?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Fix segmentation fault in minimize
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse2_single / nb_kernel_ElecEwSh_VdwNone_GeomW4W4_sse2_single.cpp
blobe01a0420813a015596faf9083e41c23df6132103
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015,2017,2018, 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/gmxlib/nrnb.h"
46
47 #include "kernelutil_x86_sse2_single.h"
48
49 /*
50 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW4W4_VF_sse2_single
51 * Electrostatics interaction: Ewald
52 * VdW interaction: None
53 * Geometry: Water4-Water4
54 * Calculate force/pot: PotentialAndForce
55 */
56 void
57 nb_kernel_ElecEwSh_VdwNone_GeomW4W4_VF_sse2_single
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
65 {
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
70 */
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
73 int jnrA,jnrB,jnrC,jnrD;
74 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
75 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
77 real rcutoff_scalar;
78 real *shiftvec,*fshift,*x,*f;
79 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
80 real scratch[4*DIM];
81 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
82 int vdwioffset1;
83 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
84 int vdwioffset2;
85 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
86 int vdwioffset3;
87 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
88 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
89 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
90 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
91 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
92 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
93 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
94 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
95 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
96 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
97 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
98 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
99 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
100 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
101 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
102 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
103 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
104 real *charge;
105 __m128i ewitab;
106 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
107 real *ewtab;
108 __m128 dummy_mask,cutoff_mask;
109 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
110 __m128 one = _mm_set1_ps(1.0);
111 __m128 two = _mm_set1_ps(2.0);
112 x = xx[0];
113 f = ff[0];
114
115 nri = nlist->nri;
116 iinr = nlist->iinr;
117 jindex = nlist->jindex;
118 jjnr = nlist->jjnr;
119 shiftidx = nlist->shift;
120 gid = nlist->gid;
121 shiftvec = fr->shift_vec[0];
122 fshift = fr->fshift[0];
123 facel = _mm_set1_ps(fr->ic->epsfac);
124 charge = mdatoms->chargeA;
125
126 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
127 ewtab = fr->ic->tabq_coul_FDV0;
128 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
129 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
130
131 /* Setup water-specific parameters */
132 inr = nlist->iinr[0];
133 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
134 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
135 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
136
137 jq1 = _mm_set1_ps(charge[inr+1]);
138 jq2 = _mm_set1_ps(charge[inr+2]);
139 jq3 = _mm_set1_ps(charge[inr+3]);
140 qq11 = _mm_mul_ps(iq1,jq1);
141 qq12 = _mm_mul_ps(iq1,jq2);
142 qq13 = _mm_mul_ps(iq1,jq3);
143 qq21 = _mm_mul_ps(iq2,jq1);
144 qq22 = _mm_mul_ps(iq2,jq2);
145 qq23 = _mm_mul_ps(iq2,jq3);
146 qq31 = _mm_mul_ps(iq3,jq1);
147 qq32 = _mm_mul_ps(iq3,jq2);
148 qq33 = _mm_mul_ps(iq3,jq3);
149
150 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
151 rcutoff_scalar = fr->ic->rcoulomb;
152 rcutoff = _mm_set1_ps(rcutoff_scalar);
153 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
154
155 /* Avoid stupid compiler warnings */
156 jnrA = jnrB = jnrC = jnrD = 0;
157 j_coord_offsetA = 0;
158 j_coord_offsetB = 0;
159 j_coord_offsetC = 0;
160 j_coord_offsetD = 0;
161
162 outeriter = 0;
163 inneriter = 0;
164
165 for(iidx=0;iidx<4*DIM;iidx++)
166 {
167 scratch[iidx] = 0.0;
168 }
169
170 /* Start outer loop over neighborlists */
171 for(iidx=0; iidx<nri; iidx++)
172 {
173 /* Load shift vector for this list */
174 i_shift_offset = DIM*shiftidx[iidx];
175
176 /* Load limits for loop over neighbors */
177 j_index_start = jindex[iidx];
178 j_index_end = jindex[iidx+1];
179
180 /* Get outer coordinate index */
181 inr = iinr[iidx];
182 i_coord_offset = DIM*inr;
183
184 /* Load i particle coords and add shift vector */
185 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
186 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
187
188 fix1 = _mm_setzero_ps();
189 fiy1 = _mm_setzero_ps();
190 fiz1 = _mm_setzero_ps();
191 fix2 = _mm_setzero_ps();
192 fiy2 = _mm_setzero_ps();
193 fiz2 = _mm_setzero_ps();
194 fix3 = _mm_setzero_ps();
195 fiy3 = _mm_setzero_ps();
196 fiz3 = _mm_setzero_ps();
197
198 /* Reset potential sums */
199 velecsum = _mm_setzero_ps();
200
201 /* Start inner kernel loop */
202 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
203 {
204
205 /* Get j neighbor index, and coordinate index */
206 jnrA = jjnr[jidx];
207 jnrB = jjnr[jidx+1];
208 jnrC = jjnr[jidx+2];
209 jnrD = jjnr[jidx+3];
210 j_coord_offsetA = DIM*jnrA;
211 j_coord_offsetB = DIM*jnrB;
212 j_coord_offsetC = DIM*jnrC;
213 j_coord_offsetD = DIM*jnrD;
214
215 /* load j atom coordinates */
216 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
217 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
218 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
219
220 /* Calculate displacement vector */
221 dx11 = _mm_sub_ps(ix1,jx1);
222 dy11 = _mm_sub_ps(iy1,jy1);
223 dz11 = _mm_sub_ps(iz1,jz1);
224 dx12 = _mm_sub_ps(ix1,jx2);
225 dy12 = _mm_sub_ps(iy1,jy2);
226 dz12 = _mm_sub_ps(iz1,jz2);
227 dx13 = _mm_sub_ps(ix1,jx3);
228 dy13 = _mm_sub_ps(iy1,jy3);
229 dz13 = _mm_sub_ps(iz1,jz3);
230 dx21 = _mm_sub_ps(ix2,jx1);
231 dy21 = _mm_sub_ps(iy2,jy1);
232 dz21 = _mm_sub_ps(iz2,jz1);
233 dx22 = _mm_sub_ps(ix2,jx2);
234 dy22 = _mm_sub_ps(iy2,jy2);
235 dz22 = _mm_sub_ps(iz2,jz2);
236 dx23 = _mm_sub_ps(ix2,jx3);
237 dy23 = _mm_sub_ps(iy2,jy3);
238 dz23 = _mm_sub_ps(iz2,jz3);
239 dx31 = _mm_sub_ps(ix3,jx1);
240 dy31 = _mm_sub_ps(iy3,jy1);
241 dz31 = _mm_sub_ps(iz3,jz1);
242 dx32 = _mm_sub_ps(ix3,jx2);
243 dy32 = _mm_sub_ps(iy3,jy2);
244 dz32 = _mm_sub_ps(iz3,jz2);
245 dx33 = _mm_sub_ps(ix3,jx3);
246 dy33 = _mm_sub_ps(iy3,jy3);
247 dz33 = _mm_sub_ps(iz3,jz3);
248
249 /* Calculate squared distance and things based on it */
250 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
251 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
252 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
253 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
254 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
255 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
256 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
257 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
258 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
259
260 rinv11 = sse2_invsqrt_f(rsq11);
261 rinv12 = sse2_invsqrt_f(rsq12);
262 rinv13 = sse2_invsqrt_f(rsq13);
263 rinv21 = sse2_invsqrt_f(rsq21);
264 rinv22 = sse2_invsqrt_f(rsq22);
265 rinv23 = sse2_invsqrt_f(rsq23);
266 rinv31 = sse2_invsqrt_f(rsq31);
267 rinv32 = sse2_invsqrt_f(rsq32);
268 rinv33 = sse2_invsqrt_f(rsq33);
269
270 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
271 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
272 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
273 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
274 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
275 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
276 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
277 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
278 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
279
280 fjx1 = _mm_setzero_ps();
281 fjy1 = _mm_setzero_ps();
282 fjz1 = _mm_setzero_ps();
283 fjx2 = _mm_setzero_ps();
284 fjy2 = _mm_setzero_ps();
285 fjz2 = _mm_setzero_ps();
286 fjx3 = _mm_setzero_ps();
287 fjy3 = _mm_setzero_ps();
288 fjz3 = _mm_setzero_ps();
289
290 /**************************
291 * CALCULATE INTERACTIONS *
292 **************************/
293
294 if (gmx_mm_any_lt(rsq11,rcutoff2))
295 {
296
297 r11 = _mm_mul_ps(rsq11,rinv11);
298
299 /* EWALD ELECTROSTATICS */
300
301 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
302 ewrt = _mm_mul_ps(r11,ewtabscale);
303 ewitab = _mm_cvttps_epi32(ewrt);
304 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
305 ewitab = _mm_slli_epi32(ewitab,2);
306 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
307 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
308 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
309 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
310 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
311 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
312 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
313 velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_sub_ps(rinv11,sh_ewald),velec));
314 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
315
316 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
317
318 /* Update potential sum for this i atom from the interaction with this j atom. */
319 velec = _mm_and_ps(velec,cutoff_mask);
320 velecsum = _mm_add_ps(velecsum,velec);
321
322 fscal = felec;
323
324 fscal = _mm_and_ps(fscal,cutoff_mask);
325
326 /* Calculate temporary vectorial force */
327 tx = _mm_mul_ps(fscal,dx11);
328 ty = _mm_mul_ps(fscal,dy11);
329 tz = _mm_mul_ps(fscal,dz11);
330
331 /* Update vectorial force */
332 fix1 = _mm_add_ps(fix1,tx);
333 fiy1 = _mm_add_ps(fiy1,ty);
334 fiz1 = _mm_add_ps(fiz1,tz);
335
336 fjx1 = _mm_add_ps(fjx1,tx);
337 fjy1 = _mm_add_ps(fjy1,ty);
338 fjz1 = _mm_add_ps(fjz1,tz);
339
340 }
341
342 /**************************
343 * CALCULATE INTERACTIONS *
344 **************************/
345
346 if (gmx_mm_any_lt(rsq12,rcutoff2))
347 {
348
349 r12 = _mm_mul_ps(rsq12,rinv12);
350
351 /* EWALD ELECTROSTATICS */
352
353 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
354 ewrt = _mm_mul_ps(r12,ewtabscale);
355 ewitab = _mm_cvttps_epi32(ewrt);
356 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
357 ewitab = _mm_slli_epi32(ewitab,2);
358 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
359 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
360 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
361 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
362 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
363 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
364 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
365 velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_sub_ps(rinv12,sh_ewald),velec));
366 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
367
368 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
369
370 /* Update potential sum for this i atom from the interaction with this j atom. */
371 velec = _mm_and_ps(velec,cutoff_mask);
372 velecsum = _mm_add_ps(velecsum,velec);
373
374 fscal = felec;
375
376 fscal = _mm_and_ps(fscal,cutoff_mask);
377
378 /* Calculate temporary vectorial force */
379 tx = _mm_mul_ps(fscal,dx12);
380 ty = _mm_mul_ps(fscal,dy12);
381 tz = _mm_mul_ps(fscal,dz12);
382
383 /* Update vectorial force */
384 fix1 = _mm_add_ps(fix1,tx);
385 fiy1 = _mm_add_ps(fiy1,ty);
386 fiz1 = _mm_add_ps(fiz1,tz);
387
388 fjx2 = _mm_add_ps(fjx2,tx);
389 fjy2 = _mm_add_ps(fjy2,ty);
390 fjz2 = _mm_add_ps(fjz2,tz);
391
392 }
393
394 /**************************
395 * CALCULATE INTERACTIONS *
396 **************************/
397
398 if (gmx_mm_any_lt(rsq13,rcutoff2))
399 {
400
401 r13 = _mm_mul_ps(rsq13,rinv13);
402
403 /* EWALD ELECTROSTATICS */
404
405 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
406 ewrt = _mm_mul_ps(r13,ewtabscale);
407 ewitab = _mm_cvttps_epi32(ewrt);
408 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
409 ewitab = _mm_slli_epi32(ewitab,2);
410 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
411 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
412 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
413 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
414 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
415 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
416 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
417 velec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_sub_ps(rinv13,sh_ewald),velec));
418 felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
419
420 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
421
422 /* Update potential sum for this i atom from the interaction with this j atom. */
423 velec = _mm_and_ps(velec,cutoff_mask);
424 velecsum = _mm_add_ps(velecsum,velec);
425
426 fscal = felec;
427
428 fscal = _mm_and_ps(fscal,cutoff_mask);
429
430 /* Calculate temporary vectorial force */
431 tx = _mm_mul_ps(fscal,dx13);
432 ty = _mm_mul_ps(fscal,dy13);
433 tz = _mm_mul_ps(fscal,dz13);
434
435 /* Update vectorial force */
436 fix1 = _mm_add_ps(fix1,tx);
437 fiy1 = _mm_add_ps(fiy1,ty);
438 fiz1 = _mm_add_ps(fiz1,tz);
439
440 fjx3 = _mm_add_ps(fjx3,tx);
441 fjy3 = _mm_add_ps(fjy3,ty);
442 fjz3 = _mm_add_ps(fjz3,tz);
443
444 }
445
446 /**************************
447 * CALCULATE INTERACTIONS *
448 **************************/
449
450 if (gmx_mm_any_lt(rsq21,rcutoff2))
451 {
452
453 r21 = _mm_mul_ps(rsq21,rinv21);
454
455 /* EWALD ELECTROSTATICS */
456
457 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
458 ewrt = _mm_mul_ps(r21,ewtabscale);
459 ewitab = _mm_cvttps_epi32(ewrt);
460 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
461 ewitab = _mm_slli_epi32(ewitab,2);
462 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
463 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
464 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
465 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
466 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
467 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
468 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
469 velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_sub_ps(rinv21,sh_ewald),velec));
470 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
471
472 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
473
474 /* Update potential sum for this i atom from the interaction with this j atom. */
475 velec = _mm_and_ps(velec,cutoff_mask);
476 velecsum = _mm_add_ps(velecsum,velec);
477
478 fscal = felec;
479
480 fscal = _mm_and_ps(fscal,cutoff_mask);
481
482 /* Calculate temporary vectorial force */
483 tx = _mm_mul_ps(fscal,dx21);
484 ty = _mm_mul_ps(fscal,dy21);
485 tz = _mm_mul_ps(fscal,dz21);
486
487 /* Update vectorial force */
488 fix2 = _mm_add_ps(fix2,tx);
489 fiy2 = _mm_add_ps(fiy2,ty);
490 fiz2 = _mm_add_ps(fiz2,tz);
491
492 fjx1 = _mm_add_ps(fjx1,tx);
493 fjy1 = _mm_add_ps(fjy1,ty);
494 fjz1 = _mm_add_ps(fjz1,tz);
495
496 }
497
498 /**************************
499 * CALCULATE INTERACTIONS *
500 **************************/
501
502 if (gmx_mm_any_lt(rsq22,rcutoff2))
503 {
504
505 r22 = _mm_mul_ps(rsq22,rinv22);
506
507 /* EWALD ELECTROSTATICS */
508
509 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
510 ewrt = _mm_mul_ps(r22,ewtabscale);
511 ewitab = _mm_cvttps_epi32(ewrt);
512 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
513 ewitab = _mm_slli_epi32(ewitab,2);
514 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
515 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
516 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
517 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
518 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
519 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
520 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
521 velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_sub_ps(rinv22,sh_ewald),velec));
522 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
523
524 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
525
526 /* Update potential sum for this i atom from the interaction with this j atom. */
527 velec = _mm_and_ps(velec,cutoff_mask);
528 velecsum = _mm_add_ps(velecsum,velec);
529
530 fscal = felec;
531
532 fscal = _mm_and_ps(fscal,cutoff_mask);
533
534 /* Calculate temporary vectorial force */
535 tx = _mm_mul_ps(fscal,dx22);
536 ty = _mm_mul_ps(fscal,dy22);
537 tz = _mm_mul_ps(fscal,dz22);
538
539 /* Update vectorial force */
540 fix2 = _mm_add_ps(fix2,tx);
541 fiy2 = _mm_add_ps(fiy2,ty);
542 fiz2 = _mm_add_ps(fiz2,tz);
543
544 fjx2 = _mm_add_ps(fjx2,tx);
545 fjy2 = _mm_add_ps(fjy2,ty);
546 fjz2 = _mm_add_ps(fjz2,tz);
547
548 }
549
550 /**************************
551 * CALCULATE INTERACTIONS *
552 **************************/
553
554 if (gmx_mm_any_lt(rsq23,rcutoff2))
555 {
556
557 r23 = _mm_mul_ps(rsq23,rinv23);
558
559 /* EWALD ELECTROSTATICS */
560
561 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
562 ewrt = _mm_mul_ps(r23,ewtabscale);
563 ewitab = _mm_cvttps_epi32(ewrt);
564 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
565 ewitab = _mm_slli_epi32(ewitab,2);
566 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
567 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
568 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
569 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
570 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
571 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
572 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
573 velec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_sub_ps(rinv23,sh_ewald),velec));
574 felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
575
576 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
577
578 /* Update potential sum for this i atom from the interaction with this j atom. */
579 velec = _mm_and_ps(velec,cutoff_mask);
580 velecsum = _mm_add_ps(velecsum,velec);
581
582 fscal = felec;
583
584 fscal = _mm_and_ps(fscal,cutoff_mask);
585
586 /* Calculate temporary vectorial force */
587 tx = _mm_mul_ps(fscal,dx23);
588 ty = _mm_mul_ps(fscal,dy23);
589 tz = _mm_mul_ps(fscal,dz23);
590
591 /* Update vectorial force */
592 fix2 = _mm_add_ps(fix2,tx);
593 fiy2 = _mm_add_ps(fiy2,ty);
594 fiz2 = _mm_add_ps(fiz2,tz);
595
596 fjx3 = _mm_add_ps(fjx3,tx);
597 fjy3 = _mm_add_ps(fjy3,ty);
598 fjz3 = _mm_add_ps(fjz3,tz);
599
600 }
601
602 /**************************
603 * CALCULATE INTERACTIONS *
604 **************************/
605
606 if (gmx_mm_any_lt(rsq31,rcutoff2))
607 {
608
609 r31 = _mm_mul_ps(rsq31,rinv31);
610
611 /* EWALD ELECTROSTATICS */
612
613 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
614 ewrt = _mm_mul_ps(r31,ewtabscale);
615 ewitab = _mm_cvttps_epi32(ewrt);
616 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
617 ewitab = _mm_slli_epi32(ewitab,2);
618 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
619 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
620 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
621 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
622 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
623 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
624 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
625 velec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_sub_ps(rinv31,sh_ewald),velec));
626 felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
627
628 cutoff_mask = _mm_cmplt_ps(rsq31,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 velecsum = _mm_add_ps(velecsum,velec);
633
634 fscal = felec;
635
636 fscal = _mm_and_ps(fscal,cutoff_mask);
637
638 /* Calculate temporary vectorial force */
639 tx = _mm_mul_ps(fscal,dx31);
640 ty = _mm_mul_ps(fscal,dy31);
641 tz = _mm_mul_ps(fscal,dz31);
642
643 /* Update vectorial force */
644 fix3 = _mm_add_ps(fix3,tx);
645 fiy3 = _mm_add_ps(fiy3,ty);
646 fiz3 = _mm_add_ps(fiz3,tz);
647
648 fjx1 = _mm_add_ps(fjx1,tx);
649 fjy1 = _mm_add_ps(fjy1,ty);
650 fjz1 = _mm_add_ps(fjz1,tz);
651
652 }
653
654 /**************************
655 * CALCULATE INTERACTIONS *
656 **************************/
657
658 if (gmx_mm_any_lt(rsq32,rcutoff2))
659 {
660
661 r32 = _mm_mul_ps(rsq32,rinv32);
662
663 /* EWALD ELECTROSTATICS */
664
665 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
666 ewrt = _mm_mul_ps(r32,ewtabscale);
667 ewitab = _mm_cvttps_epi32(ewrt);
668 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
669 ewitab = _mm_slli_epi32(ewitab,2);
670 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
671 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
672 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
673 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
674 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
675 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
676 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
677 velec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_sub_ps(rinv32,sh_ewald),velec));
678 felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
679
680 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
681
682 /* Update potential sum for this i atom from the interaction with this j atom. */
683 velec = _mm_and_ps(velec,cutoff_mask);
684 velecsum = _mm_add_ps(velecsum,velec);
685
686 fscal = felec;
687
688 fscal = _mm_and_ps(fscal,cutoff_mask);
689
690 /* Calculate temporary vectorial force */
691 tx = _mm_mul_ps(fscal,dx32);
692 ty = _mm_mul_ps(fscal,dy32);
693 tz = _mm_mul_ps(fscal,dz32);
694
695 /* Update vectorial force */
696 fix3 = _mm_add_ps(fix3,tx);
697 fiy3 = _mm_add_ps(fiy3,ty);
698 fiz3 = _mm_add_ps(fiz3,tz);
699
700 fjx2 = _mm_add_ps(fjx2,tx);
701 fjy2 = _mm_add_ps(fjy2,ty);
702 fjz2 = _mm_add_ps(fjz2,tz);
703
704 }
705
706 /**************************
707 * CALCULATE INTERACTIONS *
708 **************************/
709
710 if (gmx_mm_any_lt(rsq33,rcutoff2))
711 {
712
713 r33 = _mm_mul_ps(rsq33,rinv33);
714
715 /* EWALD ELECTROSTATICS */
716
717 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
718 ewrt = _mm_mul_ps(r33,ewtabscale);
719 ewitab = _mm_cvttps_epi32(ewrt);
720 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
721 ewitab = _mm_slli_epi32(ewitab,2);
722 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
723 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
724 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
725 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
726 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
727 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
728 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
729 velec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_sub_ps(rinv33,sh_ewald),velec));
730 felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
731
732 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
733
734 /* Update potential sum for this i atom from the interaction with this j atom. */
735 velec = _mm_and_ps(velec,cutoff_mask);
736 velecsum = _mm_add_ps(velecsum,velec);
737
738 fscal = felec;
739
740 fscal = _mm_and_ps(fscal,cutoff_mask);
741
742 /* Calculate temporary vectorial force */
743 tx = _mm_mul_ps(fscal,dx33);
744 ty = _mm_mul_ps(fscal,dy33);
745 tz = _mm_mul_ps(fscal,dz33);
746
747 /* Update vectorial force */
748 fix3 = _mm_add_ps(fix3,tx);
749 fiy3 = _mm_add_ps(fiy3,ty);
750 fiz3 = _mm_add_ps(fiz3,tz);
751
752 fjx3 = _mm_add_ps(fjx3,tx);
753 fjy3 = _mm_add_ps(fjy3,ty);
754 fjz3 = _mm_add_ps(fjz3,tz);
755
756 }
757
758 fjptrA = f+j_coord_offsetA;
759 fjptrB = f+j_coord_offsetB;
760 fjptrC = f+j_coord_offsetC;
761 fjptrD = f+j_coord_offsetD;
762
763 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
764 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
765
766 /* Inner loop uses 414 flops */
767 }
768
769 if(jidx<j_index_end)
770 {
771
772 /* Get j neighbor index, and coordinate index */
773 jnrlistA = jjnr[jidx];
774 jnrlistB = jjnr[jidx+1];
775 jnrlistC = jjnr[jidx+2];
776 jnrlistD = jjnr[jidx+3];
777 /* Sign of each element will be negative for non-real atoms.
778 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
779 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
780 */
781 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
782 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
783 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
784 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
785 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
786 j_coord_offsetA = DIM*jnrA;
787 j_coord_offsetB = DIM*jnrB;
788 j_coord_offsetC = DIM*jnrC;
789 j_coord_offsetD = DIM*jnrD;
790
791 /* load j atom coordinates */
792 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
793 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
794 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
795
796 /* Calculate displacement vector */
797 dx11 = _mm_sub_ps(ix1,jx1);
798 dy11 = _mm_sub_ps(iy1,jy1);
799 dz11 = _mm_sub_ps(iz1,jz1);
800 dx12 = _mm_sub_ps(ix1,jx2);
801 dy12 = _mm_sub_ps(iy1,jy2);
802 dz12 = _mm_sub_ps(iz1,jz2);
803 dx13 = _mm_sub_ps(ix1,jx3);
804 dy13 = _mm_sub_ps(iy1,jy3);
805 dz13 = _mm_sub_ps(iz1,jz3);
806 dx21 = _mm_sub_ps(ix2,jx1);
807 dy21 = _mm_sub_ps(iy2,jy1);
808 dz21 = _mm_sub_ps(iz2,jz1);
809 dx22 = _mm_sub_ps(ix2,jx2);
810 dy22 = _mm_sub_ps(iy2,jy2);
811 dz22 = _mm_sub_ps(iz2,jz2);
812 dx23 = _mm_sub_ps(ix2,jx3);
813 dy23 = _mm_sub_ps(iy2,jy3);
814 dz23 = _mm_sub_ps(iz2,jz3);
815 dx31 = _mm_sub_ps(ix3,jx1);
816 dy31 = _mm_sub_ps(iy3,jy1);
817 dz31 = _mm_sub_ps(iz3,jz1);
818 dx32 = _mm_sub_ps(ix3,jx2);
819 dy32 = _mm_sub_ps(iy3,jy2);
820 dz32 = _mm_sub_ps(iz3,jz2);
821 dx33 = _mm_sub_ps(ix3,jx3);
822 dy33 = _mm_sub_ps(iy3,jy3);
823 dz33 = _mm_sub_ps(iz3,jz3);
824
825 /* Calculate squared distance and things based on it */
826 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
827 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
828 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
829 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
830 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
831 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
832 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
833 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
834 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
835
836 rinv11 = sse2_invsqrt_f(rsq11);
837 rinv12 = sse2_invsqrt_f(rsq12);
838 rinv13 = sse2_invsqrt_f(rsq13);
839 rinv21 = sse2_invsqrt_f(rsq21);
840 rinv22 = sse2_invsqrt_f(rsq22);
841 rinv23 = sse2_invsqrt_f(rsq23);
842 rinv31 = sse2_invsqrt_f(rsq31);
843 rinv32 = sse2_invsqrt_f(rsq32);
844 rinv33 = sse2_invsqrt_f(rsq33);
845
846 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
847 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
848 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
849 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
850 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
851 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
852 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
853 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
854 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
855
856 fjx1 = _mm_setzero_ps();
857 fjy1 = _mm_setzero_ps();
858 fjz1 = _mm_setzero_ps();
859 fjx2 = _mm_setzero_ps();
860 fjy2 = _mm_setzero_ps();
861 fjz2 = _mm_setzero_ps();
862 fjx3 = _mm_setzero_ps();
863 fjy3 = _mm_setzero_ps();
864 fjz3 = _mm_setzero_ps();
865
866 /**************************
867 * CALCULATE INTERACTIONS *
868 **************************/
869
870 if (gmx_mm_any_lt(rsq11,rcutoff2))
871 {
872
873 r11 = _mm_mul_ps(rsq11,rinv11);
874 r11 = _mm_andnot_ps(dummy_mask,r11);
875
876 /* EWALD ELECTROSTATICS */
877
878 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
879 ewrt = _mm_mul_ps(r11,ewtabscale);
880 ewitab = _mm_cvttps_epi32(ewrt);
881 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
882 ewitab = _mm_slli_epi32(ewitab,2);
883 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
884 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
885 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
886 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
887 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
888 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
889 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
890 velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_sub_ps(rinv11,sh_ewald),velec));
891 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
892
893 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
894
895 /* Update potential sum for this i atom from the interaction with this j atom. */
896 velec = _mm_and_ps(velec,cutoff_mask);
897 velec = _mm_andnot_ps(dummy_mask,velec);
898 velecsum = _mm_add_ps(velecsum,velec);
899
900 fscal = felec;
901
902 fscal = _mm_and_ps(fscal,cutoff_mask);
903
904 fscal = _mm_andnot_ps(dummy_mask,fscal);
905
906 /* Calculate temporary vectorial force */
907 tx = _mm_mul_ps(fscal,dx11);
908 ty = _mm_mul_ps(fscal,dy11);
909 tz = _mm_mul_ps(fscal,dz11);
910
911 /* Update vectorial force */
912 fix1 = _mm_add_ps(fix1,tx);
913 fiy1 = _mm_add_ps(fiy1,ty);
914 fiz1 = _mm_add_ps(fiz1,tz);
915
916 fjx1 = _mm_add_ps(fjx1,tx);
917 fjy1 = _mm_add_ps(fjy1,ty);
918 fjz1 = _mm_add_ps(fjz1,tz);
919
920 }
921
922 /**************************
923 * CALCULATE INTERACTIONS *
924 **************************/
925
926 if (gmx_mm_any_lt(rsq12,rcutoff2))
927 {
928
929 r12 = _mm_mul_ps(rsq12,rinv12);
930 r12 = _mm_andnot_ps(dummy_mask,r12);
931
932 /* EWALD ELECTROSTATICS */
933
934 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
935 ewrt = _mm_mul_ps(r12,ewtabscale);
936 ewitab = _mm_cvttps_epi32(ewrt);
937 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
938 ewitab = _mm_slli_epi32(ewitab,2);
939 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
940 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
941 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
942 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
943 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
944 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
945 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
946 velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_sub_ps(rinv12,sh_ewald),velec));
947 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
948
949 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
950
951 /* Update potential sum for this i atom from the interaction with this j atom. */
952 velec = _mm_and_ps(velec,cutoff_mask);
953 velec = _mm_andnot_ps(dummy_mask,velec);
954 velecsum = _mm_add_ps(velecsum,velec);
955
956 fscal = felec;
957
958 fscal = _mm_and_ps(fscal,cutoff_mask);
959
960 fscal = _mm_andnot_ps(dummy_mask,fscal);
961
962 /* Calculate temporary vectorial force */
963 tx = _mm_mul_ps(fscal,dx12);
964 ty = _mm_mul_ps(fscal,dy12);
965 tz = _mm_mul_ps(fscal,dz12);
966
967 /* Update vectorial force */
968 fix1 = _mm_add_ps(fix1,tx);
969 fiy1 = _mm_add_ps(fiy1,ty);
970 fiz1 = _mm_add_ps(fiz1,tz);
971
972 fjx2 = _mm_add_ps(fjx2,tx);
973 fjy2 = _mm_add_ps(fjy2,ty);
974 fjz2 = _mm_add_ps(fjz2,tz);
975
976 }
977
978 /**************************
979 * CALCULATE INTERACTIONS *
980 **************************/
981
982 if (gmx_mm_any_lt(rsq13,rcutoff2))
983 {
984
985 r13 = _mm_mul_ps(rsq13,rinv13);
986 r13 = _mm_andnot_ps(dummy_mask,r13);
987
988 /* EWALD ELECTROSTATICS */
989
990 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
991 ewrt = _mm_mul_ps(r13,ewtabscale);
992 ewitab = _mm_cvttps_epi32(ewrt);
993 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
994 ewitab = _mm_slli_epi32(ewitab,2);
995 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
996 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
997 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
998 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
999 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1000 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1001 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1002 velec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_sub_ps(rinv13,sh_ewald),velec));
1003 felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
1004
1005 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
1006
1007 /* Update potential sum for this i atom from the interaction with this j atom. */
1008 velec = _mm_and_ps(velec,cutoff_mask);
1009 velec = _mm_andnot_ps(dummy_mask,velec);
1010 velecsum = _mm_add_ps(velecsum,velec);
1011
1012 fscal = felec;
1013
1014 fscal = _mm_and_ps(fscal,cutoff_mask);
1015
1016 fscal = _mm_andnot_ps(dummy_mask,fscal);
1017
1018 /* Calculate temporary vectorial force */
1019 tx = _mm_mul_ps(fscal,dx13);
1020 ty = _mm_mul_ps(fscal,dy13);
1021 tz = _mm_mul_ps(fscal,dz13);
1022
1023 /* Update vectorial force */
1024 fix1 = _mm_add_ps(fix1,tx);
1025 fiy1 = _mm_add_ps(fiy1,ty);
1026 fiz1 = _mm_add_ps(fiz1,tz);
1027
1028 fjx3 = _mm_add_ps(fjx3,tx);
1029 fjy3 = _mm_add_ps(fjy3,ty);
1030 fjz3 = _mm_add_ps(fjz3,tz);
1031
1032 }
1033
1034 /**************************
1035 * CALCULATE INTERACTIONS *
1036 **************************/
1037
1038 if (gmx_mm_any_lt(rsq21,rcutoff2))
1039 {
1040
1041 r21 = _mm_mul_ps(rsq21,rinv21);
1042 r21 = _mm_andnot_ps(dummy_mask,r21);
1043
1044 /* EWALD ELECTROSTATICS */
1045
1046 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1047 ewrt = _mm_mul_ps(r21,ewtabscale);
1048 ewitab = _mm_cvttps_epi32(ewrt);
1049 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1050 ewitab = _mm_slli_epi32(ewitab,2);
1051 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1052 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1053 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1054 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1055 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1056 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1057 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1058 velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_sub_ps(rinv21,sh_ewald),velec));
1059 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1060
1061 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
1062
1063 /* Update potential sum for this i atom from the interaction with this j atom. */
1064 velec = _mm_and_ps(velec,cutoff_mask);
1065 velec = _mm_andnot_ps(dummy_mask,velec);
1066 velecsum = _mm_add_ps(velecsum,velec);
1067
1068 fscal = felec;
1069
1070 fscal = _mm_and_ps(fscal,cutoff_mask);
1071
1072 fscal = _mm_andnot_ps(dummy_mask,fscal);
1073
1074 /* Calculate temporary vectorial force */
1075 tx = _mm_mul_ps(fscal,dx21);
1076 ty = _mm_mul_ps(fscal,dy21);
1077 tz = _mm_mul_ps(fscal,dz21);
1078
1079 /* Update vectorial force */
1080 fix2 = _mm_add_ps(fix2,tx);
1081 fiy2 = _mm_add_ps(fiy2,ty);
1082 fiz2 = _mm_add_ps(fiz2,tz);
1083
1084 fjx1 = _mm_add_ps(fjx1,tx);
1085 fjy1 = _mm_add_ps(fjy1,ty);
1086 fjz1 = _mm_add_ps(fjz1,tz);
1087
1088 }
1089
1090 /**************************
1091 * CALCULATE INTERACTIONS *
1092 **************************/
1093
1094 if (gmx_mm_any_lt(rsq22,rcutoff2))
1095 {
1096
1097 r22 = _mm_mul_ps(rsq22,rinv22);
1098 r22 = _mm_andnot_ps(dummy_mask,r22);
1099
1100 /* EWALD ELECTROSTATICS */
1101
1102 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1103 ewrt = _mm_mul_ps(r22,ewtabscale);
1104 ewitab = _mm_cvttps_epi32(ewrt);
1105 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1106 ewitab = _mm_slli_epi32(ewitab,2);
1107 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1108 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1109 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1110 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1111 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1112 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1113 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1114 velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_sub_ps(rinv22,sh_ewald),velec));
1115 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
1116
1117 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
1118
1119 /* Update potential sum for this i atom from the interaction with this j atom. */
1120 velec = _mm_and_ps(velec,cutoff_mask);
1121 velec = _mm_andnot_ps(dummy_mask,velec);
1122 velecsum = _mm_add_ps(velecsum,velec);
1123
1124 fscal = felec;
1125
1126 fscal = _mm_and_ps(fscal,cutoff_mask);
1127
1128 fscal = _mm_andnot_ps(dummy_mask,fscal);
1129
1130 /* Calculate temporary vectorial force */
1131 tx = _mm_mul_ps(fscal,dx22);
1132 ty = _mm_mul_ps(fscal,dy22);
1133 tz = _mm_mul_ps(fscal,dz22);
1134
1135 /* Update vectorial force */
1136 fix2 = _mm_add_ps(fix2,tx);
1137 fiy2 = _mm_add_ps(fiy2,ty);
1138 fiz2 = _mm_add_ps(fiz2,tz);
1139
1140 fjx2 = _mm_add_ps(fjx2,tx);
1141 fjy2 = _mm_add_ps(fjy2,ty);
1142 fjz2 = _mm_add_ps(fjz2,tz);
1143
1144 }
1145
1146 /**************************
1147 * CALCULATE INTERACTIONS *
1148 **************************/
1149
1150 if (gmx_mm_any_lt(rsq23,rcutoff2))
1151 {
1152
1153 r23 = _mm_mul_ps(rsq23,rinv23);
1154 r23 = _mm_andnot_ps(dummy_mask,r23);
1155
1156 /* EWALD ELECTROSTATICS */
1157
1158 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1159 ewrt = _mm_mul_ps(r23,ewtabscale);
1160 ewitab = _mm_cvttps_epi32(ewrt);
1161 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1162 ewitab = _mm_slli_epi32(ewitab,2);
1163 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1164 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1165 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1166 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1167 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1168 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1169 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1170 velec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_sub_ps(rinv23,sh_ewald),velec));
1171 felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
1172
1173 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
1174
1175 /* Update potential sum for this i atom from the interaction with this j atom. */
1176 velec = _mm_and_ps(velec,cutoff_mask);
1177 velec = _mm_andnot_ps(dummy_mask,velec);
1178 velecsum = _mm_add_ps(velecsum,velec);
1179
1180 fscal = felec;
1181
1182 fscal = _mm_and_ps(fscal,cutoff_mask);
1183
1184 fscal = _mm_andnot_ps(dummy_mask,fscal);
1185
1186 /* Calculate temporary vectorial force */
1187 tx = _mm_mul_ps(fscal,dx23);
1188 ty = _mm_mul_ps(fscal,dy23);
1189 tz = _mm_mul_ps(fscal,dz23);
1190
1191 /* Update vectorial force */
1192 fix2 = _mm_add_ps(fix2,tx);
1193 fiy2 = _mm_add_ps(fiy2,ty);
1194 fiz2 = _mm_add_ps(fiz2,tz);
1195
1196 fjx3 = _mm_add_ps(fjx3,tx);
1197 fjy3 = _mm_add_ps(fjy3,ty);
1198 fjz3 = _mm_add_ps(fjz3,tz);
1199
1200 }
1201
1202 /**************************
1203 * CALCULATE INTERACTIONS *
1204 **************************/
1205
1206 if (gmx_mm_any_lt(rsq31,rcutoff2))
1207 {
1208
1209 r31 = _mm_mul_ps(rsq31,rinv31);
1210 r31 = _mm_andnot_ps(dummy_mask,r31);
1211
1212 /* EWALD ELECTROSTATICS */
1213
1214 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1215 ewrt = _mm_mul_ps(r31,ewtabscale);
1216 ewitab = _mm_cvttps_epi32(ewrt);
1217 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1218 ewitab = _mm_slli_epi32(ewitab,2);
1219 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1220 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1221 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1222 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1223 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1224 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1225 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1226 velec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_sub_ps(rinv31,sh_ewald),velec));
1227 felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
1228
1229 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
1230
1231 /* Update potential sum for this i atom from the interaction with this j atom. */
1232 velec = _mm_and_ps(velec,cutoff_mask);
1233 velec = _mm_andnot_ps(dummy_mask,velec);
1234 velecsum = _mm_add_ps(velecsum,velec);
1235
1236 fscal = felec;
1237
1238 fscal = _mm_and_ps(fscal,cutoff_mask);
1239
1240 fscal = _mm_andnot_ps(dummy_mask,fscal);
1241
1242 /* Calculate temporary vectorial force */
1243 tx = _mm_mul_ps(fscal,dx31);
1244 ty = _mm_mul_ps(fscal,dy31);
1245 tz = _mm_mul_ps(fscal,dz31);
1246
1247 /* Update vectorial force */
1248 fix3 = _mm_add_ps(fix3,tx);
1249 fiy3 = _mm_add_ps(fiy3,ty);
1250 fiz3 = _mm_add_ps(fiz3,tz);
1251
1252 fjx1 = _mm_add_ps(fjx1,tx);
1253 fjy1 = _mm_add_ps(fjy1,ty);
1254 fjz1 = _mm_add_ps(fjz1,tz);
1255
1256 }
1257
1258 /**************************
1259 * CALCULATE INTERACTIONS *
1260 **************************/
1261
1262 if (gmx_mm_any_lt(rsq32,rcutoff2))
1263 {
1264
1265 r32 = _mm_mul_ps(rsq32,rinv32);
1266 r32 = _mm_andnot_ps(dummy_mask,r32);
1267
1268 /* EWALD ELECTROSTATICS */
1269
1270 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1271 ewrt = _mm_mul_ps(r32,ewtabscale);
1272 ewitab = _mm_cvttps_epi32(ewrt);
1273 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1274 ewitab = _mm_slli_epi32(ewitab,2);
1275 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1276 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1277 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1278 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1279 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1280 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1281 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1282 velec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_sub_ps(rinv32,sh_ewald),velec));
1283 felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
1284
1285 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
1286
1287 /* Update potential sum for this i atom from the interaction with this j atom. */
1288 velec = _mm_and_ps(velec,cutoff_mask);
1289 velec = _mm_andnot_ps(dummy_mask,velec);
1290 velecsum = _mm_add_ps(velecsum,velec);
1291
1292 fscal = felec;
1293
1294 fscal = _mm_and_ps(fscal,cutoff_mask);
1295
1296 fscal = _mm_andnot_ps(dummy_mask,fscal);
1297
1298 /* Calculate temporary vectorial force */
1299 tx = _mm_mul_ps(fscal,dx32);
1300 ty = _mm_mul_ps(fscal,dy32);
1301 tz = _mm_mul_ps(fscal,dz32);
1302
1303 /* Update vectorial force */
1304 fix3 = _mm_add_ps(fix3,tx);
1305 fiy3 = _mm_add_ps(fiy3,ty);
1306 fiz3 = _mm_add_ps(fiz3,tz);
1307
1308 fjx2 = _mm_add_ps(fjx2,tx);
1309 fjy2 = _mm_add_ps(fjy2,ty);
1310 fjz2 = _mm_add_ps(fjz2,tz);
1311
1312 }
1313
1314 /**************************
1315 * CALCULATE INTERACTIONS *
1316 **************************/
1317
1318 if (gmx_mm_any_lt(rsq33,rcutoff2))
1319 {
1320
1321 r33 = _mm_mul_ps(rsq33,rinv33);
1322 r33 = _mm_andnot_ps(dummy_mask,r33);
1323
1324 /* EWALD ELECTROSTATICS */
1325
1326 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1327 ewrt = _mm_mul_ps(r33,ewtabscale);
1328 ewitab = _mm_cvttps_epi32(ewrt);
1329 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1330 ewitab = _mm_slli_epi32(ewitab,2);
1331 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1332 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1333 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1334 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1335 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1336 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1337 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1338 velec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_sub_ps(rinv33,sh_ewald),velec));
1339 felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
1340
1341 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
1342
1343 /* Update potential sum for this i atom from the interaction with this j atom. */
1344 velec = _mm_and_ps(velec,cutoff_mask);
1345 velec = _mm_andnot_ps(dummy_mask,velec);
1346 velecsum = _mm_add_ps(velecsum,velec);
1347
1348 fscal = felec;
1349
1350 fscal = _mm_and_ps(fscal,cutoff_mask);
1351
1352 fscal = _mm_andnot_ps(dummy_mask,fscal);
1353
1354 /* Calculate temporary vectorial force */
1355 tx = _mm_mul_ps(fscal,dx33);
1356 ty = _mm_mul_ps(fscal,dy33);
1357 tz = _mm_mul_ps(fscal,dz33);
1358
1359 /* Update vectorial force */
1360 fix3 = _mm_add_ps(fix3,tx);
1361 fiy3 = _mm_add_ps(fiy3,ty);
1362 fiz3 = _mm_add_ps(fiz3,tz);
1363
1364 fjx3 = _mm_add_ps(fjx3,tx);
1365 fjy3 = _mm_add_ps(fjy3,ty);
1366 fjz3 = _mm_add_ps(fjz3,tz);
1367
1368 }
1369
1370 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1371 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1372 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1373 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1374
1375 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1376 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1377
1378 /* Inner loop uses 423 flops */
1379 }
1380
1381 /* End of innermost loop */
1382
1383 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1384 f+i_coord_offset+DIM,fshift+i_shift_offset);
1385
1386 ggid = gid[iidx];
1387 /* Update potential energies */
1388 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1389
1390 /* Increment number of inner iterations */
1391 inneriter += j_index_end - j_index_start;
1392
1393 /* Outer loop uses 19 flops */
1394 }
1395
1396 /* Increment number of outer iterations */
1397 outeriter += nri;
1398
1399 /* Update outer/inner flops */
1400
1401 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*423);
1402 }
1403 /*
1404 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_sse2_single
1405 * Electrostatics interaction: Ewald
1406 * VdW interaction: None
1407 * Geometry: Water4-Water4
1408 * Calculate force/pot: Force
1409 */
1410 void
1411 nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_sse2_single
1412 (t_nblist * gmx_restrict nlist,
1413 rvec * gmx_restrict xx,
1414 rvec * gmx_restrict ff,
1415 struct t_forcerec * gmx_restrict fr,
1416 t_mdatoms * gmx_restrict mdatoms,
1417 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1418 t_nrnb * gmx_restrict nrnb)
1419 {
1420 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1421 * just 0 for non-waters.
1422 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1423 * jnr indices corresponding to data put in the four positions in the SIMD register.
1424 */
1425 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1426 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1427 int jnrA,jnrB,jnrC,jnrD;
1428 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1429 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1430 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1431 real rcutoff_scalar;
1432 real *shiftvec,*fshift,*x,*f;
1433 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1434 real scratch[4*DIM];
1435 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1436 int vdwioffset1;
1437 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1438 int vdwioffset2;
1439 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1440 int vdwioffset3;
1441 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1442 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1443 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1444 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1445 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1446 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1447 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1448 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1449 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1450 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1451 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1452 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1453 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1454 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1455 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1456 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1457 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
1458 real *charge;
1459 __m128i ewitab;
1460 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1461 real *ewtab;
1462 __m128 dummy_mask,cutoff_mask;
1463 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1464 __m128 one = _mm_set1_ps(1.0);
1465 __m128 two = _mm_set1_ps(2.0);
1466 x = xx[0];
1467 f = ff[0];
1468
1469 nri = nlist->nri;
1470 iinr = nlist->iinr;
1471 jindex = nlist->jindex;
1472 jjnr = nlist->jjnr;
1473 shiftidx = nlist->shift;
1474 gid = nlist->gid;
1475 shiftvec = fr->shift_vec[0];
1476 fshift = fr->fshift[0];
1477 facel = _mm_set1_ps(fr->ic->epsfac);
1478 charge = mdatoms->chargeA;
1479
1480 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
1481 ewtab = fr->ic->tabq_coul_F;
1482 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
1483 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1484
1485 /* Setup water-specific parameters */
1486 inr = nlist->iinr[0];
1487 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1488 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1489 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1490
1491 jq1 = _mm_set1_ps(charge[inr+1]);
1492 jq2 = _mm_set1_ps(charge[inr+2]);
1493 jq3 = _mm_set1_ps(charge[inr+3]);
1494 qq11 = _mm_mul_ps(iq1,jq1);
1495 qq12 = _mm_mul_ps(iq1,jq2);
1496 qq13 = _mm_mul_ps(iq1,jq3);
1497 qq21 = _mm_mul_ps(iq2,jq1);
1498 qq22 = _mm_mul_ps(iq2,jq2);
1499 qq23 = _mm_mul_ps(iq2,jq3);
1500 qq31 = _mm_mul_ps(iq3,jq1);
1501 qq32 = _mm_mul_ps(iq3,jq2);
1502 qq33 = _mm_mul_ps(iq3,jq3);
1503
1504 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1505 rcutoff_scalar = fr->ic->rcoulomb;
1506 rcutoff = _mm_set1_ps(rcutoff_scalar);
1507 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
1508
1509 /* Avoid stupid compiler warnings */
1510 jnrA = jnrB = jnrC = jnrD = 0;
1511 j_coord_offsetA = 0;
1512 j_coord_offsetB = 0;
1513 j_coord_offsetC = 0;
1514 j_coord_offsetD = 0;
1515
1516 outeriter = 0;
1517 inneriter = 0;
1518
1519 for(iidx=0;iidx<4*DIM;iidx++)
1520 {
1521 scratch[iidx] = 0.0;
1522 }
1523
1524 /* Start outer loop over neighborlists */
1525 for(iidx=0; iidx<nri; iidx++)
1526 {
1527 /* Load shift vector for this list */
1528 i_shift_offset = DIM*shiftidx[iidx];
1529
1530 /* Load limits for loop over neighbors */
1531 j_index_start = jindex[iidx];
1532 j_index_end = jindex[iidx+1];
1533
1534 /* Get outer coordinate index */
1535 inr = iinr[iidx];
1536 i_coord_offset = DIM*inr;
1537
1538 /* Load i particle coords and add shift vector */
1539 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1540 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1541
1542 fix1 = _mm_setzero_ps();
1543 fiy1 = _mm_setzero_ps();
1544 fiz1 = _mm_setzero_ps();
1545 fix2 = _mm_setzero_ps();
1546 fiy2 = _mm_setzero_ps();
1547 fiz2 = _mm_setzero_ps();
1548 fix3 = _mm_setzero_ps();
1549 fiy3 = _mm_setzero_ps();
1550 fiz3 = _mm_setzero_ps();
1551
1552 /* Start inner kernel loop */
1553 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1554 {
1555
1556 /* Get j neighbor index, and coordinate index */
1557 jnrA = jjnr[jidx];
1558 jnrB = jjnr[jidx+1];
1559 jnrC = jjnr[jidx+2];
1560 jnrD = jjnr[jidx+3];
1561 j_coord_offsetA = DIM*jnrA;
1562 j_coord_offsetB = DIM*jnrB;
1563 j_coord_offsetC = DIM*jnrC;
1564 j_coord_offsetD = DIM*jnrD;
1565
1566 /* load j atom coordinates */
1567 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1568 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1569 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1570
1571 /* Calculate displacement vector */
1572 dx11 = _mm_sub_ps(ix1,jx1);
1573 dy11 = _mm_sub_ps(iy1,jy1);
1574 dz11 = _mm_sub_ps(iz1,jz1);
1575 dx12 = _mm_sub_ps(ix1,jx2);
1576 dy12 = _mm_sub_ps(iy1,jy2);
1577 dz12 = _mm_sub_ps(iz1,jz2);
1578 dx13 = _mm_sub_ps(ix1,jx3);
1579 dy13 = _mm_sub_ps(iy1,jy3);
1580 dz13 = _mm_sub_ps(iz1,jz3);
1581 dx21 = _mm_sub_ps(ix2,jx1);
1582 dy21 = _mm_sub_ps(iy2,jy1);
1583 dz21 = _mm_sub_ps(iz2,jz1);
1584 dx22 = _mm_sub_ps(ix2,jx2);
1585 dy22 = _mm_sub_ps(iy2,jy2);
1586 dz22 = _mm_sub_ps(iz2,jz2);
1587 dx23 = _mm_sub_ps(ix2,jx3);
1588 dy23 = _mm_sub_ps(iy2,jy3);
1589 dz23 = _mm_sub_ps(iz2,jz3);
1590 dx31 = _mm_sub_ps(ix3,jx1);
1591 dy31 = _mm_sub_ps(iy3,jy1);
1592 dz31 = _mm_sub_ps(iz3,jz1);
1593 dx32 = _mm_sub_ps(ix3,jx2);
1594 dy32 = _mm_sub_ps(iy3,jy2);
1595 dz32 = _mm_sub_ps(iz3,jz2);
1596 dx33 = _mm_sub_ps(ix3,jx3);
1597 dy33 = _mm_sub_ps(iy3,jy3);
1598 dz33 = _mm_sub_ps(iz3,jz3);
1599
1600 /* Calculate squared distance and things based on it */
1601 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1602 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1603 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1604 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1605 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1606 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1607 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1608 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1609 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1610
1611 rinv11 = sse2_invsqrt_f(rsq11);
1612 rinv12 = sse2_invsqrt_f(rsq12);
1613 rinv13 = sse2_invsqrt_f(rsq13);
1614 rinv21 = sse2_invsqrt_f(rsq21);
1615 rinv22 = sse2_invsqrt_f(rsq22);
1616 rinv23 = sse2_invsqrt_f(rsq23);
1617 rinv31 = sse2_invsqrt_f(rsq31);
1618 rinv32 = sse2_invsqrt_f(rsq32);
1619 rinv33 = sse2_invsqrt_f(rsq33);
1620
1621 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
1622 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
1623 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
1624 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
1625 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
1626 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
1627 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
1628 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
1629 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
1630
1631 fjx1 = _mm_setzero_ps();
1632 fjy1 = _mm_setzero_ps();
1633 fjz1 = _mm_setzero_ps();
1634 fjx2 = _mm_setzero_ps();
1635 fjy2 = _mm_setzero_ps();
1636 fjz2 = _mm_setzero_ps();
1637 fjx3 = _mm_setzero_ps();
1638 fjy3 = _mm_setzero_ps();
1639 fjz3 = _mm_setzero_ps();
1640
1641 /**************************
1642 * CALCULATE INTERACTIONS *
1643 **************************/
1644
1645 if (gmx_mm_any_lt(rsq11,rcutoff2))
1646 {
1647
1648 r11 = _mm_mul_ps(rsq11,rinv11);
1649
1650 /* EWALD ELECTROSTATICS */
1651
1652 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1653 ewrt = _mm_mul_ps(r11,ewtabscale);
1654 ewitab = _mm_cvttps_epi32(ewrt);
1655 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1656 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1657 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1658 &ewtabF,&ewtabFn);
1659 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1660 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1661
1662 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
1663
1664 fscal = felec;
1665
1666 fscal = _mm_and_ps(fscal,cutoff_mask);
1667
1668 /* Calculate temporary vectorial force */
1669 tx = _mm_mul_ps(fscal,dx11);
1670 ty = _mm_mul_ps(fscal,dy11);
1671 tz = _mm_mul_ps(fscal,dz11);
1672
1673 /* Update vectorial force */
1674 fix1 = _mm_add_ps(fix1,tx);
1675 fiy1 = _mm_add_ps(fiy1,ty);
1676 fiz1 = _mm_add_ps(fiz1,tz);
1677
1678 fjx1 = _mm_add_ps(fjx1,tx);
1679 fjy1 = _mm_add_ps(fjy1,ty);
1680 fjz1 = _mm_add_ps(fjz1,tz);
1681
1682 }
1683
1684 /**************************
1685 * CALCULATE INTERACTIONS *
1686 **************************/
1687
1688 if (gmx_mm_any_lt(rsq12,rcutoff2))
1689 {
1690
1691 r12 = _mm_mul_ps(rsq12,rinv12);
1692
1693 /* EWALD ELECTROSTATICS */
1694
1695 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1696 ewrt = _mm_mul_ps(r12,ewtabscale);
1697 ewitab = _mm_cvttps_epi32(ewrt);
1698 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1699 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1700 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1701 &ewtabF,&ewtabFn);
1702 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1703 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1704
1705 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
1706
1707 fscal = felec;
1708
1709 fscal = _mm_and_ps(fscal,cutoff_mask);
1710
1711 /* Calculate temporary vectorial force */
1712 tx = _mm_mul_ps(fscal,dx12);
1713 ty = _mm_mul_ps(fscal,dy12);
1714 tz = _mm_mul_ps(fscal,dz12);
1715
1716 /* Update vectorial force */
1717 fix1 = _mm_add_ps(fix1,tx);
1718 fiy1 = _mm_add_ps(fiy1,ty);
1719 fiz1 = _mm_add_ps(fiz1,tz);
1720
1721 fjx2 = _mm_add_ps(fjx2,tx);
1722 fjy2 = _mm_add_ps(fjy2,ty);
1723 fjz2 = _mm_add_ps(fjz2,tz);
1724
1725 }
1726
1727 /**************************
1728 * CALCULATE INTERACTIONS *
1729 **************************/
1730
1731 if (gmx_mm_any_lt(rsq13,rcutoff2))
1732 {
1733
1734 r13 = _mm_mul_ps(rsq13,rinv13);
1735
1736 /* EWALD ELECTROSTATICS */
1737
1738 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1739 ewrt = _mm_mul_ps(r13,ewtabscale);
1740 ewitab = _mm_cvttps_epi32(ewrt);
1741 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1742 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1743 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1744 &ewtabF,&ewtabFn);
1745 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1746 felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
1747
1748 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
1749
1750 fscal = felec;
1751
1752 fscal = _mm_and_ps(fscal,cutoff_mask);
1753
1754 /* Calculate temporary vectorial force */
1755 tx = _mm_mul_ps(fscal,dx13);
1756 ty = _mm_mul_ps(fscal,dy13);
1757 tz = _mm_mul_ps(fscal,dz13);
1758
1759 /* Update vectorial force */
1760 fix1 = _mm_add_ps(fix1,tx);
1761 fiy1 = _mm_add_ps(fiy1,ty);
1762 fiz1 = _mm_add_ps(fiz1,tz);
1763
1764 fjx3 = _mm_add_ps(fjx3,tx);
1765 fjy3 = _mm_add_ps(fjy3,ty);
1766 fjz3 = _mm_add_ps(fjz3,tz);
1767
1768 }
1769
1770 /**************************
1771 * CALCULATE INTERACTIONS *
1772 **************************/
1773
1774 if (gmx_mm_any_lt(rsq21,rcutoff2))
1775 {
1776
1777 r21 = _mm_mul_ps(rsq21,rinv21);
1778
1779 /* EWALD ELECTROSTATICS */
1780
1781 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1782 ewrt = _mm_mul_ps(r21,ewtabscale);
1783 ewitab = _mm_cvttps_epi32(ewrt);
1784 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1785 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1786 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1787 &ewtabF,&ewtabFn);
1788 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1789 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1790
1791 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
1792
1793 fscal = felec;
1794
1795 fscal = _mm_and_ps(fscal,cutoff_mask);
1796
1797 /* Calculate temporary vectorial force */
1798 tx = _mm_mul_ps(fscal,dx21);
1799 ty = _mm_mul_ps(fscal,dy21);
1800 tz = _mm_mul_ps(fscal,dz21);
1801
1802 /* Update vectorial force */
1803 fix2 = _mm_add_ps(fix2,tx);
1804 fiy2 = _mm_add_ps(fiy2,ty);
1805 fiz2 = _mm_add_ps(fiz2,tz);
1806
1807 fjx1 = _mm_add_ps(fjx1,tx);
1808 fjy1 = _mm_add_ps(fjy1,ty);
1809 fjz1 = _mm_add_ps(fjz1,tz);
1810
1811 }
1812
1813 /**************************
1814 * CALCULATE INTERACTIONS *
1815 **************************/
1816
1817 if (gmx_mm_any_lt(rsq22,rcutoff2))
1818 {
1819
1820 r22 = _mm_mul_ps(rsq22,rinv22);
1821
1822 /* EWALD ELECTROSTATICS */
1823
1824 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1825 ewrt = _mm_mul_ps(r22,ewtabscale);
1826 ewitab = _mm_cvttps_epi32(ewrt);
1827 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1828 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1829 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1830 &ewtabF,&ewtabFn);
1831 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1832 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
1833
1834 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
1835
1836 fscal = felec;
1837
1838 fscal = _mm_and_ps(fscal,cutoff_mask);
1839
1840 /* Calculate temporary vectorial force */
1841 tx = _mm_mul_ps(fscal,dx22);
1842 ty = _mm_mul_ps(fscal,dy22);
1843 tz = _mm_mul_ps(fscal,dz22);
1844
1845 /* Update vectorial force */
1846 fix2 = _mm_add_ps(fix2,tx);
1847 fiy2 = _mm_add_ps(fiy2,ty);
1848 fiz2 = _mm_add_ps(fiz2,tz);
1849
1850 fjx2 = _mm_add_ps(fjx2,tx);
1851 fjy2 = _mm_add_ps(fjy2,ty);
1852 fjz2 = _mm_add_ps(fjz2,tz);
1853
1854 }
1855
1856 /**************************
1857 * CALCULATE INTERACTIONS *
1858 **************************/
1859
1860 if (gmx_mm_any_lt(rsq23,rcutoff2))
1861 {
1862
1863 r23 = _mm_mul_ps(rsq23,rinv23);
1864
1865 /* EWALD ELECTROSTATICS */
1866
1867 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1868 ewrt = _mm_mul_ps(r23,ewtabscale);
1869 ewitab = _mm_cvttps_epi32(ewrt);
1870 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1871 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1872 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1873 &ewtabF,&ewtabFn);
1874 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1875 felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
1876
1877 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
1878
1879 fscal = felec;
1880
1881 fscal = _mm_and_ps(fscal,cutoff_mask);
1882
1883 /* Calculate temporary vectorial force */
1884 tx = _mm_mul_ps(fscal,dx23);
1885 ty = _mm_mul_ps(fscal,dy23);
1886 tz = _mm_mul_ps(fscal,dz23);
1887
1888 /* Update vectorial force */
1889 fix2 = _mm_add_ps(fix2,tx);
1890 fiy2 = _mm_add_ps(fiy2,ty);
1891 fiz2 = _mm_add_ps(fiz2,tz);
1892
1893 fjx3 = _mm_add_ps(fjx3,tx);
1894 fjy3 = _mm_add_ps(fjy3,ty);
1895 fjz3 = _mm_add_ps(fjz3,tz);
1896
1897 }
1898
1899 /**************************
1900 * CALCULATE INTERACTIONS *
1901 **************************/
1902
1903 if (gmx_mm_any_lt(rsq31,rcutoff2))
1904 {
1905
1906 r31 = _mm_mul_ps(rsq31,rinv31);
1907
1908 /* EWALD ELECTROSTATICS */
1909
1910 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1911 ewrt = _mm_mul_ps(r31,ewtabscale);
1912 ewitab = _mm_cvttps_epi32(ewrt);
1913 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1914 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1915 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1916 &ewtabF,&ewtabFn);
1917 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1918 felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
1919
1920 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
1921
1922 fscal = felec;
1923
1924 fscal = _mm_and_ps(fscal,cutoff_mask);
1925
1926 /* Calculate temporary vectorial force */
1927 tx = _mm_mul_ps(fscal,dx31);
1928 ty = _mm_mul_ps(fscal,dy31);
1929 tz = _mm_mul_ps(fscal,dz31);
1930
1931 /* Update vectorial force */
1932 fix3 = _mm_add_ps(fix3,tx);
1933 fiy3 = _mm_add_ps(fiy3,ty);
1934 fiz3 = _mm_add_ps(fiz3,tz);
1935
1936 fjx1 = _mm_add_ps(fjx1,tx);
1937 fjy1 = _mm_add_ps(fjy1,ty);
1938 fjz1 = _mm_add_ps(fjz1,tz);
1939
1940 }
1941
1942 /**************************
1943 * CALCULATE INTERACTIONS *
1944 **************************/
1945
1946 if (gmx_mm_any_lt(rsq32,rcutoff2))
1947 {
1948
1949 r32 = _mm_mul_ps(rsq32,rinv32);
1950
1951 /* EWALD ELECTROSTATICS */
1952
1953 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1954 ewrt = _mm_mul_ps(r32,ewtabscale);
1955 ewitab = _mm_cvttps_epi32(ewrt);
1956 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1957 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1958 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1959 &ewtabF,&ewtabFn);
1960 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1961 felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
1962
1963 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
1964
1965 fscal = felec;
1966
1967 fscal = _mm_and_ps(fscal,cutoff_mask);
1968
1969 /* Calculate temporary vectorial force */
1970 tx = _mm_mul_ps(fscal,dx32);
1971 ty = _mm_mul_ps(fscal,dy32);
1972 tz = _mm_mul_ps(fscal,dz32);
1973
1974 /* Update vectorial force */
1975 fix3 = _mm_add_ps(fix3,tx);
1976 fiy3 = _mm_add_ps(fiy3,ty);
1977 fiz3 = _mm_add_ps(fiz3,tz);
1978
1979 fjx2 = _mm_add_ps(fjx2,tx);
1980 fjy2 = _mm_add_ps(fjy2,ty);
1981 fjz2 = _mm_add_ps(fjz2,tz);
1982
1983 }
1984
1985 /**************************
1986 * CALCULATE INTERACTIONS *
1987 **************************/
1988
1989 if (gmx_mm_any_lt(rsq33,rcutoff2))
1990 {
1991
1992 r33 = _mm_mul_ps(rsq33,rinv33);
1993
1994 /* EWALD ELECTROSTATICS */
1995
1996 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1997 ewrt = _mm_mul_ps(r33,ewtabscale);
1998 ewitab = _mm_cvttps_epi32(ewrt);
1999 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2000 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2001 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2002 &ewtabF,&ewtabFn);
2003 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2004 felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
2005
2006 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
2007
2008 fscal = felec;
2009
2010 fscal = _mm_and_ps(fscal,cutoff_mask);
2011
2012 /* Calculate temporary vectorial force */
2013 tx = _mm_mul_ps(fscal,dx33);
2014 ty = _mm_mul_ps(fscal,dy33);
2015 tz = _mm_mul_ps(fscal,dz33);
2016
2017 /* Update vectorial force */
2018 fix3 = _mm_add_ps(fix3,tx);
2019 fiy3 = _mm_add_ps(fiy3,ty);
2020 fiz3 = _mm_add_ps(fiz3,tz);
2021
2022 fjx3 = _mm_add_ps(fjx3,tx);
2023 fjy3 = _mm_add_ps(fjy3,ty);
2024 fjz3 = _mm_add_ps(fjz3,tz);
2025
2026 }
2027
2028 fjptrA = f+j_coord_offsetA;
2029 fjptrB = f+j_coord_offsetB;
2030 fjptrC = f+j_coord_offsetC;
2031 fjptrD = f+j_coord_offsetD;
2032
2033 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2034 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2035
2036 /* Inner loop uses 351 flops */
2037 }
2038
2039 if(jidx<j_index_end)
2040 {
2041
2042 /* Get j neighbor index, and coordinate index */
2043 jnrlistA = jjnr[jidx];
2044 jnrlistB = jjnr[jidx+1];
2045 jnrlistC = jjnr[jidx+2];
2046 jnrlistD = jjnr[jidx+3];
2047 /* Sign of each element will be negative for non-real atoms.
2048 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2049 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2050 */
2051 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2052 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
2053 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
2054 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
2055 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
2056 j_coord_offsetA = DIM*jnrA;
2057 j_coord_offsetB = DIM*jnrB;
2058 j_coord_offsetC = DIM*jnrC;
2059 j_coord_offsetD = DIM*jnrD;
2060
2061 /* load j atom coordinates */
2062 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
2063 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
2064 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
2065
2066 /* Calculate displacement vector */
2067 dx11 = _mm_sub_ps(ix1,jx1);
2068 dy11 = _mm_sub_ps(iy1,jy1);
2069 dz11 = _mm_sub_ps(iz1,jz1);
2070 dx12 = _mm_sub_ps(ix1,jx2);
2071 dy12 = _mm_sub_ps(iy1,jy2);
2072 dz12 = _mm_sub_ps(iz1,jz2);
2073 dx13 = _mm_sub_ps(ix1,jx3);
2074 dy13 = _mm_sub_ps(iy1,jy3);
2075 dz13 = _mm_sub_ps(iz1,jz3);
2076 dx21 = _mm_sub_ps(ix2,jx1);
2077 dy21 = _mm_sub_ps(iy2,jy1);
2078 dz21 = _mm_sub_ps(iz2,jz1);
2079 dx22 = _mm_sub_ps(ix2,jx2);
2080 dy22 = _mm_sub_ps(iy2,jy2);
2081 dz22 = _mm_sub_ps(iz2,jz2);
2082 dx23 = _mm_sub_ps(ix2,jx3);
2083 dy23 = _mm_sub_ps(iy2,jy3);
2084 dz23 = _mm_sub_ps(iz2,jz3);
2085 dx31 = _mm_sub_ps(ix3,jx1);
2086 dy31 = _mm_sub_ps(iy3,jy1);
2087 dz31 = _mm_sub_ps(iz3,jz1);
2088 dx32 = _mm_sub_ps(ix3,jx2);
2089 dy32 = _mm_sub_ps(iy3,jy2);
2090 dz32 = _mm_sub_ps(iz3,jz2);
2091 dx33 = _mm_sub_ps(ix3,jx3);
2092 dy33 = _mm_sub_ps(iy3,jy3);
2093 dz33 = _mm_sub_ps(iz3,jz3);
2094
2095 /* Calculate squared distance and things based on it */
2096 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
2097 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
2098 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
2099 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
2100 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
2101 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
2102 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
2103 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
2104 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
2105
2106 rinv11 = sse2_invsqrt_f(rsq11);
2107 rinv12 = sse2_invsqrt_f(rsq12);
2108 rinv13 = sse2_invsqrt_f(rsq13);
2109 rinv21 = sse2_invsqrt_f(rsq21);
2110 rinv22 = sse2_invsqrt_f(rsq22);
2111 rinv23 = sse2_invsqrt_f(rsq23);
2112 rinv31 = sse2_invsqrt_f(rsq31);
2113 rinv32 = sse2_invsqrt_f(rsq32);
2114 rinv33 = sse2_invsqrt_f(rsq33);
2115
2116 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
2117 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
2118 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
2119 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
2120 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
2121 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
2122 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
2123 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
2124 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
2125
2126 fjx1 = _mm_setzero_ps();
2127 fjy1 = _mm_setzero_ps();
2128 fjz1 = _mm_setzero_ps();
2129 fjx2 = _mm_setzero_ps();
2130 fjy2 = _mm_setzero_ps();
2131 fjz2 = _mm_setzero_ps();
2132 fjx3 = _mm_setzero_ps();
2133 fjy3 = _mm_setzero_ps();
2134 fjz3 = _mm_setzero_ps();
2135
2136 /**************************
2137 * CALCULATE INTERACTIONS *
2138 **************************/
2139
2140 if (gmx_mm_any_lt(rsq11,rcutoff2))
2141 {
2142
2143 r11 = _mm_mul_ps(rsq11,rinv11);
2144 r11 = _mm_andnot_ps(dummy_mask,r11);
2145
2146 /* EWALD ELECTROSTATICS */
2147
2148 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2149 ewrt = _mm_mul_ps(r11,ewtabscale);
2150 ewitab = _mm_cvttps_epi32(ewrt);
2151 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2152 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2153 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2154 &ewtabF,&ewtabFn);
2155 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2156 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
2157
2158 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
2159
2160 fscal = felec;
2161
2162 fscal = _mm_and_ps(fscal,cutoff_mask);
2163
2164 fscal = _mm_andnot_ps(dummy_mask,fscal);
2165
2166 /* Calculate temporary vectorial force */
2167 tx = _mm_mul_ps(fscal,dx11);
2168 ty = _mm_mul_ps(fscal,dy11);
2169 tz = _mm_mul_ps(fscal,dz11);
2170
2171 /* Update vectorial force */
2172 fix1 = _mm_add_ps(fix1,tx);
2173 fiy1 = _mm_add_ps(fiy1,ty);
2174 fiz1 = _mm_add_ps(fiz1,tz);
2175
2176 fjx1 = _mm_add_ps(fjx1,tx);
2177 fjy1 = _mm_add_ps(fjy1,ty);
2178 fjz1 = _mm_add_ps(fjz1,tz);
2179
2180 }
2181
2182 /**************************
2183 * CALCULATE INTERACTIONS *
2184 **************************/
2185
2186 if (gmx_mm_any_lt(rsq12,rcutoff2))
2187 {
2188
2189 r12 = _mm_mul_ps(rsq12,rinv12);
2190 r12 = _mm_andnot_ps(dummy_mask,r12);
2191
2192 /* EWALD ELECTROSTATICS */
2193
2194 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2195 ewrt = _mm_mul_ps(r12,ewtabscale);
2196 ewitab = _mm_cvttps_epi32(ewrt);
2197 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2198 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2199 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2200 &ewtabF,&ewtabFn);
2201 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2202 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
2203
2204 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
2205
2206 fscal = felec;
2207
2208 fscal = _mm_and_ps(fscal,cutoff_mask);
2209
2210 fscal = _mm_andnot_ps(dummy_mask,fscal);
2211
2212 /* Calculate temporary vectorial force */
2213 tx = _mm_mul_ps(fscal,dx12);
2214 ty = _mm_mul_ps(fscal,dy12);
2215 tz = _mm_mul_ps(fscal,dz12);
2216
2217 /* Update vectorial force */
2218 fix1 = _mm_add_ps(fix1,tx);
2219 fiy1 = _mm_add_ps(fiy1,ty);
2220 fiz1 = _mm_add_ps(fiz1,tz);
2221
2222 fjx2 = _mm_add_ps(fjx2,tx);
2223 fjy2 = _mm_add_ps(fjy2,ty);
2224 fjz2 = _mm_add_ps(fjz2,tz);
2225
2226 }
2227
2228 /**************************
2229 * CALCULATE INTERACTIONS *
2230 **************************/
2231
2232 if (gmx_mm_any_lt(rsq13,rcutoff2))
2233 {
2234
2235 r13 = _mm_mul_ps(rsq13,rinv13);
2236 r13 = _mm_andnot_ps(dummy_mask,r13);
2237
2238 /* EWALD ELECTROSTATICS */
2239
2240 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2241 ewrt = _mm_mul_ps(r13,ewtabscale);
2242 ewitab = _mm_cvttps_epi32(ewrt);
2243 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2244 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2245 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2246 &ewtabF,&ewtabFn);
2247 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2248 felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec));
2249
2250 cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2);
2251
2252 fscal = felec;
2253
2254 fscal = _mm_and_ps(fscal,cutoff_mask);
2255
2256 fscal = _mm_andnot_ps(dummy_mask,fscal);
2257
2258 /* Calculate temporary vectorial force */
2259 tx = _mm_mul_ps(fscal,dx13);
2260 ty = _mm_mul_ps(fscal,dy13);
2261 tz = _mm_mul_ps(fscal,dz13);
2262
2263 /* Update vectorial force */
2264 fix1 = _mm_add_ps(fix1,tx);
2265 fiy1 = _mm_add_ps(fiy1,ty);
2266 fiz1 = _mm_add_ps(fiz1,tz);
2267
2268 fjx3 = _mm_add_ps(fjx3,tx);
2269 fjy3 = _mm_add_ps(fjy3,ty);
2270 fjz3 = _mm_add_ps(fjz3,tz);
2271
2272 }
2273
2274 /**************************
2275 * CALCULATE INTERACTIONS *
2276 **************************/
2277
2278 if (gmx_mm_any_lt(rsq21,rcutoff2))
2279 {
2280
2281 r21 = _mm_mul_ps(rsq21,rinv21);
2282 r21 = _mm_andnot_ps(dummy_mask,r21);
2283
2284 /* EWALD ELECTROSTATICS */
2285
2286 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2287 ewrt = _mm_mul_ps(r21,ewtabscale);
2288 ewitab = _mm_cvttps_epi32(ewrt);
2289 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2290 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2291 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2292 &ewtabF,&ewtabFn);
2293 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2294 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
2295
2296 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
2297
2298 fscal = felec;
2299
2300 fscal = _mm_and_ps(fscal,cutoff_mask);
2301
2302 fscal = _mm_andnot_ps(dummy_mask,fscal);
2303
2304 /* Calculate temporary vectorial force */
2305 tx = _mm_mul_ps(fscal,dx21);
2306 ty = _mm_mul_ps(fscal,dy21);
2307 tz = _mm_mul_ps(fscal,dz21);
2308
2309 /* Update vectorial force */
2310 fix2 = _mm_add_ps(fix2,tx);
2311 fiy2 = _mm_add_ps(fiy2,ty);
2312 fiz2 = _mm_add_ps(fiz2,tz);
2313
2314 fjx1 = _mm_add_ps(fjx1,tx);
2315 fjy1 = _mm_add_ps(fjy1,ty);
2316 fjz1 = _mm_add_ps(fjz1,tz);
2317
2318 }
2319
2320 /**************************
2321 * CALCULATE INTERACTIONS *
2322 **************************/
2323
2324 if (gmx_mm_any_lt(rsq22,rcutoff2))
2325 {
2326
2327 r22 = _mm_mul_ps(rsq22,rinv22);
2328 r22 = _mm_andnot_ps(dummy_mask,r22);
2329
2330 /* EWALD ELECTROSTATICS */
2331
2332 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2333 ewrt = _mm_mul_ps(r22,ewtabscale);
2334 ewitab = _mm_cvttps_epi32(ewrt);
2335 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2336 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2337 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2338 &ewtabF,&ewtabFn);
2339 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2340 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2341
2342 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
2343
2344 fscal = felec;
2345
2346 fscal = _mm_and_ps(fscal,cutoff_mask);
2347
2348 fscal = _mm_andnot_ps(dummy_mask,fscal);
2349
2350 /* Calculate temporary vectorial force */
2351 tx = _mm_mul_ps(fscal,dx22);
2352 ty = _mm_mul_ps(fscal,dy22);
2353 tz = _mm_mul_ps(fscal,dz22);
2354
2355 /* Update vectorial force */
2356 fix2 = _mm_add_ps(fix2,tx);
2357 fiy2 = _mm_add_ps(fiy2,ty);
2358 fiz2 = _mm_add_ps(fiz2,tz);
2359
2360 fjx2 = _mm_add_ps(fjx2,tx);
2361 fjy2 = _mm_add_ps(fjy2,ty);
2362 fjz2 = _mm_add_ps(fjz2,tz);
2363
2364 }
2365
2366 /**************************
2367 * CALCULATE INTERACTIONS *
2368 **************************/
2369
2370 if (gmx_mm_any_lt(rsq23,rcutoff2))
2371 {
2372
2373 r23 = _mm_mul_ps(rsq23,rinv23);
2374 r23 = _mm_andnot_ps(dummy_mask,r23);
2375
2376 /* EWALD ELECTROSTATICS */
2377
2378 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2379 ewrt = _mm_mul_ps(r23,ewtabscale);
2380 ewitab = _mm_cvttps_epi32(ewrt);
2381 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2382 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2383 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2384 &ewtabF,&ewtabFn);
2385 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2386 felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec));
2387
2388 cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2);
2389
2390 fscal = felec;
2391
2392 fscal = _mm_and_ps(fscal,cutoff_mask);
2393
2394 fscal = _mm_andnot_ps(dummy_mask,fscal);
2395
2396 /* Calculate temporary vectorial force */
2397 tx = _mm_mul_ps(fscal,dx23);
2398 ty = _mm_mul_ps(fscal,dy23);
2399 tz = _mm_mul_ps(fscal,dz23);
2400
2401 /* Update vectorial force */
2402 fix2 = _mm_add_ps(fix2,tx);
2403 fiy2 = _mm_add_ps(fiy2,ty);
2404 fiz2 = _mm_add_ps(fiz2,tz);
2405
2406 fjx3 = _mm_add_ps(fjx3,tx);
2407 fjy3 = _mm_add_ps(fjy3,ty);
2408 fjz3 = _mm_add_ps(fjz3,tz);
2409
2410 }
2411
2412 /**************************
2413 * CALCULATE INTERACTIONS *
2414 **************************/
2415
2416 if (gmx_mm_any_lt(rsq31,rcutoff2))
2417 {
2418
2419 r31 = _mm_mul_ps(rsq31,rinv31);
2420 r31 = _mm_andnot_ps(dummy_mask,r31);
2421
2422 /* EWALD ELECTROSTATICS */
2423
2424 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2425 ewrt = _mm_mul_ps(r31,ewtabscale);
2426 ewitab = _mm_cvttps_epi32(ewrt);
2427 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2428 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2429 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2430 &ewtabF,&ewtabFn);
2431 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2432 felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec));
2433
2434 cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2);
2435
2436 fscal = felec;
2437
2438 fscal = _mm_and_ps(fscal,cutoff_mask);
2439
2440 fscal = _mm_andnot_ps(dummy_mask,fscal);
2441
2442 /* Calculate temporary vectorial force */
2443 tx = _mm_mul_ps(fscal,dx31);
2444 ty = _mm_mul_ps(fscal,dy31);
2445 tz = _mm_mul_ps(fscal,dz31);
2446
2447 /* Update vectorial force */
2448 fix3 = _mm_add_ps(fix3,tx);
2449 fiy3 = _mm_add_ps(fiy3,ty);
2450 fiz3 = _mm_add_ps(fiz3,tz);
2451
2452 fjx1 = _mm_add_ps(fjx1,tx);
2453 fjy1 = _mm_add_ps(fjy1,ty);
2454 fjz1 = _mm_add_ps(fjz1,tz);
2455
2456 }
2457
2458 /**************************
2459 * CALCULATE INTERACTIONS *
2460 **************************/
2461
2462 if (gmx_mm_any_lt(rsq32,rcutoff2))
2463 {
2464
2465 r32 = _mm_mul_ps(rsq32,rinv32);
2466 r32 = _mm_andnot_ps(dummy_mask,r32);
2467
2468 /* EWALD ELECTROSTATICS */
2469
2470 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2471 ewrt = _mm_mul_ps(r32,ewtabscale);
2472 ewitab = _mm_cvttps_epi32(ewrt);
2473 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2474 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2475 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2476 &ewtabF,&ewtabFn);
2477 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2478 felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec));
2479
2480 cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2);
2481
2482 fscal = felec;
2483
2484 fscal = _mm_and_ps(fscal,cutoff_mask);
2485
2486 fscal = _mm_andnot_ps(dummy_mask,fscal);
2487
2488 /* Calculate temporary vectorial force */
2489 tx = _mm_mul_ps(fscal,dx32);
2490 ty = _mm_mul_ps(fscal,dy32);
2491 tz = _mm_mul_ps(fscal,dz32);
2492
2493 /* Update vectorial force */
2494 fix3 = _mm_add_ps(fix3,tx);
2495 fiy3 = _mm_add_ps(fiy3,ty);
2496 fiz3 = _mm_add_ps(fiz3,tz);
2497
2498 fjx2 = _mm_add_ps(fjx2,tx);
2499 fjy2 = _mm_add_ps(fjy2,ty);
2500 fjz2 = _mm_add_ps(fjz2,tz);
2501
2502 }
2503
2504 /**************************
2505 * CALCULATE INTERACTIONS *
2506 **************************/
2507
2508 if (gmx_mm_any_lt(rsq33,rcutoff2))
2509 {
2510
2511 r33 = _mm_mul_ps(rsq33,rinv33);
2512 r33 = _mm_andnot_ps(dummy_mask,r33);
2513
2514 /* EWALD ELECTROSTATICS */
2515
2516 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2517 ewrt = _mm_mul_ps(r33,ewtabscale);
2518 ewitab = _mm_cvttps_epi32(ewrt);
2519 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2520 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
2521 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
2522 &ewtabF,&ewtabFn);
2523 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2524 felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec));
2525
2526 cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2);
2527
2528 fscal = felec;
2529
2530 fscal = _mm_and_ps(fscal,cutoff_mask);
2531
2532 fscal = _mm_andnot_ps(dummy_mask,fscal);
2533
2534 /* Calculate temporary vectorial force */
2535 tx = _mm_mul_ps(fscal,dx33);
2536 ty = _mm_mul_ps(fscal,dy33);
2537 tz = _mm_mul_ps(fscal,dz33);
2538
2539 /* Update vectorial force */
2540 fix3 = _mm_add_ps(fix3,tx);
2541 fiy3 = _mm_add_ps(fiy3,ty);
2542 fiz3 = _mm_add_ps(fiz3,tz);
2543
2544 fjx3 = _mm_add_ps(fjx3,tx);
2545 fjy3 = _mm_add_ps(fjy3,ty);
2546 fjz3 = _mm_add_ps(fjz3,tz);
2547
2548 }
2549
2550 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2551 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2552 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2553 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2554
2555 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2556 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2557
2558 /* Inner loop uses 360 flops */
2559 }
2560
2561 /* End of innermost loop */
2562
2563 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2564 f+i_coord_offset+DIM,fshift+i_shift_offset);
2565
2566 /* Increment number of inner iterations */
2567 inneriter += j_index_end - j_index_start;
2568
2569 /* Outer loop uses 18 flops */
2570 }
2571
2572 /* Increment number of outer iterations */
2573 outeriter += nri;
2574
2575 /* Update outer/inner flops */
2576
2577 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*360);
2578 }
The ultimate molecular dynamics simulation package