?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed simple.h from nb_kernel_sse2_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse2_single / nb_kernel_ElecEw_VdwNone_GeomW3P1_sse2_single.c
blobb73fbb873a7a12bbdbdcf6e5e8645f5a4725d9d6
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 sse2_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_sse2_single.h"
49 #include "kernelutil_x86_sse2_single.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse2_single
53 * Electrostatics interaction: Ewald
54 * VdW interaction: None
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse2_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 SSE, 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 tx,ty,tz,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 vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
94 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
95 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
96 real *charge;
97 __m128i ewitab;
98 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
99 real *ewtab;
100 __m128 dummy_mask,cutoff_mask;
101 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
102 __m128 one = _mm_set1_ps(1.0);
103 __m128 two = _mm_set1_ps(2.0);
104 x = xx[0];
105 f = ff[0];
106
107 nri = nlist->nri;
108 iinr = nlist->iinr;
109 jindex = nlist->jindex;
110 jjnr = nlist->jjnr;
111 shiftidx = nlist->shift;
112 gid = nlist->gid;
113 shiftvec = fr->shift_vec[0];
114 fshift = fr->fshift[0];
115 facel = _mm_set1_ps(fr->epsfac);
116 charge = mdatoms->chargeA;
117
118 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
119 ewtab = fr->ic->tabq_coul_FDV0;
120 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
121 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
122
123 /* Setup water-specific parameters */
124 inr = nlist->iinr[0];
125 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
126 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
127 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
128
129 /* Avoid stupid compiler warnings */
130 jnrA = jnrB = jnrC = jnrD = 0;
131 j_coord_offsetA = 0;
132 j_coord_offsetB = 0;
133 j_coord_offsetC = 0;
134 j_coord_offsetD = 0;
135
136 outeriter = 0;
137 inneriter = 0;
138
139 for(iidx=0;iidx<4*DIM;iidx++)
140 {
141 scratch[iidx] = 0.0;
142 }
143
144 /* Start outer loop over neighborlists */
145 for(iidx=0; iidx<nri; iidx++)
146 {
147 /* Load shift vector for this list */
148 i_shift_offset = DIM*shiftidx[iidx];
149
150 /* Load limits for loop over neighbors */
151 j_index_start = jindex[iidx];
152 j_index_end = jindex[iidx+1];
153
154 /* Get outer coordinate index */
155 inr = iinr[iidx];
156 i_coord_offset = DIM*inr;
157
158 /* Load i particle coords and add shift vector */
159 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
160 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
161
162 fix0 = _mm_setzero_ps();
163 fiy0 = _mm_setzero_ps();
164 fiz0 = _mm_setzero_ps();
165 fix1 = _mm_setzero_ps();
166 fiy1 = _mm_setzero_ps();
167 fiz1 = _mm_setzero_ps();
168 fix2 = _mm_setzero_ps();
169 fiy2 = _mm_setzero_ps();
170 fiz2 = _mm_setzero_ps();
171
172 /* Reset potential sums */
173 velecsum = _mm_setzero_ps();
174
175 /* Start inner kernel loop */
176 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
177 {
178
179 /* Get j neighbor index, and coordinate index */
180 jnrA = jjnr[jidx];
181 jnrB = jjnr[jidx+1];
182 jnrC = jjnr[jidx+2];
183 jnrD = jjnr[jidx+3];
184 j_coord_offsetA = DIM*jnrA;
185 j_coord_offsetB = DIM*jnrB;
186 j_coord_offsetC = DIM*jnrC;
187 j_coord_offsetD = DIM*jnrD;
188
189 /* load j atom coordinates */
190 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
191 x+j_coord_offsetC,x+j_coord_offsetD,
192 &jx0,&jy0,&jz0);
193
194 /* Calculate displacement vector */
195 dx00 = _mm_sub_ps(ix0,jx0);
196 dy00 = _mm_sub_ps(iy0,jy0);
197 dz00 = _mm_sub_ps(iz0,jz0);
198 dx10 = _mm_sub_ps(ix1,jx0);
199 dy10 = _mm_sub_ps(iy1,jy0);
200 dz10 = _mm_sub_ps(iz1,jz0);
201 dx20 = _mm_sub_ps(ix2,jx0);
202 dy20 = _mm_sub_ps(iy2,jy0);
203 dz20 = _mm_sub_ps(iz2,jz0);
204
205 /* Calculate squared distance and things based on it */
206 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
207 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
208 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
209
210 rinv00 = gmx_mm_invsqrt_ps(rsq00);
211 rinv10 = gmx_mm_invsqrt_ps(rsq10);
212 rinv20 = gmx_mm_invsqrt_ps(rsq20);
213
214 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
215 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
216 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
217
218 /* Load parameters for j particles */
219 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
220 charge+jnrC+0,charge+jnrD+0);
221
222 fjx0 = _mm_setzero_ps();
223 fjy0 = _mm_setzero_ps();
224 fjz0 = _mm_setzero_ps();
225
226 /**************************
227 * CALCULATE INTERACTIONS *
228 **************************/
229
230 r00 = _mm_mul_ps(rsq00,rinv00);
231
232 /* Compute parameters for interactions between i and j atoms */
233 qq00 = _mm_mul_ps(iq0,jq0);
234
235 /* EWALD ELECTROSTATICS */
236
237 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
238 ewrt = _mm_mul_ps(r00,ewtabscale);
239 ewitab = _mm_cvttps_epi32(ewrt);
240 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
241 ewitab = _mm_slli_epi32(ewitab,2);
242 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
243 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
244 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
245 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
246 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
247 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
248 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
249 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
250 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
251
252 /* Update potential sum for this i atom from the interaction with this j atom. */
253 velecsum = _mm_add_ps(velecsum,velec);
254
255 fscal = felec;
256
257 /* Calculate temporary vectorial force */
258 tx = _mm_mul_ps(fscal,dx00);
259 ty = _mm_mul_ps(fscal,dy00);
260 tz = _mm_mul_ps(fscal,dz00);
261
262 /* Update vectorial force */
263 fix0 = _mm_add_ps(fix0,tx);
264 fiy0 = _mm_add_ps(fiy0,ty);
265 fiz0 = _mm_add_ps(fiz0,tz);
266
267 fjx0 = _mm_add_ps(fjx0,tx);
268 fjy0 = _mm_add_ps(fjy0,ty);
269 fjz0 = _mm_add_ps(fjz0,tz);
270
271 /**************************
272 * CALCULATE INTERACTIONS *
273 **************************/
274
275 r10 = _mm_mul_ps(rsq10,rinv10);
276
277 /* Compute parameters for interactions between i and j atoms */
278 qq10 = _mm_mul_ps(iq1,jq0);
279
280 /* EWALD ELECTROSTATICS */
281
282 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
283 ewrt = _mm_mul_ps(r10,ewtabscale);
284 ewitab = _mm_cvttps_epi32(ewrt);
285 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
286 ewitab = _mm_slli_epi32(ewitab,2);
287 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
288 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
289 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
290 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
291 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
292 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
293 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
294 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
295 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
296
297 /* Update potential sum for this i atom from the interaction with this j atom. */
298 velecsum = _mm_add_ps(velecsum,velec);
299
300 fscal = felec;
301
302 /* Calculate temporary vectorial force */
303 tx = _mm_mul_ps(fscal,dx10);
304 ty = _mm_mul_ps(fscal,dy10);
305 tz = _mm_mul_ps(fscal,dz10);
306
307 /* Update vectorial force */
308 fix1 = _mm_add_ps(fix1,tx);
309 fiy1 = _mm_add_ps(fiy1,ty);
310 fiz1 = _mm_add_ps(fiz1,tz);
311
312 fjx0 = _mm_add_ps(fjx0,tx);
313 fjy0 = _mm_add_ps(fjy0,ty);
314 fjz0 = _mm_add_ps(fjz0,tz);
315
316 /**************************
317 * CALCULATE INTERACTIONS *
318 **************************/
319
320 r20 = _mm_mul_ps(rsq20,rinv20);
321
322 /* Compute parameters for interactions between i and j atoms */
323 qq20 = _mm_mul_ps(iq2,jq0);
324
325 /* EWALD ELECTROSTATICS */
326
327 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
328 ewrt = _mm_mul_ps(r20,ewtabscale);
329 ewitab = _mm_cvttps_epi32(ewrt);
330 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
331 ewitab = _mm_slli_epi32(ewitab,2);
332 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
333 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
334 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
335 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
336 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
337 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
338 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
339 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
340 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
341
342 /* Update potential sum for this i atom from the interaction with this j atom. */
343 velecsum = _mm_add_ps(velecsum,velec);
344
345 fscal = felec;
346
347 /* Calculate temporary vectorial force */
348 tx = _mm_mul_ps(fscal,dx20);
349 ty = _mm_mul_ps(fscal,dy20);
350 tz = _mm_mul_ps(fscal,dz20);
351
352 /* Update vectorial force */
353 fix2 = _mm_add_ps(fix2,tx);
354 fiy2 = _mm_add_ps(fiy2,ty);
355 fiz2 = _mm_add_ps(fiz2,tz);
356
357 fjx0 = _mm_add_ps(fjx0,tx);
358 fjy0 = _mm_add_ps(fjy0,ty);
359 fjz0 = _mm_add_ps(fjz0,tz);
360
361 fjptrA = f+j_coord_offsetA;
362 fjptrB = f+j_coord_offsetB;
363 fjptrC = f+j_coord_offsetC;
364 fjptrD = f+j_coord_offsetD;
365
366 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
367
368 /* Inner loop uses 123 flops */
369 }
370
371 if(jidx<j_index_end)
372 {
373
374 /* Get j neighbor index, and coordinate index */
375 jnrlistA = jjnr[jidx];
376 jnrlistB = jjnr[jidx+1];
377 jnrlistC = jjnr[jidx+2];
378 jnrlistD = jjnr[jidx+3];
379 /* Sign of each element will be negative for non-real atoms.
380 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
381 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
382 */
383 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
384 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
385 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
386 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
387 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
388 j_coord_offsetA = DIM*jnrA;
389 j_coord_offsetB = DIM*jnrB;
390 j_coord_offsetC = DIM*jnrC;
391 j_coord_offsetD = DIM*jnrD;
392
393 /* load j atom coordinates */
394 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
395 x+j_coord_offsetC,x+j_coord_offsetD,
396 &jx0,&jy0,&jz0);
397
398 /* Calculate displacement vector */
399 dx00 = _mm_sub_ps(ix0,jx0);
400 dy00 = _mm_sub_ps(iy0,jy0);
401 dz00 = _mm_sub_ps(iz0,jz0);
402 dx10 = _mm_sub_ps(ix1,jx0);
403 dy10 = _mm_sub_ps(iy1,jy0);
404 dz10 = _mm_sub_ps(iz1,jz0);
405 dx20 = _mm_sub_ps(ix2,jx0);
406 dy20 = _mm_sub_ps(iy2,jy0);
407 dz20 = _mm_sub_ps(iz2,jz0);
408
409 /* Calculate squared distance and things based on it */
410 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
411 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
412 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
413
414 rinv00 = gmx_mm_invsqrt_ps(rsq00);
415 rinv10 = gmx_mm_invsqrt_ps(rsq10);
416 rinv20 = gmx_mm_invsqrt_ps(rsq20);
417
418 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
419 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
420 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
421
422 /* Load parameters for j particles */
423 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
424 charge+jnrC+0,charge+jnrD+0);
425
426 fjx0 = _mm_setzero_ps();
427 fjy0 = _mm_setzero_ps();
428 fjz0 = _mm_setzero_ps();
429
430 /**************************
431 * CALCULATE INTERACTIONS *
432 **************************/
433
434 r00 = _mm_mul_ps(rsq00,rinv00);
435 r00 = _mm_andnot_ps(dummy_mask,r00);
436
437 /* Compute parameters for interactions between i and j atoms */
438 qq00 = _mm_mul_ps(iq0,jq0);
439
440 /* EWALD ELECTROSTATICS */
441
442 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
443 ewrt = _mm_mul_ps(r00,ewtabscale);
444 ewitab = _mm_cvttps_epi32(ewrt);
445 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
446 ewitab = _mm_slli_epi32(ewitab,2);
447 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
448 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
449 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
450 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
451 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
452 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
453 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
454 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
455 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
456
457 /* Update potential sum for this i atom from the interaction with this j atom. */
458 velec = _mm_andnot_ps(dummy_mask,velec);
459 velecsum = _mm_add_ps(velecsum,velec);
460
461 fscal = felec;
462
463 fscal = _mm_andnot_ps(dummy_mask,fscal);
464
465 /* Calculate temporary vectorial force */
466 tx = _mm_mul_ps(fscal,dx00);
467 ty = _mm_mul_ps(fscal,dy00);
468 tz = _mm_mul_ps(fscal,dz00);
469
470 /* Update vectorial force */
471 fix0 = _mm_add_ps(fix0,tx);
472 fiy0 = _mm_add_ps(fiy0,ty);
473 fiz0 = _mm_add_ps(fiz0,tz);
474
475 fjx0 = _mm_add_ps(fjx0,tx);
476 fjy0 = _mm_add_ps(fjy0,ty);
477 fjz0 = _mm_add_ps(fjz0,tz);
478
479 /**************************
480 * CALCULATE INTERACTIONS *
481 **************************/
482
483 r10 = _mm_mul_ps(rsq10,rinv10);
484 r10 = _mm_andnot_ps(dummy_mask,r10);
485
486 /* Compute parameters for interactions between i and j atoms */
487 qq10 = _mm_mul_ps(iq1,jq0);
488
489 /* EWALD ELECTROSTATICS */
490
491 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
492 ewrt = _mm_mul_ps(r10,ewtabscale);
493 ewitab = _mm_cvttps_epi32(ewrt);
494 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
495 ewitab = _mm_slli_epi32(ewitab,2);
496 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
497 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
498 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
499 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
500 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
501 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
502 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
503 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
504 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
505
506 /* Update potential sum for this i atom from the interaction with this j atom. */
507 velec = _mm_andnot_ps(dummy_mask,velec);
508 velecsum = _mm_add_ps(velecsum,velec);
509
510 fscal = felec;
511
512 fscal = _mm_andnot_ps(dummy_mask,fscal);
513
514 /* Calculate temporary vectorial force */
515 tx = _mm_mul_ps(fscal,dx10);
516 ty = _mm_mul_ps(fscal,dy10);
517 tz = _mm_mul_ps(fscal,dz10);
518
519 /* Update vectorial force */
520 fix1 = _mm_add_ps(fix1,tx);
521 fiy1 = _mm_add_ps(fiy1,ty);
522 fiz1 = _mm_add_ps(fiz1,tz);
523
524 fjx0 = _mm_add_ps(fjx0,tx);
525 fjy0 = _mm_add_ps(fjy0,ty);
526 fjz0 = _mm_add_ps(fjz0,tz);
527
528 /**************************
529 * CALCULATE INTERACTIONS *
530 **************************/
531
532 r20 = _mm_mul_ps(rsq20,rinv20);
533 r20 = _mm_andnot_ps(dummy_mask,r20);
534
535 /* Compute parameters for interactions between i and j atoms */
536 qq20 = _mm_mul_ps(iq2,jq0);
537
538 /* EWALD ELECTROSTATICS */
539
540 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
541 ewrt = _mm_mul_ps(r20,ewtabscale);
542 ewitab = _mm_cvttps_epi32(ewrt);
543 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
544 ewitab = _mm_slli_epi32(ewitab,2);
545 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
546 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
547 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
548 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
549 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
550 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
551 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
552 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
553 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
554
555 /* Update potential sum for this i atom from the interaction with this j atom. */
556 velec = _mm_andnot_ps(dummy_mask,velec);
557 velecsum = _mm_add_ps(velecsum,velec);
558
559 fscal = felec;
560
561 fscal = _mm_andnot_ps(dummy_mask,fscal);
562
563 /* Calculate temporary vectorial force */
564 tx = _mm_mul_ps(fscal,dx20);
565 ty = _mm_mul_ps(fscal,dy20);
566 tz = _mm_mul_ps(fscal,dz20);
567
568 /* Update vectorial force */
569 fix2 = _mm_add_ps(fix2,tx);
570 fiy2 = _mm_add_ps(fiy2,ty);
571 fiz2 = _mm_add_ps(fiz2,tz);
572
573 fjx0 = _mm_add_ps(fjx0,tx);
574 fjy0 = _mm_add_ps(fjy0,ty);
575 fjz0 = _mm_add_ps(fjz0,tz);
576
577 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
578 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
579 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
580 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
581
582 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
583
584 /* Inner loop uses 126 flops */
585 }
586
587 /* End of innermost loop */
588
589 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
590 f+i_coord_offset,fshift+i_shift_offset);
591
592 ggid = gid[iidx];
593 /* Update potential energies */
594 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
595
596 /* Increment number of inner iterations */
597 inneriter += j_index_end - j_index_start;
598
599 /* Outer loop uses 19 flops */
600 }
601
602 /* Increment number of outer iterations */
603 outeriter += nri;
604
605 /* Update outer/inner flops */
606
607 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*126);
608 }
609 /*
610 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse2_single
611 * Electrostatics interaction: Ewald
612 * VdW interaction: None
613 * Geometry: Water3-Particle
614 * Calculate force/pot: Force
615 */
616 void
617 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse2_single
618 (t_nblist * gmx_restrict nlist,
619 rvec * gmx_restrict xx,
620 rvec * gmx_restrict ff,
621 t_forcerec * gmx_restrict fr,
622 t_mdatoms * gmx_restrict mdatoms,
623 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
624 t_nrnb * gmx_restrict nrnb)
625 {
626 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
627 * just 0 for non-waters.
628 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
629 * jnr indices corresponding to data put in the four positions in the SIMD register.
630 */
631 int i_shift_offset,i_coord_offset,outeriter,inneriter;
632 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
633 int jnrA,jnrB,jnrC,jnrD;
634 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
635 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
636 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
637 real rcutoff_scalar;
638 real *shiftvec,*fshift,*x,*f;
639 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
640 real scratch[4*DIM];
641 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
642 int vdwioffset0;
643 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
644 int vdwioffset1;
645 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
646 int vdwioffset2;
647 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
648 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
649 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
650 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
651 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
652 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
653 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
654 real *charge;
655 __m128i ewitab;
656 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
657 real *ewtab;
658 __m128 dummy_mask,cutoff_mask;
659 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
660 __m128 one = _mm_set1_ps(1.0);
661 __m128 two = _mm_set1_ps(2.0);
662 x = xx[0];
663 f = ff[0];
664
665 nri = nlist->nri;
666 iinr = nlist->iinr;
667 jindex = nlist->jindex;
668 jjnr = nlist->jjnr;
669 shiftidx = nlist->shift;
670 gid = nlist->gid;
671 shiftvec = fr->shift_vec[0];
672 fshift = fr->fshift[0];
673 facel = _mm_set1_ps(fr->epsfac);
674 charge = mdatoms->chargeA;
675
676 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
677 ewtab = fr->ic->tabq_coul_F;
678 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
679 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
680
681 /* Setup water-specific parameters */
682 inr = nlist->iinr[0];
683 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
684 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
685 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
686
687 /* Avoid stupid compiler warnings */
688 jnrA = jnrB = jnrC = jnrD = 0;
689 j_coord_offsetA = 0;
690 j_coord_offsetB = 0;
691 j_coord_offsetC = 0;
692 j_coord_offsetD = 0;
693
694 outeriter = 0;
695 inneriter = 0;
696
697 for(iidx=0;iidx<4*DIM;iidx++)
698 {
699 scratch[iidx] = 0.0;
700 }
701
702 /* Start outer loop over neighborlists */
703 for(iidx=0; iidx<nri; iidx++)
704 {
705 /* Load shift vector for this list */
706 i_shift_offset = DIM*shiftidx[iidx];
707
708 /* Load limits for loop over neighbors */
709 j_index_start = jindex[iidx];
710 j_index_end = jindex[iidx+1];
711
712 /* Get outer coordinate index */
713 inr = iinr[iidx];
714 i_coord_offset = DIM*inr;
715
716 /* Load i particle coords and add shift vector */
717 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
718 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
719
720 fix0 = _mm_setzero_ps();
721 fiy0 = _mm_setzero_ps();
722 fiz0 = _mm_setzero_ps();
723 fix1 = _mm_setzero_ps();
724 fiy1 = _mm_setzero_ps();
725 fiz1 = _mm_setzero_ps();
726 fix2 = _mm_setzero_ps();
727 fiy2 = _mm_setzero_ps();
728 fiz2 = _mm_setzero_ps();
729
730 /* Start inner kernel loop */
731 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
732 {
733
734 /* Get j neighbor index, and coordinate index */
735 jnrA = jjnr[jidx];
736 jnrB = jjnr[jidx+1];
737 jnrC = jjnr[jidx+2];
738 jnrD = jjnr[jidx+3];
739 j_coord_offsetA = DIM*jnrA;
740 j_coord_offsetB = DIM*jnrB;
741 j_coord_offsetC = DIM*jnrC;
742 j_coord_offsetD = DIM*jnrD;
743
744 /* load j atom coordinates */
745 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
746 x+j_coord_offsetC,x+j_coord_offsetD,
747 &jx0,&jy0,&jz0);
748
749 /* Calculate displacement vector */
750 dx00 = _mm_sub_ps(ix0,jx0);
751 dy00 = _mm_sub_ps(iy0,jy0);
752 dz00 = _mm_sub_ps(iz0,jz0);
753 dx10 = _mm_sub_ps(ix1,jx0);
754 dy10 = _mm_sub_ps(iy1,jy0);
755 dz10 = _mm_sub_ps(iz1,jz0);
756 dx20 = _mm_sub_ps(ix2,jx0);
757 dy20 = _mm_sub_ps(iy2,jy0);
758 dz20 = _mm_sub_ps(iz2,jz0);
759
760 /* Calculate squared distance and things based on it */
761 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
762 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
763 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
764
765 rinv00 = gmx_mm_invsqrt_ps(rsq00);
766 rinv10 = gmx_mm_invsqrt_ps(rsq10);
767 rinv20 = gmx_mm_invsqrt_ps(rsq20);
768
769 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
770 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
771 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
772
773 /* Load parameters for j particles */
774 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
775 charge+jnrC+0,charge+jnrD+0);
776
777 fjx0 = _mm_setzero_ps();
778 fjy0 = _mm_setzero_ps();
779 fjz0 = _mm_setzero_ps();
780
781 /**************************
782 * CALCULATE INTERACTIONS *
783 **************************/
784
785 r00 = _mm_mul_ps(rsq00,rinv00);
786
787 /* Compute parameters for interactions between i and j atoms */
788 qq00 = _mm_mul_ps(iq0,jq0);
789
790 /* EWALD ELECTROSTATICS */
791
792 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
793 ewrt = _mm_mul_ps(r00,ewtabscale);
794 ewitab = _mm_cvttps_epi32(ewrt);
795 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
796 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
797 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
798 &ewtabF,&ewtabFn);
799 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
800 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
801
802 fscal = felec;
803
804 /* Calculate temporary vectorial force */
805 tx = _mm_mul_ps(fscal,dx00);
806 ty = _mm_mul_ps(fscal,dy00);
807 tz = _mm_mul_ps(fscal,dz00);
808
809 /* Update vectorial force */
810 fix0 = _mm_add_ps(fix0,tx);
811 fiy0 = _mm_add_ps(fiy0,ty);
812 fiz0 = _mm_add_ps(fiz0,tz);
813
814 fjx0 = _mm_add_ps(fjx0,tx);
815 fjy0 = _mm_add_ps(fjy0,ty);
816 fjz0 = _mm_add_ps(fjz0,tz);
817
818 /**************************
819 * CALCULATE INTERACTIONS *
820 **************************/
821
822 r10 = _mm_mul_ps(rsq10,rinv10);
823
824 /* Compute parameters for interactions between i and j atoms */
825 qq10 = _mm_mul_ps(iq1,jq0);
826
827 /* EWALD ELECTROSTATICS */
828
829 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
830 ewrt = _mm_mul_ps(r10,ewtabscale);
831 ewitab = _mm_cvttps_epi32(ewrt);
832 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
833 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
834 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
835 &ewtabF,&ewtabFn);
836 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
837 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
838
839 fscal = felec;
840
841 /* Calculate temporary vectorial force */
842 tx = _mm_mul_ps(fscal,dx10);
843 ty = _mm_mul_ps(fscal,dy10);
844 tz = _mm_mul_ps(fscal,dz10);
845
846 /* Update vectorial force */
847 fix1 = _mm_add_ps(fix1,tx);
848 fiy1 = _mm_add_ps(fiy1,ty);
849 fiz1 = _mm_add_ps(fiz1,tz);
850
851 fjx0 = _mm_add_ps(fjx0,tx);
852 fjy0 = _mm_add_ps(fjy0,ty);
853 fjz0 = _mm_add_ps(fjz0,tz);
854
855 /**************************
856 * CALCULATE INTERACTIONS *
857 **************************/
858
859 r20 = _mm_mul_ps(rsq20,rinv20);
860
861 /* Compute parameters for interactions between i and j atoms */
862 qq20 = _mm_mul_ps(iq2,jq0);
863
864 /* EWALD ELECTROSTATICS */
865
866 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
867 ewrt = _mm_mul_ps(r20,ewtabscale);
868 ewitab = _mm_cvttps_epi32(ewrt);
869 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
870 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
871 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
872 &ewtabF,&ewtabFn);
873 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
874 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
875
876 fscal = felec;
877
878 /* Calculate temporary vectorial force */
879 tx = _mm_mul_ps(fscal,dx20);
880 ty = _mm_mul_ps(fscal,dy20);
881 tz = _mm_mul_ps(fscal,dz20);
882
883 /* Update vectorial force */
884 fix2 = _mm_add_ps(fix2,tx);
885 fiy2 = _mm_add_ps(fiy2,ty);
886 fiz2 = _mm_add_ps(fiz2,tz);
887
888 fjx0 = _mm_add_ps(fjx0,tx);
889 fjy0 = _mm_add_ps(fjy0,ty);
890 fjz0 = _mm_add_ps(fjz0,tz);
891
892 fjptrA = f+j_coord_offsetA;
893 fjptrB = f+j_coord_offsetB;
894 fjptrC = f+j_coord_offsetC;
895 fjptrD = f+j_coord_offsetD;
896
897 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
898
899 /* Inner loop uses 108 flops */
900 }
901
902 if(jidx<j_index_end)
903 {
904
905 /* Get j neighbor index, and coordinate index */
906 jnrlistA = jjnr[jidx];
907 jnrlistB = jjnr[jidx+1];
908 jnrlistC = jjnr[jidx+2];
909 jnrlistD = jjnr[jidx+3];
910 /* Sign of each element will be negative for non-real atoms.
911 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
912 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
913 */
914 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
915 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
916 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
917 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
918 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
919 j_coord_offsetA = DIM*jnrA;
920 j_coord_offsetB = DIM*jnrB;
921 j_coord_offsetC = DIM*jnrC;
922 j_coord_offsetD = DIM*jnrD;
923
924 /* load j atom coordinates */
925 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
926 x+j_coord_offsetC,x+j_coord_offsetD,
927 &jx0,&jy0,&jz0);
928
929 /* Calculate displacement vector */
930 dx00 = _mm_sub_ps(ix0,jx0);
931 dy00 = _mm_sub_ps(iy0,jy0);
932 dz00 = _mm_sub_ps(iz0,jz0);
933 dx10 = _mm_sub_ps(ix1,jx0);
934 dy10 = _mm_sub_ps(iy1,jy0);
935 dz10 = _mm_sub_ps(iz1,jz0);
936 dx20 = _mm_sub_ps(ix2,jx0);
937 dy20 = _mm_sub_ps(iy2,jy0);
938 dz20 = _mm_sub_ps(iz2,jz0);
939
940 /* Calculate squared distance and things based on it */
941 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
942 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
943 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
944
945 rinv00 = gmx_mm_invsqrt_ps(rsq00);
946 rinv10 = gmx_mm_invsqrt_ps(rsq10);
947 rinv20 = gmx_mm_invsqrt_ps(rsq20);
948
949 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
950 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
951 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
952
953 /* Load parameters for j particles */
954 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
955 charge+jnrC+0,charge+jnrD+0);
956
957 fjx0 = _mm_setzero_ps();
958 fjy0 = _mm_setzero_ps();
959 fjz0 = _mm_setzero_ps();
960
961 /**************************
962 * CALCULATE INTERACTIONS *
963 **************************/
964
965 r00 = _mm_mul_ps(rsq00,rinv00);
966 r00 = _mm_andnot_ps(dummy_mask,r00);
967
968 /* Compute parameters for interactions between i and j atoms */
969 qq00 = _mm_mul_ps(iq0,jq0);
970
971 /* EWALD ELECTROSTATICS */
972
973 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
974 ewrt = _mm_mul_ps(r00,ewtabscale);
975 ewitab = _mm_cvttps_epi32(ewrt);
976 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
977 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
978 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
979 &ewtabF,&ewtabFn);
980 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
981 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
982
983 fscal = felec;
984
985 fscal = _mm_andnot_ps(dummy_mask,fscal);
986
987 /* Calculate temporary vectorial force */
988 tx = _mm_mul_ps(fscal,dx00);
989 ty = _mm_mul_ps(fscal,dy00);
990 tz = _mm_mul_ps(fscal,dz00);
991
992 /* Update vectorial force */
993 fix0 = _mm_add_ps(fix0,tx);
994 fiy0 = _mm_add_ps(fiy0,ty);
995 fiz0 = _mm_add_ps(fiz0,tz);
996
997 fjx0 = _mm_add_ps(fjx0,tx);
998 fjy0 = _mm_add_ps(fjy0,ty);
999 fjz0 = _mm_add_ps(fjz0,tz);
1000
1001 /**************************
1002 * CALCULATE INTERACTIONS *
1003 **************************/
1004
1005 r10 = _mm_mul_ps(rsq10,rinv10);
1006 r10 = _mm_andnot_ps(dummy_mask,r10);
1007
1008 /* Compute parameters for interactions between i and j atoms */
1009 qq10 = _mm_mul_ps(iq1,jq0);
1010
1011 /* EWALD ELECTROSTATICS */
1012
1013 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1014 ewrt = _mm_mul_ps(r10,ewtabscale);
1015 ewitab = _mm_cvttps_epi32(ewrt);
1016 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1017 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1018 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1019 &ewtabF,&ewtabFn);
1020 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1021 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1022
1023 fscal = felec;
1024
1025 fscal = _mm_andnot_ps(dummy_mask,fscal);
1026
1027 /* Calculate temporary vectorial force */
1028 tx = _mm_mul_ps(fscal,dx10);
1029 ty = _mm_mul_ps(fscal,dy10);
1030 tz = _mm_mul_ps(fscal,dz10);
1031
1032 /* Update vectorial force */
1033 fix1 = _mm_add_ps(fix1,tx);
1034 fiy1 = _mm_add_ps(fiy1,ty);
1035 fiz1 = _mm_add_ps(fiz1,tz);
1036
1037 fjx0 = _mm_add_ps(fjx0,tx);
1038 fjy0 = _mm_add_ps(fjy0,ty);
1039 fjz0 = _mm_add_ps(fjz0,tz);
1040
1041 /**************************
1042 * CALCULATE INTERACTIONS *
1043 **************************/
1044
1045 r20 = _mm_mul_ps(rsq20,rinv20);
1046 r20 = _mm_andnot_ps(dummy_mask,r20);
1047
1048 /* Compute parameters for interactions between i and j atoms */
1049 qq20 = _mm_mul_ps(iq2,jq0);
1050
1051 /* EWALD ELECTROSTATICS */
1052
1053 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1054 ewrt = _mm_mul_ps(r20,ewtabscale);
1055 ewitab = _mm_cvttps_epi32(ewrt);
1056 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1057 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1058 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1059 &ewtabF,&ewtabFn);
1060 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1061 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1062
1063 fscal = felec;
1064
1065 fscal = _mm_andnot_ps(dummy_mask,fscal);
1066
1067 /* Calculate temporary vectorial force */
1068 tx = _mm_mul_ps(fscal,dx20);
1069 ty = _mm_mul_ps(fscal,dy20);
1070 tz = _mm_mul_ps(fscal,dz20);
1071
1072 /* Update vectorial force */
1073 fix2 = _mm_add_ps(fix2,tx);
1074 fiy2 = _mm_add_ps(fiy2,ty);
1075 fiz2 = _mm_add_ps(fiz2,tz);
1076
1077 fjx0 = _mm_add_ps(fjx0,tx);
1078 fjy0 = _mm_add_ps(fjy0,ty);
1079 fjz0 = _mm_add_ps(fjz0,tz);
1080
1081 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1082 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1083 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1084 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1085
1086 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1087
1088 /* Inner loop uses 111 flops */
1089 }
1090
1091 /* End of innermost loop */
1092
1093 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1094 f+i_coord_offset,fshift+i_shift_offset);
1095
1096 /* Increment number of inner iterations */
1097 inneriter += j_index_end - j_index_start;
1098
1099 /* Outer loop uses 18 flops */
1100 }
1101
1102 /* Increment number of outer iterations */
1103 outeriter += nri;
1104
1105 /* Update outer/inner flops */
1106
1107 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*111);
1108 }
The ultimate molecular dynamics simulation package