i atom from the interaction with this j atom. */
341 velec = _mm_and_ps(velec,cutoff_mask);
342 velecsum = _mm_add_ps(velecsum,velec);
343
344 fscal = felec;
345
346 fscal = _mm_and_ps(fscal,cutoff_mask);
347
348 /* Update vectorial force */
349 fix1 = _mm_macc_ps(dx10,fscal,fix1);
350 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
351 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
352
353 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
354 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
355 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
356
357 }
358
359 /**************************
360 * CALCULATE INTERACTIONS *
361 **************************/
362
363 if (gmx_mm_any_lt(rsq20,rcutoff2))
364 {
365
366 /* Compute parameters for interactions between i and j atoms */
367 qq20 = _mm_mul_ps(iq2,jq0);
368
369 /* REACTION-FIELD ELECTROSTATICS */
370 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
371 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
372
373 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
374
375 /* Update potential sum for this i atom from the interaction with this j atom. */
376 velec = _mm_and_ps(velec,cutoff_mask);
377 velecsum = _mm_add_ps(velecsum,velec);
378
379 fscal = felec;
380
381 fscal = _mm_and_ps(fscal,cutoff_mask);
382
383 /* Update vectorial force */
384 fix2 = _mm_macc_ps(dx20,fscal,fix2);
385 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
386 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
387
388 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
389 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
390 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
391
392 }
393
394 /**************************
395 * CALCULATE INTERACTIONS *
396 **************************/
397
398 if (gmx_mm_any_lt(rsq30,rcutoff2))
399 {
400
401 /* Compute parameters for interactions between i and j atoms */
402 qq30 = _mm_mul_ps(iq3,jq0);
403
404 /* REACTION-FIELD ELECTROSTATICS */
405 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
406 felec = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
407
408 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
409
410 /* Update potential sum for this i atom from the interaction with this j atom. */
411 velec = _mm_and_ps(velec,cutoff_mask);
412 velecsum = _mm_add_ps(velecsum,velec);
413
414 fscal = felec;
415
416 fscal = _mm_and_ps(fscal,cutoff_mask);
417
418 /* Update vectorial force */
419 fix3 = _mm_macc_ps(dx30,fscal,fix3);
420 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
421 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
422
423 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
424 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
425 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
426
427 }
428
429 fjptrA = f+j_coord_offsetA;
430 fjptrB = f+j_coord_offsetB;
431 fjptrC = f+j_coord_offsetC;
432 fjptrD = f+j_coord_offsetD;
433
434 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
435
436 /* Inner loop uses 176 flops */
437 }
438
439 if(jidx<j_index_end)
440 {
441
442 /* Get j neighbor index, and coordinate index */
443 jnrlistA = jjnr[jidx];
444 jnrlistB = jjnr[jidx+1];
445 jnrlistC = jjnr[jidx+2];
446 jnrlistD = jjnr[jidx+3];
447 /* Sign of each element will be negative for non-real atoms.
448 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
449 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
450 */
451 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
452 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
453 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
454 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
455 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
456 j_coord_offsetA = DIM*jnrA;
457 j_coord_offsetB = DIM*jnrB;
458 j_coord_offsetC = DIM*jnrC;
459 j_coord_offsetD = DIM*jnrD;
460
461 /* load j atom coordinates */
462 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
463 x+j_coord_offsetC,x+j_coord_offsetD,
464 &jx0,&jy0,&jz0);
465
466 /* Calculate displacement vector */
467 dx00 = _mm_sub_ps(ix0,jx0);
468 dy00 = _mm_sub_ps(iy0,jy0);
469 dz00 = _mm_sub_ps(iz0,jz0);
470 dx10 = _mm_sub_ps(ix1,jx0);
471 dy10 = _mm_sub_ps(iy1,jy0);
472 dz10 = _mm_sub_ps(iz1,jz0);
473 dx20 = _mm_sub_ps(ix2,jx0);
474 dy20 = _mm_sub_ps(iy2,jy0);
475 dz20 = _mm_sub_ps(iz2,jz0);
476 dx30 = _mm_sub_ps(ix3,jx0);
477 dy30 = _mm_sub_ps(iy3,jy0);
478 dz30 = _mm_sub_ps(iz3,jz0);
479
480 /* Calculate squared distance and things based on it */
481 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
482 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
483 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
484 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
485
486 rinv00 = gmx_mm_invsqrt_ps(rsq00);
487 rinv10 = gmx_mm_invsqrt_ps(rsq10);
488 rinv20 = gmx_mm_invsqrt_ps(rsq20);
489 rinv30 = gmx_mm_invsqrt_ps(rsq30);
490
491 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
492 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
493 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
494
495 /* Load parameters for j particles */
496 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
497 charge+jnrC+0,charge+jnrD+0);
498 vdwjidx0A = 2*vdwtype[jnrA+0];
499 vdwjidx0B = 2*vdwtype[jnrB+0];
500 vdwjidx0C = 2*vdwtype[jnrC+0];
501 vdwjidx0D = 2*vdwtype[jnrD+0];
502
503 fjx0 = _mm_setzero_ps();
504 fjy0 = _mm_setzero_ps();
505 fjz0 = _mm_setzero_ps();
506
507 /**************************
508 * CALCULATE INTERACTIONS *
509 **************************/
510
511 r00 = _mm_mul_ps(rsq00,rinv00);
512 r00 = _mm_andnot_ps(dummy_mask,r00);
513
514 /* Compute parameters for interactions between i and j atoms */
515 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
516 vdwparam+vdwioffset0+vdwjidx0B,
517 vdwparam+vdwioffset0+vdwjidx0C,
518 vdwparam+vdwioffset0+vdwjidx0D,
519 &c6_00,&c12_00);
520
521 /* Calculate table index by multiplying r with table scale and truncate to integer */
522 rt = _mm_mul_ps(r00,vftabscale);
523 vfitab = _mm_cvttps_epi32(rt);
524 #ifdef __XOP__
525 vfeps = _mm_frcz_ps(rt);
526 #else
527 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
528 #endif
529 twovfeps = _mm_add_ps(vfeps,vfeps);
530 vfitab = _mm_slli_epi32(vfitab,3);
531
532 /* CUBIC SPLINE TABLE DISPERSION */
533 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
534 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
535 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
536 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
537 _MM_TRANSPOSE4_PS(Y,F,G,H);
538 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
539 VV = _mm_macc_ps(vfeps,Fp,Y);
540 vvdw6 = _mm_mul_ps(c6_00,VV);
541 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
542 fvdw6 = _mm_mul_ps(c6_00,FF);
543
544 /* CUBIC SPLINE TABLE REPULSION */
545 vfitab = _mm_add_epi32(vfitab,ifour);
546 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
547 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
548 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
549 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
550 _MM_TRANSPOSE4_PS(Y,F,G,H);
551 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
552 VV = _mm_macc_ps(vfeps,Fp,Y);
553 vvdw12 = _mm_mul_ps(c12_00,VV);
554 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
555 fvdw12 = _mm_mul_ps(c12_00,FF);
556 vvdw = _mm_add_ps(vvdw12,vvdw6);
557 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
558
559 /* Update potential sum for this i atom from the interaction with this j atom. */
560 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
561 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
562
563 fscal = fvdw;
564
565 fscal = _mm_andnot_ps(dummy_mask,fscal);
566
567 /* Update vectorial force */
568 fix0 = _mm_macc_ps(dx00,fscal,fix0);
569 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
570 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
571
572 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
573 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
574 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
575
576 /**************************
577 * CALCULATE INTERACTIONS *
578 **************************/
579
580 if (gmx_mm_any_lt(rsq10,rcutoff2))
581 {
582
583 /* Compute parameters for interactions between i and j atoms */
584 qq10 = _mm_mul_ps(iq1,jq0);
585
586 /* REACTION-FIELD ELECTROSTATICS */
587 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
588 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
589
590 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
591
592 /* Update potential sum for this i atom from the interaction with this j atom. */
593 velec = _mm_and_ps(velec,cutoff_mask);
594 velec = _mm_andnot_ps(dummy_mask,velec);
595 velecsum = _mm_add_ps(velecsum,velec);
596
597 fscal = felec;
598
599 fscal = _mm_and_ps(fscal,cutoff_mask);
600
601 fscal = _mm_andnot_ps(dummy_mask,fscal);
602
603 /* Update vectorial force */
604 fix1 = _mm_macc_ps(dx10,fscal,fix1);
605 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
606 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
607
608 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
609 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
610 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
611
612 }
613
614 /**************************
615 * CALCULATE INTERACTIONS *
616 **************************/
617
618 if (gmx_mm_any_lt(rsq20,rcutoff2))
619 {
620
621 /* Compute parameters for interactions between i and j atoms */
622 qq20 = _mm_mul_ps(iq2,jq0);
623
624 /* REACTION-FIELD ELECTROSTATICS */
625 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
626 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
627
628 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
629
630 /* Update potential sum for this i atom from the interaction with this j atom. */
631 velec = _mm_and_ps(velec,cutoff_mask);
632 velec = _mm_andnot_ps(dummy_mask,velec);
633 velecsum = _mm_add_ps(velecsum,velec);
634
635 fscal = felec;
636
637 fscal = _mm_and_ps(fscal,cutoff_mask);
638
639 fscal = _mm_andnot_ps(dummy_mask,fscal);
640
641 /* Update vectorial force */
642 fix2 = _mm_macc_ps(dx20,fscal,fix2);
643 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
644 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
645
646 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
647 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
648 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
649
650 }
651
652 /**************************
653 * CALCULATE INTERACTIONS *
654 **************************/
655
656 if (gmx_mm_any_lt(rsq30,rcutoff2))
657 {
658
659 /* Compute parameters for interactions between i and j atoms */
660 qq30 = _mm_mul_ps(iq3,jq0);
661
662 /* REACTION-FIELD ELECTROSTATICS */
663 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
664 felec = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
665
666 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
667
668 /* Update potential sum for this i atom from the interaction with this j atom. */
669 velec = _mm_and_ps(velec,cutoff_mask);
670 velec = _mm_andnot_ps(dummy_mask,velec);
671 velecsum = _mm_add_ps(velecsum,velec);
672
673 fscal = felec;
674
675 fscal = _mm_and_ps(fscal,cutoff_mask);
676
677 fscal = _mm_andnot_ps(dummy_mask,fscal);
678
679 /* Update vectorial force */
680 fix3 = _mm_macc_ps(dx30,fscal,fix3);
681 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
682 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
683
684 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
685 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
686 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
687
688 }
689
690 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
691 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
692 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
693 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
694
695 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
696
697 /* Inner loop uses 177 flops */
698 }
699
700 /* End of innermost loop */
701
702 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
703 f+i_coord_offset,fshift+i_shift_offset);
704
705 ggid = gid[iidx];
706 /* Update potential energies */
707 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
708 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
709
710 /* Increment number of inner iterations */
711 inneriter += j_index_end - j_index_start;
712
713 /* Outer loop uses 26 flops */
714 }
715
716 /* Increment number of outer iterations */
717 outeriter += nri;
718
719 /* Update outer/inner flops */
720
721 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*177);
722 }
723 /*
724 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_single
725 * Electrostatics interaction: ReactionField
726 * VdW interaction: CubicSplineTable
727 * Geometry: Water4-Particle
728 * Calculate force/pot: Force
729 */
730 void
731 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_single
732 (t_nblist * gmx_restrict nlist,
733 rvec * gmx_restrict xx,
734 rvec * gmx_restrict ff,
735 t_forcerec * gmx_restrict fr,
736 t_mdatoms * gmx_restrict mdatoms,
737 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
738 t_nrnb * gmx_restrict nrnb)
739 {
740 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
741 * just 0 for non-waters.
742 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
743 * jnr indices corresponding to data put in the four positions in the SIMD register.
744 */
745 int i_shift_offset,i_coord_offset,outeriter,inneriter;
746 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
747 int jnrA,jnrB,jnrC,jnrD;
748 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
749 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
750 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
751 real rcutoff_scalar;
752 real *shiftvec,*fshift,*x,*f;
753 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
754 real scratch[4*DIM];
755 __m128 fscal,rcutoff,rcutoff2,jidxall;
756 int vdwioffset0;
757 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
758 int vdwioffset1;
759 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
760 int vdwioffset2;
761 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
762 int vdwioffset3;
763 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
764 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
765 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
766 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
767 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
768 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
769 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
770 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
771 real *charge;
772 int nvdwtype;
773 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
774 int *vdwtype;
775 real *vdwparam;
776 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
777 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
778 __m128i vfitab;
779 __m128i ifour = _mm_set1_epi32(4);
780 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
781 real *vftab;
782 __m128 dummy_mask,cutoff_mask;
783 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
784 __m128 one = _mm_set1_ps(1.0);
785 __m128 two = _mm_set1_ps(2.0);
786 x = xx[0];
787 f = ff[0];
788
789 nri = nlist->nri;
790 iinr = nlist->iinr;
791 jindex = nlist->jindex;
792 jjnr = nlist->jjnr;
793 shiftidx = nlist->shift;
794 gid = nlist->gid;
795 shiftvec = fr->shift_vec[0];
796 fshift = fr->fshift[0];
797 facel = _mm_set1_ps(fr->epsfac);
798 charge = mdatoms->chargeA;
799 krf = _mm_set1_ps(fr->ic->k_rf);
800 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
801 crf = _mm_set1_ps(fr->ic->c_rf);
802 nvdwtype = fr->ntype;
803 vdwparam = fr->nbfp;
804 vdwtype = mdatoms->typeA;
805
806 vftab = kernel_data->table_vdw->data;
807 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
808
809 /* Setup water-specific parameters */
810 inr = nlist->iinr[0];
811 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
812 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
813 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
814 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
815
816 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
817 rcutoff_scalar = fr->rcoulomb;
818 rcutoff = _mm_set1_ps(rcutoff_scalar);
819 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
820
821 /* Avoid stupid compiler warnings */
822 jnrA = jnrB = jnrC = jnrD = 0;
823 j_coord_offsetA = 0;
824 j_coord_offsetB = 0;
825 j_coord_offsetC = 0;
826 j_coord_offsetD = 0;
827
828 outeriter = 0;
829 inneriter = 0;
830
831 for(iidx=0;iidx<4*DIM;iidx++)
832 {
833 scratch[iidx] = 0.0;
834 }
835
836 /* Start outer loop over neighborlists */
837 for(iidx=0; iidx<nri; iidx++)
838 {
839 /* Load shift vector for this list */
840 i_shift_offset = DIM*shiftidx[iidx];
841
842 /* Load limits for loop over neighbors */
843 j_index_start = jindex[iidx];
844 j_index_end = jindex[iidx+1];
845
846 /* Get outer coordinate index */
847 inr = iinr[iidx];
848 i_coord_offset = DIM*inr;
849
850 /* Load i particle coords and add shift vector */
851 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
852 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
853
854 fix0 = _mm_setzero_ps();
855 fiy0 = _mm_setzero_ps();
856 fiz0 = _mm_setzero_ps();
857 fix1 = _mm_setzero_ps();
858 fiy1 = _mm_setzero_ps();
859 fiz1 = _mm_setzero_ps();
860 fix2 = _mm_setzero_ps();
861 fiy2 = _mm_setzero_ps();
862 fiz2 = _mm_setzero_ps();
863 fix3 = _mm_setzero_ps();
864 fiy3 = _mm_setzero_ps();
865 fiz3 = _mm_setzero_ps();
866
867 /* Start inner kernel loop */
868 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
869 {
870
871 /* Get j neighbor index, and coordinate index */
872 jnrA = jjnr[jidx];
873 jnrB = jjnr[jidx+1];
874 jnrC = jjnr[jidx+2];
875 jnrD = jjnr[jidx+3];
876 j_coord_offsetA = DIM*jnrA;
877 j_coord_offsetB = DIM*jnrB;
878 j_coord_offsetC = DIM*jnrC;
879 j_coord_offsetD = DIM*jnrD;
880
881 /* load j atom coordinates */
882 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
883 x+j_coord_offsetC,x+j_coord_offsetD,
884 &jx0,&jy0,&jz0);
885
886 /* Calculate displacement vector */
887 dx00 = _mm_sub_ps(ix0,jx0);
888 dy00 = _mm_sub_ps(iy0,jy0);
889 dz00 = _mm_sub_ps(iz0,jz0);
890 dx10 = _mm_sub_ps(ix1,jx0);
891 dy10 = _mm_sub_ps(iy1,jy0);
892 dz10 = _mm_sub_ps(iz1,jz0);
893 dx20 = _mm_sub_ps(ix2,jx0);
894 dy20 = _mm_sub_ps(iy2,jy0);
895 dz20 = _mm_sub_ps(iz2,jz0);
896 dx30 = _mm_sub_ps(ix3,jx0);
897 dy30 = _mm_sub_ps(iy3,jy0);
898 dz30 = _mm_sub_ps(iz3,jz0);
899
900 /* Calculate squared distance and things based on it */
901 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
902 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
903 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
904 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
905
906 rinv00 = gmx_mm_invsqrt_ps(rsq00);
907 rinv10 = gmx_mm_invsqrt_ps(rsq10);
908 rinv20 = gmx_mm_invsqrt_ps(rsq20);
909 rinv30 = gmx_mm_invsqrt_ps(rsq30);
910
911 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
912 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
913 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
914
915 /* Load parameters for j particles */
916 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
917 charge+jnrC+0,charge+jnrD+0);
918 vdwjidx0A = 2*vdwtype[jnrA+0];
919 vdwjidx0B = 2*vdwtype[jnrB+0];
920 vdwjidx0C = 2*vdwtype[jnrC+0];
921 vdwjidx0D = 2*vdwtype[jnrD+0];
922
923 fjx0 = _mm_setzero_ps();
924 fjy0 = _mm_setzero_ps();
925 fjz0 = _mm_setzero_ps();
926
927 /**************************
928 * CALCULATE INTERACTIONS *
929 **************************/
930
931 r00 = _mm_mul_ps(rsq00,rinv00);
932
933 /* Compute parameters for interactions between i and j atoms */
934 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
935 vdwparam+vdwioffset0+vdwjidx0B,
936 vdwparam+vdwioffset0+vdwjidx0C,
937 vdwparam+vdwioffset0+vdwjidx0D,
938 &c6_00,&c12_00);
939
940 /* Calculate table index by multiplying r with table scale and truncate to integer */
941 rt = _mm_mul_ps(r00,vftabscale);
942 vfitab = _mm_cvttps_epi32(rt);
943 #ifdef __XOP__
944 vfeps = _mm_frcz_ps(rt);
945 #else
946 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
947 #endif
948 twovfeps = _mm_add_ps(vfeps,vfeps);
949 vfitab = _mm_slli_epi32(vfitab,3);
950
951 /* CUBIC SPLINE TABLE DISPERSION */
952 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
953 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
954 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
955 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
956 _MM_TRANSPOSE4_PS(Y,F,G,H);
957 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
958 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
959 fvdw6 = _mm_mul_ps(c6_00,FF);
960
961 /* CUBIC SPLINE TABLE REPULSION */
962 vfitab = _mm_add_epi32(vfitab,ifour);
963 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
964 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
965 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
966 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
967 _MM_TRANSPOSE4_PS(Y,F,G,H);
968 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
969 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
970 fvdw12 = _mm_mul_ps(c12_00,FF);
971 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
972
973 fscal = fvdw;
974
975 /* Update vectorial force */
976 fix0 = _mm_macc_ps(dx00,fscal,fix0);
977 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
978 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
979
980 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
981 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
982 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
983
984 /**************************
985 * CALCULATE INTERACTIONS *
986 **************************/
987
988 if (gmx_mm_any_lt(rsq10,rcutoff2))
989 {
990
991 /* Compute parameters for interactions between i and j atoms */
992 qq10 = _mm_mul_ps(iq1,jq0);
993
994 /* REACTION-FIELD ELECTROSTATICS */
995 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
996
997 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
998
999 fscal = felec;
1000
1001 fscal = _mm_and_ps(fscal,cutoff_mask);
1002
1003 /* Update vectorial force */
1004 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1005 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1006 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1007
1008 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1009 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1010 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1011
1012 }
1013
1014 /**************************
1015 * CALCULATE INTERACTIONS *
1016 **************************/
1017
1018 if (gmx_mm_any_lt(rsq20,rcutoff2))
1019 {
1020
1021 /* Compute parameters for interactions between i and j atoms */
1022 qq20 = _mm_mul_ps(iq2,jq0);
1023
1024 /* REACTION-FIELD ELECTROSTATICS */
1025 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1026
1027 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1028
1029 fscal = felec;
1030
1031 fscal = _mm_and_ps(fscal,cutoff_mask);
1032
1033 /* Update vectorial force */
1034 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1035 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1036 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1037
1038 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1039 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1040 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1041
1042 }
1043
1044 /**************************
1045 * CALCULATE INTERACTIONS *
1046 **************************/
1047
1048 if (gmx_mm_any_lt(rsq30,rcutoff2))
1049 {
1050
1051 /* Compute parameters for interactions between i and j atoms */
1052 qq30 = _mm_mul_ps(iq3,jq0);
1053
1054 /* REACTION-FIELD ELECTROSTATICS */
1055 felec = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1056
1057 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1058
1059 fscal = felec;
1060
1061 fscal = _mm_and_ps(fscal,cutoff_mask);
1062
1063 /* Update vectorial force */
1064 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1065 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1066 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1067
1068 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1069 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1070 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1071
1072 }
1073
1074 fjptrA = f+j_coord_offsetA;
1075 fjptrB = f+j_coord_offsetB;
1076 fjptrC = f+j_coord_offsetC;
1077 fjptrD = f+j_coord_offsetD;
1078
1079 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1080
1081 /* Inner loop uses 150 flops */
1082 }
1083
1084 if(jidx<j_index_end)
1085 {
1086
1087 /* Get j neighbor index, and coordinate index */
1088 jnrlistA = jjnr[jidx];
1089 jnrlistB = jjnr[jidx+1];
1090 jnrlistC = jjnr[jidx+2];
1091 jnrlistD = jjnr[jidx+3];
1092 /* Sign of each element will be negative for non-real atoms.
1093 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1094 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1095 */
1096 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1097 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1098 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1099 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1100 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1101 j_coord_offsetA = DIM*jnrA;
1102 j_coord_offsetB = DIM*jnrB;
1103 j_coord_offsetC = DIM*jnrC;
1104 j_coord_offsetD = DIM*jnrD;
1105
1106 /* load j atom coordinates */
1107 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1108 x+j_coord_offsetC,x+j_coord_offsetD,
1109 &jx0,&jy0,&jz0);
1110
1111 /* Calculate displacement vector */
1112 dx00 = _mm_sub_ps(ix0,jx0);
1113 dy00 = _mm_sub_ps(iy0,jy0);
1114 dz00 = _mm_sub_ps(iz0,jz0);
1115 dx10 = _mm_sub_ps(ix1,jx0);
1116 dy10 = _mm_sub_ps(iy1,jy0);
1117 dz10 = _mm_sub_ps(iz1,jz0);
1118 dx20 = _mm_sub_ps(ix2,jx0);
1119 dy20 = _mm_sub_ps(iy2,jy0);
1120 dz20 = _mm_sub_ps(iz2,jz0);
1121 dx30 = _mm_sub_ps(ix3,jx0);
1122 dy30 = _mm_sub_ps(iy3,jy0);
1123 dz30 = _mm_sub_ps(iz3,jz0);
1124
1125 /* Calculate squared distance and things based on it */
1126 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1127 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1128 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1129 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1130
1131 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1132 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1133 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1134 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1135
1136 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1137 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1138 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1139
1140 /* Load parameters for j particles */
1141 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1142 charge+jnrC+0,charge+jnrD+0);
1143 vdwjidx0A = 2*vdwtype[jnrA+0];
1144 vdwjidx0B = 2*vdwtype[jnrB+0];
1145 vdwjidx0C = 2*vdwtype[jnrC+0];
1146 vdwjidx0D = 2*vdwtype[jnrD+0];
1147
1148 fjx0 = _mm_setzero_ps();
1149 fjy0 = _mm_setzero_ps();
1150 fjz0 = _mm_setzero_ps();
1151
1152 /**************************
1153 * CALCULATE INTERACTIONS *
1154 **************************/
1155
1156 r00 = _mm_mul_ps(rsq00,rinv00);
1157 r00 = _mm_andnot_ps(dummy_mask,r00);
1158
1159 /* Compute parameters for interactions between i and j atoms */
1160 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1161 vdwparam+vdwioffset0+vdwjidx0B,
1162 vdwparam+vdwioffset0+vdwjidx0C,
1163 vdwparam+vdwioffset0+vdwjidx0D,
1164 &c6_00,&c12_00);
1165
1166 /* Calculate table index by multiplying r with table scale and truncate to integer */
1167 rt = _mm_mul_ps(r00,vftabscale);
1168 vfitab = _mm_cvttps_epi32(rt);
1169 #ifdef __XOP__
1170 vfeps = _mm_frcz_ps(rt);
1171 #else
1172 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1173 #endif
1174 twovfeps = _mm_add_ps(vfeps,vfeps);
1175 vfitab = _mm_slli_epi32(vfitab,3);
1176
1177 /* CUBIC SPLINE TABLE DISPERSION */
1178 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1179 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1180 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1181 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1182 _MM_TRANSPOSE4_PS(Y,F,G,H);
1183 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1184 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1185 fvdw6 = _mm_mul_ps(c6_00,FF);
1186
1187 /* CUBIC SPLINE TABLE REPULSION */
1188 vfitab = _mm_add_epi32(vfitab,ifour);
1189 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1190 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1191 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1192 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1193 _MM_TRANSPOSE4_PS(Y,F,G,H);
1194 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1195 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1196 fvdw12 = _mm_mul_ps(c12_00,FF);
1197 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1198
1199 fscal = fvdw;
1200
1201 fscal = _mm_andnot_ps(dummy_mask,fscal);
1202
1203 /* Update vectorial force */
1204 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1205 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1206 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1207
1208 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1209 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1210 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1211
1212 /**************************
1213 * CALCULATE INTERACTIONS *
1214 **************************/
1215
1216 if (gmx_mm_any_lt(rsq10,rcutoff2))
1217 {
1218
1219 /* Compute parameters for interactions between i and j atoms */
1220 qq10 = _mm_mul_ps(iq1,jq0);
1221
1222 /* REACTION-FIELD ELECTROSTATICS */
1223 felec = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1224
1225 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1226
1227 fscal = felec;
1228
1229 fscal = _mm_and_ps(fscal,cutoff_mask);
1230
1231 fscal = _mm_andnot_ps(dummy_mask,fscal);
1232
1233 /* Update vectorial force */
1234 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1235 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1236 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1237
1238 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1239 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1240 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1241
1242 }
1243
1244 /**************************
1245 * CALCULATE INTERACTIONS *
1246 **************************/
1247
1248 if (gmx_mm_any_lt(rsq20,rcutoff2))
1249 {
1250
1251 /* Compute parameters for interactions between i and j atoms */
1252 qq20 = _mm_mul_ps(iq2,jq0);
1253
1254 /* REACTION-FIELD ELECTROSTATICS */
1255 felec = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1256
1257 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1258
1259 fscal = felec;
1260
1261 fscal = _mm_and_ps(fscal,cutoff_mask);
1262
1263 fscal = _mm_andnot_ps(dummy_mask,fscal);
1264
1265 /* Update vectorial force */
1266 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1267 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1268 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1269
1270 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1271 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1272 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1273
1274 }
1275
1276 /**************************
1277 * CALCULATE INTERACTIONS *
1278 **************************/
1279
1280 if (gmx_mm_any_lt(rsq30,rcutoff2))
1281 {
1282
1283 /* Compute parameters for interactions between i and j atoms */
1284 qq30 = _mm_mul_ps(iq3,jq0);
1285
1286 /* REACTION-FIELD ELECTROSTATICS */
1287 felec = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1288
1289 cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2);
1290
1291 fscal = felec;
1292
1293 fscal = _mm_and_ps(fscal,cutoff_mask);
1294
1295 fscal = _mm_andnot_ps(dummy_mask,fscal);
1296
1297 /* Update vectorial force */
1298 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1299 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1300 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1301
1302 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1303 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1304 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1305
1306 }
1307
1308 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1309 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1310 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1311 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1312
1313 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1314
1315 /* Inner loop uses 151 flops */
1316 }
1317
1318 /* End of innermost loop */
1319
1320 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1321 f+i_coord_offset,fshift+i_shift_offset);
1322
1323 /* Increment number of inner iterations */
1324 inneriter += j_index_end - j_index_start;
1325
1326 /* Outer loop uses 24 flops */
1327 }
1328
1329 /* Increment number of outer iterations */
1330 outeriter += nri;
1331
1332 /* Update outer/inner flops */
1333
1334 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*151);
1335 }
The ultimate molecular dynamics simulation package