2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS sse2_single kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse2_single.h"
48 #include "kernelutil_x86_sse2_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4W4_VF_sse2_single
52 * Electrostatics interaction: CubicSplineTable
53 * VdW interaction: None
54 * Geometry: Water4-Water4
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_VF_sse2_single
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
86 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
89 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
90 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
91 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
92 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
93 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
94 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
95 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
96 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
97 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
98 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
99 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
100 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
101 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
102 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
103 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
104 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
107 __m128i ifour = _mm_set1_epi32(4);
108 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
110 __m128 dummy_mask,cutoff_mask;
111 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
112 __m128 one = _mm_set1_ps(1.0);
113 __m128 two = _mm_set1_ps(2.0);
119 jindex = nlist->jindex;
121 shiftidx = nlist->shift;
123 shiftvec = fr->shift_vec[0];
124 fshift = fr->fshift[0];
125 facel = _mm_set1_ps(fr->epsfac);
126 charge = mdatoms->chargeA;
128 vftab = kernel_data->table_elec->data;
129 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
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]));
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);
150 /* Avoid stupid compiler warnings */
151 jnrA = jnrB = jnrC = jnrD = 0;
160 for(iidx=0;iidx<4*DIM;iidx++)
165 /* Start outer loop over neighborlists */
166 for(iidx=0; iidx<nri; iidx++)
168 /* Load shift vector for this list */
169 i_shift_offset = DIM*shiftidx[iidx];
171 /* Load limits for loop over neighbors */
172 j_index_start = jindex[iidx];
173 j_index_end = jindex[iidx+1];
175 /* Get outer coordinate index */
177 i_coord_offset = DIM*inr;
179 /* Load i particle coords and add shift vector */
180 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
181 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
183 fix1 = _mm_setzero_ps();
184 fiy1 = _mm_setzero_ps();
185 fiz1 = _mm_setzero_ps();
186 fix2 = _mm_setzero_ps();
187 fiy2 = _mm_setzero_ps();
188 fiz2 = _mm_setzero_ps();
189 fix3 = _mm_setzero_ps();
190 fiy3 = _mm_setzero_ps();
191 fiz3 = _mm_setzero_ps();
193 /* Reset potential sums */
194 velecsum = _mm_setzero_ps();
196 /* Start inner kernel loop */
197 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
200 /* Get j neighbor index, and coordinate index */
205 j_coord_offsetA = DIM*jnrA;
206 j_coord_offsetB = DIM*jnrB;
207 j_coord_offsetC = DIM*jnrC;
208 j_coord_offsetD = DIM*jnrD;
210 /* load j atom coordinates */
211 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
212 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
213 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
215 /* Calculate displacement vector */
216 dx11 = _mm_sub_ps(ix1,jx1);
217 dy11 = _mm_sub_ps(iy1,jy1);
218 dz11 = _mm_sub_ps(iz1,jz1);
219 dx12 = _mm_sub_ps(ix1,jx2);
220 dy12 = _mm_sub_ps(iy1,jy2);
221 dz12 = _mm_sub_ps(iz1,jz2);
222 dx13 = _mm_sub_ps(ix1,jx3);
223 dy13 = _mm_sub_ps(iy1,jy3);
224 dz13 = _mm_sub_ps(iz1,jz3);
225 dx21 = _mm_sub_ps(ix2,jx1);
226 dy21 = _mm_sub_ps(iy2,jy1);
227 dz21 = _mm_sub_ps(iz2,jz1);
228 dx22 = _mm_sub_ps(ix2,jx2);
229 dy22 = _mm_sub_ps(iy2,jy2);
230 dz22 = _mm_sub_ps(iz2,jz2);
231 dx23 = _mm_sub_ps(ix2,jx3);
232 dy23 = _mm_sub_ps(iy2,jy3);
233 dz23 = _mm_sub_ps(iz2,jz3);
234 dx31 = _mm_sub_ps(ix3,jx1);
235 dy31 = _mm_sub_ps(iy3,jy1);
236 dz31 = _mm_sub_ps(iz3,jz1);
237 dx32 = _mm_sub_ps(ix3,jx2);
238 dy32 = _mm_sub_ps(iy3,jy2);
239 dz32 = _mm_sub_ps(iz3,jz2);
240 dx33 = _mm_sub_ps(ix3,jx3);
241 dy33 = _mm_sub_ps(iy3,jy3);
242 dz33 = _mm_sub_ps(iz3,jz3);
244 /* Calculate squared distance and things based on it */
245 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
246 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
247 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
248 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
249 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
250 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
251 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
252 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
253 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
255 rinv11 = gmx_mm_invsqrt_ps(rsq11);
256 rinv12 = gmx_mm_invsqrt_ps(rsq12);
257 rinv13 = gmx_mm_invsqrt_ps(rsq13);
258 rinv21 = gmx_mm_invsqrt_ps(rsq21);
259 rinv22 = gmx_mm_invsqrt_ps(rsq22);
260 rinv23 = gmx_mm_invsqrt_ps(rsq23);
261 rinv31 = gmx_mm_invsqrt_ps(rsq31);
262 rinv32 = gmx_mm_invsqrt_ps(rsq32);
263 rinv33 = gmx_mm_invsqrt_ps(rsq33);
265 fjx1 = _mm_setzero_ps();
266 fjy1 = _mm_setzero_ps();
267 fjz1 = _mm_setzero_ps();
268 fjx2 = _mm_setzero_ps();
269 fjy2 = _mm_setzero_ps();
270 fjz2 = _mm_setzero_ps();
271 fjx3 = _mm_setzero_ps();
272 fjy3 = _mm_setzero_ps();
273 fjz3 = _mm_setzero_ps();
275 /**************************
276 * CALCULATE INTERACTIONS *
277 **************************/
279 r11 = _mm_mul_ps(rsq11,rinv11);
281 /* Calculate table index by multiplying r with table scale and truncate to integer */
282 rt = _mm_mul_ps(r11,vftabscale);
283 vfitab = _mm_cvttps_epi32(rt);
284 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
285 vfitab = _mm_slli_epi32(vfitab,2);
287 /* CUBIC SPLINE TABLE ELECTROSTATICS */
288 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
289 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
290 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
291 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
292 _MM_TRANSPOSE4_PS(Y,F,G,H);
293 Heps = _mm_mul_ps(vfeps,H);
294 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
295 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
296 velec = _mm_mul_ps(qq11,VV);
297 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
298 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
300 /* Update potential sum for this i atom from the interaction with this j atom. */
301 velecsum = _mm_add_ps(velecsum,velec);
305 /* Calculate temporary vectorial force */
306 tx = _mm_mul_ps(fscal,dx11);
307 ty = _mm_mul_ps(fscal,dy11);
308 tz = _mm_mul_ps(fscal,dz11);
310 /* Update vectorial force */
311 fix1 = _mm_add_ps(fix1,tx);
312 fiy1 = _mm_add_ps(fiy1,ty);
313 fiz1 = _mm_add_ps(fiz1,tz);
315 fjx1 = _mm_add_ps(fjx1,tx);
316 fjy1 = _mm_add_ps(fjy1,ty);
317 fjz1 = _mm_add_ps(fjz1,tz);
319 /**************************
320 * CALCULATE INTERACTIONS *
321 **************************/
323 r12 = _mm_mul_ps(rsq12,rinv12);
325 /* Calculate table index by multiplying r with table scale and truncate to integer */
326 rt = _mm_mul_ps(r12,vftabscale);
327 vfitab = _mm_cvttps_epi32(rt);
328 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
329 vfitab = _mm_slli_epi32(vfitab,2);
331 /* CUBIC SPLINE TABLE ELECTROSTATICS */
332 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
333 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
334 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
335 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
336 _MM_TRANSPOSE4_PS(Y,F,G,H);
337 Heps = _mm_mul_ps(vfeps,H);
338 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
339 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
340 velec = _mm_mul_ps(qq12,VV);
341 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
342 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
344 /* Update potential sum for this i atom from the interaction with this j atom. */
345 velecsum = _mm_add_ps(velecsum,velec);
349 /* Calculate temporary vectorial force */
350 tx = _mm_mul_ps(fscal,dx12);
351 ty = _mm_mul_ps(fscal,dy12);
352 tz = _mm_mul_ps(fscal,dz12);
354 /* Update vectorial force */
355 fix1 = _mm_add_ps(fix1,tx);
356 fiy1 = _mm_add_ps(fiy1,ty);
357 fiz1 = _mm_add_ps(fiz1,tz);
359 fjx2 = _mm_add_ps(fjx2,tx);
360 fjy2 = _mm_add_ps(fjy2,ty);
361 fjz2 = _mm_add_ps(fjz2,tz);
363 /**************************
364 * CALCULATE INTERACTIONS *
365 **************************/
367 r13 = _mm_mul_ps(rsq13,rinv13);
369 /* Calculate table index by multiplying r with table scale and truncate to integer */
370 rt = _mm_mul_ps(r13,vftabscale);
371 vfitab = _mm_cvttps_epi32(rt);
372 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
373 vfitab = _mm_slli_epi32(vfitab,2);
375 /* CUBIC SPLINE TABLE ELECTROSTATICS */
376 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
377 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
378 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
379 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
380 _MM_TRANSPOSE4_PS(Y,F,G,H);
381 Heps = _mm_mul_ps(vfeps,H);
382 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
383 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
384 velec = _mm_mul_ps(qq13,VV);
385 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
386 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
388 /* Update potential sum for this i atom from the interaction with this j atom. */
389 velecsum = _mm_add_ps(velecsum,velec);
393 /* Calculate temporary vectorial force */
394 tx = _mm_mul_ps(fscal,dx13);
395 ty = _mm_mul_ps(fscal,dy13);
396 tz = _mm_mul_ps(fscal,dz13);
398 /* Update vectorial force */
399 fix1 = _mm_add_ps(fix1,tx);
400 fiy1 = _mm_add_ps(fiy1,ty);
401 fiz1 = _mm_add_ps(fiz1,tz);
403 fjx3 = _mm_add_ps(fjx3,tx);
404 fjy3 = _mm_add_ps(fjy3,ty);
405 fjz3 = _mm_add_ps(fjz3,tz);
407 /**************************
408 * CALCULATE INTERACTIONS *
409 **************************/
411 r21 = _mm_mul_ps(rsq21,rinv21);
413 /* Calculate table index by multiplying r with table scale and truncate to integer */
414 rt = _mm_mul_ps(r21,vftabscale);
415 vfitab = _mm_cvttps_epi32(rt);
416 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
417 vfitab = _mm_slli_epi32(vfitab,2);
419 /* CUBIC SPLINE TABLE ELECTROSTATICS */
420 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
421 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
422 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
423 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
424 _MM_TRANSPOSE4_PS(Y,F,G,H);
425 Heps = _mm_mul_ps(vfeps,H);
426 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
427 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
428 velec = _mm_mul_ps(qq21,VV);
429 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
430 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
432 /* Update potential sum for this i atom from the interaction with this j atom. */
433 velecsum = _mm_add_ps(velecsum,velec);
437 /* Calculate temporary vectorial force */
438 tx = _mm_mul_ps(fscal,dx21);
439 ty = _mm_mul_ps(fscal,dy21);
440 tz = _mm_mul_ps(fscal,dz21);
442 /* Update vectorial force */
443 fix2 = _mm_add_ps(fix2,tx);
444 fiy2 = _mm_add_ps(fiy2,ty);
445 fiz2 = _mm_add_ps(fiz2,tz);
447 fjx1 = _mm_add_ps(fjx1,tx);
448 fjy1 = _mm_add_ps(fjy1,ty);
449 fjz1 = _mm_add_ps(fjz1,tz);
451 /**************************
452 * CALCULATE INTERACTIONS *
453 **************************/
455 r22 = _mm_mul_ps(rsq22,rinv22);
457 /* Calculate table index by multiplying r with table scale and truncate to integer */
458 rt = _mm_mul_ps(r22,vftabscale);
459 vfitab = _mm_cvttps_epi32(rt);
460 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
461 vfitab = _mm_slli_epi32(vfitab,2);
463 /* CUBIC SPLINE TABLE ELECTROSTATICS */
464 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
465 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
466 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
467 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
468 _MM_TRANSPOSE4_PS(Y,F,G,H);
469 Heps = _mm_mul_ps(vfeps,H);
470 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
471 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
472 velec = _mm_mul_ps(qq22,VV);
473 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
474 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
476 /* Update potential sum for this i atom from the interaction with this j atom. */
477 velecsum = _mm_add_ps(velecsum,velec);
481 /* Calculate temporary vectorial force */
482 tx = _mm_mul_ps(fscal,dx22);
483 ty = _mm_mul_ps(fscal,dy22);
484 tz = _mm_mul_ps(fscal,dz22);
486 /* Update vectorial force */
487 fix2 = _mm_add_ps(fix2,tx);
488 fiy2 = _mm_add_ps(fiy2,ty);
489 fiz2 = _mm_add_ps(fiz2,tz);
491 fjx2 = _mm_add_ps(fjx2,tx);
492 fjy2 = _mm_add_ps(fjy2,ty);
493 fjz2 = _mm_add_ps(fjz2,tz);
495 /**************************
496 * CALCULATE INTERACTIONS *
497 **************************/
499 r23 = _mm_mul_ps(rsq23,rinv23);
501 /* Calculate table index by multiplying r with table scale and truncate to integer */
502 rt = _mm_mul_ps(r23,vftabscale);
503 vfitab = _mm_cvttps_epi32(rt);
504 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
505 vfitab = _mm_slli_epi32(vfitab,2);
507 /* CUBIC SPLINE TABLE ELECTROSTATICS */
508 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
509 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
510 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
511 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
512 _MM_TRANSPOSE4_PS(Y,F,G,H);
513 Heps = _mm_mul_ps(vfeps,H);
514 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
515 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
516 velec = _mm_mul_ps(qq23,VV);
517 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
518 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
520 /* Update potential sum for this i atom from the interaction with this j atom. */
521 velecsum = _mm_add_ps(velecsum,velec);
525 /* Calculate temporary vectorial force */
526 tx = _mm_mul_ps(fscal,dx23);
527 ty = _mm_mul_ps(fscal,dy23);
528 tz = _mm_mul_ps(fscal,dz23);
530 /* Update vectorial force */
531 fix2 = _mm_add_ps(fix2,tx);
532 fiy2 = _mm_add_ps(fiy2,ty);
533 fiz2 = _mm_add_ps(fiz2,tz);
535 fjx3 = _mm_add_ps(fjx3,tx);
536 fjy3 = _mm_add_ps(fjy3,ty);
537 fjz3 = _mm_add_ps(fjz3,tz);
539 /**************************
540 * CALCULATE INTERACTIONS *
541 **************************/
543 r31 = _mm_mul_ps(rsq31,rinv31);
545 /* Calculate table index by multiplying r with table scale and truncate to integer */
546 rt = _mm_mul_ps(r31,vftabscale);
547 vfitab = _mm_cvttps_epi32(rt);
548 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
549 vfitab = _mm_slli_epi32(vfitab,2);
551 /* CUBIC SPLINE TABLE ELECTROSTATICS */
552 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
553 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
554 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
555 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
556 _MM_TRANSPOSE4_PS(Y,F,G,H);
557 Heps = _mm_mul_ps(vfeps,H);
558 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
559 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
560 velec = _mm_mul_ps(qq31,VV);
561 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
562 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
564 /* Update potential sum for this i atom from the interaction with this j atom. */
565 velecsum = _mm_add_ps(velecsum,velec);
569 /* Calculate temporary vectorial force */
570 tx = _mm_mul_ps(fscal,dx31);
571 ty = _mm_mul_ps(fscal,dy31);
572 tz = _mm_mul_ps(fscal,dz31);
574 /* Update vectorial force */
575 fix3 = _mm_add_ps(fix3,tx);
576 fiy3 = _mm_add_ps(fiy3,ty);
577 fiz3 = _mm_add_ps(fiz3,tz);
579 fjx1 = _mm_add_ps(fjx1,tx);
580 fjy1 = _mm_add_ps(fjy1,ty);
581 fjz1 = _mm_add_ps(fjz1,tz);
583 /**************************
584 * CALCULATE INTERACTIONS *
585 **************************/
587 r32 = _mm_mul_ps(rsq32,rinv32);
589 /* Calculate table index by multiplying r with table scale and truncate to integer */
590 rt = _mm_mul_ps(r32,vftabscale);
591 vfitab = _mm_cvttps_epi32(rt);
592 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
593 vfitab = _mm_slli_epi32(vfitab,2);
595 /* CUBIC SPLINE TABLE ELECTROSTATICS */
596 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
597 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
598 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
599 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
600 _MM_TRANSPOSE4_PS(Y,F,G,H);
601 Heps = _mm_mul_ps(vfeps,H);
602 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
603 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
604 velec = _mm_mul_ps(qq32,VV);
605 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
606 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
608 /* Update potential sum for this i atom from the interaction with this j atom. */
609 velecsum = _mm_add_ps(velecsum,velec);
613 /* Calculate temporary vectorial force */
614 tx = _mm_mul_ps(fscal,dx32);
615 ty = _mm_mul_ps(fscal,dy32);
616 tz = _mm_mul_ps(fscal,dz32);
618 /* Update vectorial force */
619 fix3 = _mm_add_ps(fix3,tx);
620 fiy3 = _mm_add_ps(fiy3,ty);
621 fiz3 = _mm_add_ps(fiz3,tz);
623 fjx2 = _mm_add_ps(fjx2,tx);
624 fjy2 = _mm_add_ps(fjy2,ty);
625 fjz2 = _mm_add_ps(fjz2,tz);
627 /**************************
628 * CALCULATE INTERACTIONS *
629 **************************/
631 r33 = _mm_mul_ps(rsq33,rinv33);
633 /* Calculate table index by multiplying r with table scale and truncate to integer */
634 rt = _mm_mul_ps(r33,vftabscale);
635 vfitab = _mm_cvttps_epi32(rt);
636 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
637 vfitab = _mm_slli_epi32(vfitab,2);
639 /* CUBIC SPLINE TABLE ELECTROSTATICS */
640 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
641 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
642 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
643 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
644 _MM_TRANSPOSE4_PS(Y,F,G,H);
645 Heps = _mm_mul_ps(vfeps,H);
646 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
647 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
648 velec = _mm_mul_ps(qq33,VV);
649 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
650 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
652 /* Update potential sum for this i atom from the interaction with this j atom. */
653 velecsum = _mm_add_ps(velecsum,velec);
657 /* Calculate temporary vectorial force */
658 tx = _mm_mul_ps(fscal,dx33);
659 ty = _mm_mul_ps(fscal,dy33);
660 tz = _mm_mul_ps(fscal,dz33);
662 /* Update vectorial force */
663 fix3 = _mm_add_ps(fix3,tx);
664 fiy3 = _mm_add_ps(fiy3,ty);
665 fiz3 = _mm_add_ps(fiz3,tz);
667 fjx3 = _mm_add_ps(fjx3,tx);
668 fjy3 = _mm_add_ps(fjy3,ty);
669 fjz3 = _mm_add_ps(fjz3,tz);
671 fjptrA = f+j_coord_offsetA;
672 fjptrB = f+j_coord_offsetB;
673 fjptrC = f+j_coord_offsetC;
674 fjptrD = f+j_coord_offsetD;
676 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
677 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
679 /* Inner loop uses 387 flops */
685 /* Get j neighbor index, and coordinate index */
686 jnrlistA = jjnr[jidx];
687 jnrlistB = jjnr[jidx+1];
688 jnrlistC = jjnr[jidx+2];
689 jnrlistD = jjnr[jidx+3];
690 /* Sign of each element will be negative for non-real atoms.
691 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
692 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
694 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
695 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
696 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
697 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
698 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
699 j_coord_offsetA = DIM*jnrA;
700 j_coord_offsetB = DIM*jnrB;
701 j_coord_offsetC = DIM*jnrC;
702 j_coord_offsetD = DIM*jnrD;
704 /* load j atom coordinates */
705 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
706 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
707 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
709 /* Calculate displacement vector */
710 dx11 = _mm_sub_ps(ix1,jx1);
711 dy11 = _mm_sub_ps(iy1,jy1);
712 dz11 = _mm_sub_ps(iz1,jz1);
713 dx12 = _mm_sub_ps(ix1,jx2);
714 dy12 = _mm_sub_ps(iy1,jy2);
715 dz12 = _mm_sub_ps(iz1,jz2);
716 dx13 = _mm_sub_ps(ix1,jx3);
717 dy13 = _mm_sub_ps(iy1,jy3);
718 dz13 = _mm_sub_ps(iz1,jz3);
719 dx21 = _mm_sub_ps(ix2,jx1);
720 dy21 = _mm_sub_ps(iy2,jy1);
721 dz21 = _mm_sub_ps(iz2,jz1);
722 dx22 = _mm_sub_ps(ix2,jx2);
723 dy22 = _mm_sub_ps(iy2,jy2);
724 dz22 = _mm_sub_ps(iz2,jz2);
725 dx23 = _mm_sub_ps(ix2,jx3);
726 dy23 = _mm_sub_ps(iy2,jy3);
727 dz23 = _mm_sub_ps(iz2,jz3);
728 dx31 = _mm_sub_ps(ix3,jx1);
729 dy31 = _mm_sub_ps(iy3,jy1);
730 dz31 = _mm_sub_ps(iz3,jz1);
731 dx32 = _mm_sub_ps(ix3,jx2);
732 dy32 = _mm_sub_ps(iy3,jy2);
733 dz32 = _mm_sub_ps(iz3,jz2);
734 dx33 = _mm_sub_ps(ix3,jx3);
735 dy33 = _mm_sub_ps(iy3,jy3);
736 dz33 = _mm_sub_ps(iz3,jz3);
738 /* Calculate squared distance and things based on it */
739 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
740 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
741 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
742 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
743 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
744 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
745 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
746 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
747 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
749 rinv11 = gmx_mm_invsqrt_ps(rsq11);
750 rinv12 = gmx_mm_invsqrt_ps(rsq12);
751 rinv13 = gmx_mm_invsqrt_ps(rsq13);
752 rinv21 = gmx_mm_invsqrt_ps(rsq21);
753 rinv22 = gmx_mm_invsqrt_ps(rsq22);
754 rinv23 = gmx_mm_invsqrt_ps(rsq23);
755 rinv31 = gmx_mm_invsqrt_ps(rsq31);
756 rinv32 = gmx_mm_invsqrt_ps(rsq32);
757 rinv33 = gmx_mm_invsqrt_ps(rsq33);
759 fjx1 = _mm_setzero_ps();
760 fjy1 = _mm_setzero_ps();
761 fjz1 = _mm_setzero_ps();
762 fjx2 = _mm_setzero_ps();
763 fjy2 = _mm_setzero_ps();
764 fjz2 = _mm_setzero_ps();
765 fjx3 = _mm_setzero_ps();
766 fjy3 = _mm_setzero_ps();
767 fjz3 = _mm_setzero_ps();
769 /**************************
770 * CALCULATE INTERACTIONS *
771 **************************/
773 r11 = _mm_mul_ps(rsq11,rinv11);
774 r11 = _mm_andnot_ps(dummy_mask,r11);
776 /* Calculate table index by multiplying r with table scale and truncate to integer */
777 rt = _mm_mul_ps(r11,vftabscale);
778 vfitab = _mm_cvttps_epi32(rt);
779 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
780 vfitab = _mm_slli_epi32(vfitab,2);
782 /* CUBIC SPLINE TABLE ELECTROSTATICS */
783 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
784 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
785 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
786 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
787 _MM_TRANSPOSE4_PS(Y,F,G,H);
788 Heps = _mm_mul_ps(vfeps,H);
789 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
790 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
791 velec = _mm_mul_ps(qq11,VV);
792 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
793 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
795 /* Update potential sum for this i atom from the interaction with this j atom. */
796 velec = _mm_andnot_ps(dummy_mask,velec);
797 velecsum = _mm_add_ps(velecsum,velec);
801 fscal = _mm_andnot_ps(dummy_mask,fscal);
803 /* Calculate temporary vectorial force */
804 tx = _mm_mul_ps(fscal,dx11);
805 ty = _mm_mul_ps(fscal,dy11);
806 tz = _mm_mul_ps(fscal,dz11);
808 /* Update vectorial force */
809 fix1 = _mm_add_ps(fix1,tx);
810 fiy1 = _mm_add_ps(fiy1,ty);
811 fiz1 = _mm_add_ps(fiz1,tz);
813 fjx1 = _mm_add_ps(fjx1,tx);
814 fjy1 = _mm_add_ps(fjy1,ty);
815 fjz1 = _mm_add_ps(fjz1,tz);
817 /**************************
818 * CALCULATE INTERACTIONS *
819 **************************/
821 r12 = _mm_mul_ps(rsq12,rinv12);
822 r12 = _mm_andnot_ps(dummy_mask,r12);
824 /* Calculate table index by multiplying r with table scale and truncate to integer */
825 rt = _mm_mul_ps(r12,vftabscale);
826 vfitab = _mm_cvttps_epi32(rt);
827 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
828 vfitab = _mm_slli_epi32(vfitab,2);
830 /* CUBIC SPLINE TABLE ELECTROSTATICS */
831 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
832 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
833 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
834 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
835 _MM_TRANSPOSE4_PS(Y,F,G,H);
836 Heps = _mm_mul_ps(vfeps,H);
837 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
838 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
839 velec = _mm_mul_ps(qq12,VV);
840 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
841 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
843 /* Update potential sum for this i atom from the interaction with this j atom. */
844 velec = _mm_andnot_ps(dummy_mask,velec);
845 velecsum = _mm_add_ps(velecsum,velec);
849 fscal = _mm_andnot_ps(dummy_mask,fscal);
851 /* Calculate temporary vectorial force */
852 tx = _mm_mul_ps(fscal,dx12);
853 ty = _mm_mul_ps(fscal,dy12);
854 tz = _mm_mul_ps(fscal,dz12);
856 /* Update vectorial force */
857 fix1 = _mm_add_ps(fix1,tx);
858 fiy1 = _mm_add_ps(fiy1,ty);
859 fiz1 = _mm_add_ps(fiz1,tz);
861 fjx2 = _mm_add_ps(fjx2,tx);
862 fjy2 = _mm_add_ps(fjy2,ty);
863 fjz2 = _mm_add_ps(fjz2,tz);
865 /**************************
866 * CALCULATE INTERACTIONS *
867 **************************/
869 r13 = _mm_mul_ps(rsq13,rinv13);
870 r13 = _mm_andnot_ps(dummy_mask,r13);
872 /* Calculate table index by multiplying r with table scale and truncate to integer */
873 rt = _mm_mul_ps(r13,vftabscale);
874 vfitab = _mm_cvttps_epi32(rt);
875 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
876 vfitab = _mm_slli_epi32(vfitab,2);
878 /* CUBIC SPLINE TABLE ELECTROSTATICS */
879 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
880 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
881 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
882 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
883 _MM_TRANSPOSE4_PS(Y,F,G,H);
884 Heps = _mm_mul_ps(vfeps,H);
885 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
886 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
887 velec = _mm_mul_ps(qq13,VV);
888 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
889 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
891 /* Update potential sum for this i atom from the interaction with this j atom. */
892 velec = _mm_andnot_ps(dummy_mask,velec);
893 velecsum = _mm_add_ps(velecsum,velec);
897 fscal = _mm_andnot_ps(dummy_mask,fscal);
899 /* Calculate temporary vectorial force */
900 tx = _mm_mul_ps(fscal,dx13);
901 ty = _mm_mul_ps(fscal,dy13);
902 tz = _mm_mul_ps(fscal,dz13);
904 /* Update vectorial force */
905 fix1 = _mm_add_ps(fix1,tx);
906 fiy1 = _mm_add_ps(fiy1,ty);
907 fiz1 = _mm_add_ps(fiz1,tz);
909 fjx3 = _mm_add_ps(fjx3,tx);
910 fjy3 = _mm_add_ps(fjy3,ty);
911 fjz3 = _mm_add_ps(fjz3,tz);
913 /**************************
914 * CALCULATE INTERACTIONS *
915 **************************/
917 r21 = _mm_mul_ps(rsq21,rinv21);
918 r21 = _mm_andnot_ps(dummy_mask,r21);
920 /* Calculate table index by multiplying r with table scale and truncate to integer */
921 rt = _mm_mul_ps(r21,vftabscale);
922 vfitab = _mm_cvttps_epi32(rt);
923 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
924 vfitab = _mm_slli_epi32(vfitab,2);
926 /* CUBIC SPLINE TABLE ELECTROSTATICS */
927 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
928 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
929 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
930 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
931 _MM_TRANSPOSE4_PS(Y,F,G,H);
932 Heps = _mm_mul_ps(vfeps,H);
933 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
934 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
935 velec = _mm_mul_ps(qq21,VV);
936 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
937 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
939 /* Update potential sum for this i atom from the interaction with this j atom. */
940 velec = _mm_andnot_ps(dummy_mask,velec);
941 velecsum = _mm_add_ps(velecsum,velec);
945 fscal = _mm_andnot_ps(dummy_mask,fscal);
947 /* Calculate temporary vectorial force */
948 tx = _mm_mul_ps(fscal,dx21);
949 ty = _mm_mul_ps(fscal,dy21);
950 tz = _mm_mul_ps(fscal,dz21);
952 /* Update vectorial force */
953 fix2 = _mm_add_ps(fix2,tx);
954 fiy2 = _mm_add_ps(fiy2,ty);
955 fiz2 = _mm_add_ps(fiz2,tz);
957 fjx1 = _mm_add_ps(fjx1,tx);
958 fjy1 = _mm_add_ps(fjy1,ty);
959 fjz1 = _mm_add_ps(fjz1,tz);
961 /**************************
962 * CALCULATE INTERACTIONS *
963 **************************/
965 r22 = _mm_mul_ps(rsq22,rinv22);
966 r22 = _mm_andnot_ps(dummy_mask,r22);
968 /* Calculate table index by multiplying r with table scale and truncate to integer */
969 rt = _mm_mul_ps(r22,vftabscale);
970 vfitab = _mm_cvttps_epi32(rt);
971 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
972 vfitab = _mm_slli_epi32(vfitab,2);
974 /* CUBIC SPLINE TABLE ELECTROSTATICS */
975 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
976 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
977 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
978 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
979 _MM_TRANSPOSE4_PS(Y,F,G,H);
980 Heps = _mm_mul_ps(vfeps,H);
981 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
982 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
983 velec = _mm_mul_ps(qq22,VV);
984 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
985 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
987 /* Update potential sum for this i atom from the interaction with this j atom. */
988 velec = _mm_andnot_ps(dummy_mask,velec);
989 velecsum = _mm_add_ps(velecsum,velec);
993 fscal = _mm_andnot_ps(dummy_mask,fscal);
995 /* Calculate temporary vectorial force */
996 tx = _mm_mul_ps(fscal,dx22);
997 ty = _mm_mul_ps(fscal,dy22);
998 tz = _mm_mul_ps(fscal,dz22);
1000 /* Update vectorial force */
1001 fix2 = _mm_add_ps(fix2,tx);
1002 fiy2 = _mm_add_ps(fiy2,ty);
1003 fiz2 = _mm_add_ps(fiz2,tz);
1005 fjx2 = _mm_add_ps(fjx2,tx);
1006 fjy2 = _mm_add_ps(fjy2,ty);
1007 fjz2 = _mm_add_ps(fjz2,tz);
1009 /**************************
1010 * CALCULATE INTERACTIONS *
1011 **************************/
1013 r23 = _mm_mul_ps(rsq23,rinv23);
1014 r23 = _mm_andnot_ps(dummy_mask,r23);
1016 /* Calculate table index by multiplying r with table scale and truncate to integer */
1017 rt = _mm_mul_ps(r23,vftabscale);
1018 vfitab = _mm_cvttps_epi32(rt);
1019 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1020 vfitab = _mm_slli_epi32(vfitab,2);
1022 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1023 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1024 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1025 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1026 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1027 _MM_TRANSPOSE4_PS(Y,F,G,H);
1028 Heps = _mm_mul_ps(vfeps,H);
1029 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1030 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1031 velec = _mm_mul_ps(qq23,VV);
1032 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1033 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
1035 /* Update potential sum for this i atom from the interaction with this j atom. */
1036 velec = _mm_andnot_ps(dummy_mask,velec);
1037 velecsum = _mm_add_ps(velecsum,velec);
1041 fscal = _mm_andnot_ps(dummy_mask,fscal);
1043 /* Calculate temporary vectorial force */
1044 tx = _mm_mul_ps(fscal,dx23);
1045 ty = _mm_mul_ps(fscal,dy23);
1046 tz = _mm_mul_ps(fscal,dz23);
1048 /* Update vectorial force */
1049 fix2 = _mm_add_ps(fix2,tx);
1050 fiy2 = _mm_add_ps(fiy2,ty);
1051 fiz2 = _mm_add_ps(fiz2,tz);
1053 fjx3 = _mm_add_ps(fjx3,tx);
1054 fjy3 = _mm_add_ps(fjy3,ty);
1055 fjz3 = _mm_add_ps(fjz3,tz);
1057 /**************************
1058 * CALCULATE INTERACTIONS *
1059 **************************/
1061 r31 = _mm_mul_ps(rsq31,rinv31);
1062 r31 = _mm_andnot_ps(dummy_mask,r31);
1064 /* Calculate table index by multiplying r with table scale and truncate to integer */
1065 rt = _mm_mul_ps(r31,vftabscale);
1066 vfitab = _mm_cvttps_epi32(rt);
1067 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1068 vfitab = _mm_slli_epi32(vfitab,2);
1070 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1071 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1072 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1073 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1074 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1075 _MM_TRANSPOSE4_PS(Y,F,G,H);
1076 Heps = _mm_mul_ps(vfeps,H);
1077 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1078 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1079 velec = _mm_mul_ps(qq31,VV);
1080 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1081 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
1083 /* Update potential sum for this i atom from the interaction with this j atom. */
1084 velec = _mm_andnot_ps(dummy_mask,velec);
1085 velecsum = _mm_add_ps(velecsum,velec);
1089 fscal = _mm_andnot_ps(dummy_mask,fscal);
1091 /* Calculate temporary vectorial force */
1092 tx = _mm_mul_ps(fscal,dx31);
1093 ty = _mm_mul_ps(fscal,dy31);
1094 tz = _mm_mul_ps(fscal,dz31);
1096 /* Update vectorial force */
1097 fix3 = _mm_add_ps(fix3,tx);
1098 fiy3 = _mm_add_ps(fiy3,ty);
1099 fiz3 = _mm_add_ps(fiz3,tz);
1101 fjx1 = _mm_add_ps(fjx1,tx);
1102 fjy1 = _mm_add_ps(fjy1,ty);
1103 fjz1 = _mm_add_ps(fjz1,tz);
1105 /**************************
1106 * CALCULATE INTERACTIONS *
1107 **************************/
1109 r32 = _mm_mul_ps(rsq32,rinv32);
1110 r32 = _mm_andnot_ps(dummy_mask,r32);
1112 /* Calculate table index by multiplying r with table scale and truncate to integer */
1113 rt = _mm_mul_ps(r32,vftabscale);
1114 vfitab = _mm_cvttps_epi32(rt);
1115 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1116 vfitab = _mm_slli_epi32(vfitab,2);
1118 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1119 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1120 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1121 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1122 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1123 _MM_TRANSPOSE4_PS(Y,F,G,H);
1124 Heps = _mm_mul_ps(vfeps,H);
1125 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1126 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1127 velec = _mm_mul_ps(qq32,VV);
1128 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1129 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
1131 /* Update potential sum for this i atom from the interaction with this j atom. */
1132 velec = _mm_andnot_ps(dummy_mask,velec);
1133 velecsum = _mm_add_ps(velecsum,velec);
1137 fscal = _mm_andnot_ps(dummy_mask,fscal);
1139 /* Calculate temporary vectorial force */
1140 tx = _mm_mul_ps(fscal,dx32);
1141 ty = _mm_mul_ps(fscal,dy32);
1142 tz = _mm_mul_ps(fscal,dz32);
1144 /* Update vectorial force */
1145 fix3 = _mm_add_ps(fix3,tx);
1146 fiy3 = _mm_add_ps(fiy3,ty);
1147 fiz3 = _mm_add_ps(fiz3,tz);
1149 fjx2 = _mm_add_ps(fjx2,tx);
1150 fjy2 = _mm_add_ps(fjy2,ty);
1151 fjz2 = _mm_add_ps(fjz2,tz);
1153 /**************************
1154 * CALCULATE INTERACTIONS *
1155 **************************/
1157 r33 = _mm_mul_ps(rsq33,rinv33);
1158 r33 = _mm_andnot_ps(dummy_mask,r33);
1160 /* Calculate table index by multiplying r with table scale and truncate to integer */
1161 rt = _mm_mul_ps(r33,vftabscale);
1162 vfitab = _mm_cvttps_epi32(rt);
1163 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1164 vfitab = _mm_slli_epi32(vfitab,2);
1166 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1167 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1168 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1169 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1170 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1171 _MM_TRANSPOSE4_PS(Y,F,G,H);
1172 Heps = _mm_mul_ps(vfeps,H);
1173 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1174 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1175 velec = _mm_mul_ps(qq33,VV);
1176 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1177 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
1179 /* Update potential sum for this i atom from the interaction with this j atom. */
1180 velec = _mm_andnot_ps(dummy_mask,velec);
1181 velecsum = _mm_add_ps(velecsum,velec);
1185 fscal = _mm_andnot_ps(dummy_mask,fscal);
1187 /* Calculate temporary vectorial force */
1188 tx = _mm_mul_ps(fscal,dx33);
1189 ty = _mm_mul_ps(fscal,dy33);
1190 tz = _mm_mul_ps(fscal,dz33);
1192 /* Update vectorial force */
1193 fix3 = _mm_add_ps(fix3,tx);
1194 fiy3 = _mm_add_ps(fiy3,ty);
1195 fiz3 = _mm_add_ps(fiz3,tz);
1197 fjx3 = _mm_add_ps(fjx3,tx);
1198 fjy3 = _mm_add_ps(fjy3,ty);
1199 fjz3 = _mm_add_ps(fjz3,tz);
1201 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1202 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1203 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1204 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1206 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1207 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1209 /* Inner loop uses 396 flops */
1212 /* End of innermost loop */
1214 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1215 f+i_coord_offset+DIM,fshift+i_shift_offset);
1218 /* Update potential energies */
1219 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1221 /* Increment number of inner iterations */
1222 inneriter += j_index_end - j_index_start;
1224 /* Outer loop uses 19 flops */
1227 /* Increment number of outer iterations */
1230 /* Update outer/inner flops */
1232 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*396);
1235 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_sse2_single
1236 * Electrostatics interaction: CubicSplineTable
1237 * VdW interaction: None
1238 * Geometry: Water4-Water4
1239 * Calculate force/pot: Force
1242 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_sse2_single
1243 (t_nblist * gmx_restrict nlist,
1244 rvec * gmx_restrict xx,
1245 rvec * gmx_restrict ff,
1246 t_forcerec * gmx_restrict fr,
1247 t_mdatoms * gmx_restrict mdatoms,
1248 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1249 t_nrnb * gmx_restrict nrnb)
1251 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1252 * just 0 for non-waters.
1253 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1254 * jnr indices corresponding to data put in the four positions in the SIMD register.
1256 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1257 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1258 int jnrA,jnrB,jnrC,jnrD;
1259 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1260 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1261 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1262 real rcutoff_scalar;
1263 real *shiftvec,*fshift,*x,*f;
1264 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1265 real scratch[4*DIM];
1266 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1268 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1270 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1272 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1273 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1274 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1275 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1276 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1277 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1278 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1279 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1280 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1281 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1282 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1283 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1284 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1285 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1286 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1287 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1288 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
1291 __m128i ifour = _mm_set1_epi32(4);
1292 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1294 __m128 dummy_mask,cutoff_mask;
1295 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1296 __m128 one = _mm_set1_ps(1.0);
1297 __m128 two = _mm_set1_ps(2.0);
1303 jindex = nlist->jindex;
1305 shiftidx = nlist->shift;
1307 shiftvec = fr->shift_vec[0];
1308 fshift = fr->fshift[0];
1309 facel = _mm_set1_ps(fr->epsfac);
1310 charge = mdatoms->chargeA;
1312 vftab = kernel_data->table_elec->data;
1313 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
1315 /* Setup water-specific parameters */
1316 inr = nlist->iinr[0];
1317 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1318 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1319 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1321 jq1 = _mm_set1_ps(charge[inr+1]);
1322 jq2 = _mm_set1_ps(charge[inr+2]);
1323 jq3 = _mm_set1_ps(charge[inr+3]);
1324 qq11 = _mm_mul_ps(iq1,jq1);
1325 qq12 = _mm_mul_ps(iq1,jq2);
1326 qq13 = _mm_mul_ps(iq1,jq3);
1327 qq21 = _mm_mul_ps(iq2,jq1);
1328 qq22 = _mm_mul_ps(iq2,jq2);
1329 qq23 = _mm_mul_ps(iq2,jq3);
1330 qq31 = _mm_mul_ps(iq3,jq1);
1331 qq32 = _mm_mul_ps(iq3,jq2);
1332 qq33 = _mm_mul_ps(iq3,jq3);
1334 /* Avoid stupid compiler warnings */
1335 jnrA = jnrB = jnrC = jnrD = 0;
1336 j_coord_offsetA = 0;
1337 j_coord_offsetB = 0;
1338 j_coord_offsetC = 0;
1339 j_coord_offsetD = 0;
1344 for(iidx=0;iidx<4*DIM;iidx++)
1346 scratch[iidx] = 0.0;
1349 /* Start outer loop over neighborlists */
1350 for(iidx=0; iidx<nri; iidx++)
1352 /* Load shift vector for this list */
1353 i_shift_offset = DIM*shiftidx[iidx];
1355 /* Load limits for loop over neighbors */
1356 j_index_start = jindex[iidx];
1357 j_index_end = jindex[iidx+1];
1359 /* Get outer coordinate index */
1361 i_coord_offset = DIM*inr;
1363 /* Load i particle coords and add shift vector */
1364 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1365 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1367 fix1 = _mm_setzero_ps();
1368 fiy1 = _mm_setzero_ps();
1369 fiz1 = _mm_setzero_ps();
1370 fix2 = _mm_setzero_ps();
1371 fiy2 = _mm_setzero_ps();
1372 fiz2 = _mm_setzero_ps();
1373 fix3 = _mm_setzero_ps();
1374 fiy3 = _mm_setzero_ps();
1375 fiz3 = _mm_setzero_ps();
1377 /* Start inner kernel loop */
1378 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1381 /* Get j neighbor index, and coordinate index */
1383 jnrB = jjnr[jidx+1];
1384 jnrC = jjnr[jidx+2];
1385 jnrD = jjnr[jidx+3];
1386 j_coord_offsetA = DIM*jnrA;
1387 j_coord_offsetB = DIM*jnrB;
1388 j_coord_offsetC = DIM*jnrC;
1389 j_coord_offsetD = DIM*jnrD;
1391 /* load j atom coordinates */
1392 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1393 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1394 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1396 /* Calculate displacement vector */
1397 dx11 = _mm_sub_ps(ix1,jx1);
1398 dy11 = _mm_sub_ps(iy1,jy1);
1399 dz11 = _mm_sub_ps(iz1,jz1);
1400 dx12 = _mm_sub_ps(ix1,jx2);
1401 dy12 = _mm_sub_ps(iy1,jy2);
1402 dz12 = _mm_sub_ps(iz1,jz2);
1403 dx13 = _mm_sub_ps(ix1,jx3);
1404 dy13 = _mm_sub_ps(iy1,jy3);
1405 dz13 = _mm_sub_ps(iz1,jz3);
1406 dx21 = _mm_sub_ps(ix2,jx1);
1407 dy21 = _mm_sub_ps(iy2,jy1);
1408 dz21 = _mm_sub_ps(iz2,jz1);
1409 dx22 = _mm_sub_ps(ix2,jx2);
1410 dy22 = _mm_sub_ps(iy2,jy2);
1411 dz22 = _mm_sub_ps(iz2,jz2);
1412 dx23 = _mm_sub_ps(ix2,jx3);
1413 dy23 = _mm_sub_ps(iy2,jy3);
1414 dz23 = _mm_sub_ps(iz2,jz3);
1415 dx31 = _mm_sub_ps(ix3,jx1);
1416 dy31 = _mm_sub_ps(iy3,jy1);
1417 dz31 = _mm_sub_ps(iz3,jz1);
1418 dx32 = _mm_sub_ps(ix3,jx2);
1419 dy32 = _mm_sub_ps(iy3,jy2);
1420 dz32 = _mm_sub_ps(iz3,jz2);
1421 dx33 = _mm_sub_ps(ix3,jx3);
1422 dy33 = _mm_sub_ps(iy3,jy3);
1423 dz33 = _mm_sub_ps(iz3,jz3);
1425 /* Calculate squared distance and things based on it */
1426 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1427 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1428 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1429 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1430 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1431 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1432 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1433 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1434 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1436 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1437 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1438 rinv13 = gmx_mm_invsqrt_ps(rsq13);
1439 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1440 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1441 rinv23 = gmx_mm_invsqrt_ps(rsq23);
1442 rinv31 = gmx_mm_invsqrt_ps(rsq31);
1443 rinv32 = gmx_mm_invsqrt_ps(rsq32);
1444 rinv33 = gmx_mm_invsqrt_ps(rsq33);
1446 fjx1 = _mm_setzero_ps();
1447 fjy1 = _mm_setzero_ps();
1448 fjz1 = _mm_setzero_ps();
1449 fjx2 = _mm_setzero_ps();
1450 fjy2 = _mm_setzero_ps();
1451 fjz2 = _mm_setzero_ps();
1452 fjx3 = _mm_setzero_ps();
1453 fjy3 = _mm_setzero_ps();
1454 fjz3 = _mm_setzero_ps();
1456 /**************************
1457 * CALCULATE INTERACTIONS *
1458 **************************/
1460 r11 = _mm_mul_ps(rsq11,rinv11);
1462 /* Calculate table index by multiplying r with table scale and truncate to integer */
1463 rt = _mm_mul_ps(r11,vftabscale);
1464 vfitab = _mm_cvttps_epi32(rt);
1465 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1466 vfitab = _mm_slli_epi32(vfitab,2);
1468 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1469 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1470 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1471 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1472 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1473 _MM_TRANSPOSE4_PS(Y,F,G,H);
1474 Heps = _mm_mul_ps(vfeps,H);
1475 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1476 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1477 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1481 /* Calculate temporary vectorial force */
1482 tx = _mm_mul_ps(fscal,dx11);
1483 ty = _mm_mul_ps(fscal,dy11);
1484 tz = _mm_mul_ps(fscal,dz11);
1486 /* Update vectorial force */
1487 fix1 = _mm_add_ps(fix1,tx);
1488 fiy1 = _mm_add_ps(fiy1,ty);
1489 fiz1 = _mm_add_ps(fiz1,tz);
1491 fjx1 = _mm_add_ps(fjx1,tx);
1492 fjy1 = _mm_add_ps(fjy1,ty);
1493 fjz1 = _mm_add_ps(fjz1,tz);
1495 /**************************
1496 * CALCULATE INTERACTIONS *
1497 **************************/
1499 r12 = _mm_mul_ps(rsq12,rinv12);
1501 /* Calculate table index by multiplying r with table scale and truncate to integer */
1502 rt = _mm_mul_ps(r12,vftabscale);
1503 vfitab = _mm_cvttps_epi32(rt);
1504 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1505 vfitab = _mm_slli_epi32(vfitab,2);
1507 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1508 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1509 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1510 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1511 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1512 _MM_TRANSPOSE4_PS(Y,F,G,H);
1513 Heps = _mm_mul_ps(vfeps,H);
1514 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1515 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1516 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1520 /* Calculate temporary vectorial force */
1521 tx = _mm_mul_ps(fscal,dx12);
1522 ty = _mm_mul_ps(fscal,dy12);
1523 tz = _mm_mul_ps(fscal,dz12);
1525 /* Update vectorial force */
1526 fix1 = _mm_add_ps(fix1,tx);
1527 fiy1 = _mm_add_ps(fiy1,ty);
1528 fiz1 = _mm_add_ps(fiz1,tz);
1530 fjx2 = _mm_add_ps(fjx2,tx);
1531 fjy2 = _mm_add_ps(fjy2,ty);
1532 fjz2 = _mm_add_ps(fjz2,tz);
1534 /**************************
1535 * CALCULATE INTERACTIONS *
1536 **************************/
1538 r13 = _mm_mul_ps(rsq13,rinv13);
1540 /* Calculate table index by multiplying r with table scale and truncate to integer */
1541 rt = _mm_mul_ps(r13,vftabscale);
1542 vfitab = _mm_cvttps_epi32(rt);
1543 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1544 vfitab = _mm_slli_epi32(vfitab,2);
1546 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1547 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1548 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1549 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1550 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1551 _MM_TRANSPOSE4_PS(Y,F,G,H);
1552 Heps = _mm_mul_ps(vfeps,H);
1553 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1554 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1555 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
1559 /* Calculate temporary vectorial force */
1560 tx = _mm_mul_ps(fscal,dx13);
1561 ty = _mm_mul_ps(fscal,dy13);
1562 tz = _mm_mul_ps(fscal,dz13);
1564 /* Update vectorial force */
1565 fix1 = _mm_add_ps(fix1,tx);
1566 fiy1 = _mm_add_ps(fiy1,ty);
1567 fiz1 = _mm_add_ps(fiz1,tz);
1569 fjx3 = _mm_add_ps(fjx3,tx);
1570 fjy3 = _mm_add_ps(fjy3,ty);
1571 fjz3 = _mm_add_ps(fjz3,tz);
1573 /**************************
1574 * CALCULATE INTERACTIONS *
1575 **************************/
1577 r21 = _mm_mul_ps(rsq21,rinv21);
1579 /* Calculate table index by multiplying r with table scale and truncate to integer */
1580 rt = _mm_mul_ps(r21,vftabscale);
1581 vfitab = _mm_cvttps_epi32(rt);
1582 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1583 vfitab = _mm_slli_epi32(vfitab,2);
1585 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1586 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1587 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1588 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1589 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1590 _MM_TRANSPOSE4_PS(Y,F,G,H);
1591 Heps = _mm_mul_ps(vfeps,H);
1592 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1593 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1594 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1598 /* Calculate temporary vectorial force */
1599 tx = _mm_mul_ps(fscal,dx21);
1600 ty = _mm_mul_ps(fscal,dy21);
1601 tz = _mm_mul_ps(fscal,dz21);
1603 /* Update vectorial force */
1604 fix2 = _mm_add_ps(fix2,tx);
1605 fiy2 = _mm_add_ps(fiy2,ty);
1606 fiz2 = _mm_add_ps(fiz2,tz);
1608 fjx1 = _mm_add_ps(fjx1,tx);
1609 fjy1 = _mm_add_ps(fjy1,ty);
1610 fjz1 = _mm_add_ps(fjz1,tz);
1612 /**************************
1613 * CALCULATE INTERACTIONS *
1614 **************************/
1616 r22 = _mm_mul_ps(rsq22,rinv22);
1618 /* Calculate table index by multiplying r with table scale and truncate to integer */
1619 rt = _mm_mul_ps(r22,vftabscale);
1620 vfitab = _mm_cvttps_epi32(rt);
1621 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1622 vfitab = _mm_slli_epi32(vfitab,2);
1624 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1625 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1626 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1627 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1628 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1629 _MM_TRANSPOSE4_PS(Y,F,G,H);
1630 Heps = _mm_mul_ps(vfeps,H);
1631 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1632 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1633 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1637 /* Calculate temporary vectorial force */
1638 tx = _mm_mul_ps(fscal,dx22);
1639 ty = _mm_mul_ps(fscal,dy22);
1640 tz = _mm_mul_ps(fscal,dz22);
1642 /* Update vectorial force */
1643 fix2 = _mm_add_ps(fix2,tx);
1644 fiy2 = _mm_add_ps(fiy2,ty);
1645 fiz2 = _mm_add_ps(fiz2,tz);
1647 fjx2 = _mm_add_ps(fjx2,tx);
1648 fjy2 = _mm_add_ps(fjy2,ty);
1649 fjz2 = _mm_add_ps(fjz2,tz);
1651 /**************************
1652 * CALCULATE INTERACTIONS *
1653 **************************/
1655 r23 = _mm_mul_ps(rsq23,rinv23);
1657 /* Calculate table index by multiplying r with table scale and truncate to integer */
1658 rt = _mm_mul_ps(r23,vftabscale);
1659 vfitab = _mm_cvttps_epi32(rt);
1660 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1661 vfitab = _mm_slli_epi32(vfitab,2);
1663 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1664 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1665 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1666 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1667 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1668 _MM_TRANSPOSE4_PS(Y,F,G,H);
1669 Heps = _mm_mul_ps(vfeps,H);
1670 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1671 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1672 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
1676 /* Calculate temporary vectorial force */
1677 tx = _mm_mul_ps(fscal,dx23);
1678 ty = _mm_mul_ps(fscal,dy23);
1679 tz = _mm_mul_ps(fscal,dz23);
1681 /* Update vectorial force */
1682 fix2 = _mm_add_ps(fix2,tx);
1683 fiy2 = _mm_add_ps(fiy2,ty);
1684 fiz2 = _mm_add_ps(fiz2,tz);
1686 fjx3 = _mm_add_ps(fjx3,tx);
1687 fjy3 = _mm_add_ps(fjy3,ty);
1688 fjz3 = _mm_add_ps(fjz3,tz);
1690 /**************************
1691 * CALCULATE INTERACTIONS *
1692 **************************/
1694 r31 = _mm_mul_ps(rsq31,rinv31);
1696 /* Calculate table index by multiplying r with table scale and truncate to integer */
1697 rt = _mm_mul_ps(r31,vftabscale);
1698 vfitab = _mm_cvttps_epi32(rt);
1699 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1700 vfitab = _mm_slli_epi32(vfitab,2);
1702 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1703 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1704 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1705 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1706 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1707 _MM_TRANSPOSE4_PS(Y,F,G,H);
1708 Heps = _mm_mul_ps(vfeps,H);
1709 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1710 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1711 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
1715 /* Calculate temporary vectorial force */
1716 tx = _mm_mul_ps(fscal,dx31);
1717 ty = _mm_mul_ps(fscal,dy31);
1718 tz = _mm_mul_ps(fscal,dz31);
1720 /* Update vectorial force */
1721 fix3 = _mm_add_ps(fix3,tx);
1722 fiy3 = _mm_add_ps(fiy3,ty);
1723 fiz3 = _mm_add_ps(fiz3,tz);
1725 fjx1 = _mm_add_ps(fjx1,tx);
1726 fjy1 = _mm_add_ps(fjy1,ty);
1727 fjz1 = _mm_add_ps(fjz1,tz);
1729 /**************************
1730 * CALCULATE INTERACTIONS *
1731 **************************/
1733 r32 = _mm_mul_ps(rsq32,rinv32);
1735 /* Calculate table index by multiplying r with table scale and truncate to integer */
1736 rt = _mm_mul_ps(r32,vftabscale);
1737 vfitab = _mm_cvttps_epi32(rt);
1738 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1739 vfitab = _mm_slli_epi32(vfitab,2);
1741 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1742 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1743 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1744 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1745 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1746 _MM_TRANSPOSE4_PS(Y,F,G,H);
1747 Heps = _mm_mul_ps(vfeps,H);
1748 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1749 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1750 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
1754 /* Calculate temporary vectorial force */
1755 tx = _mm_mul_ps(fscal,dx32);
1756 ty = _mm_mul_ps(fscal,dy32);
1757 tz = _mm_mul_ps(fscal,dz32);
1759 /* Update vectorial force */
1760 fix3 = _mm_add_ps(fix3,tx);
1761 fiy3 = _mm_add_ps(fiy3,ty);
1762 fiz3 = _mm_add_ps(fiz3,tz);
1764 fjx2 = _mm_add_ps(fjx2,tx);
1765 fjy2 = _mm_add_ps(fjy2,ty);
1766 fjz2 = _mm_add_ps(fjz2,tz);
1768 /**************************
1769 * CALCULATE INTERACTIONS *
1770 **************************/
1772 r33 = _mm_mul_ps(rsq33,rinv33);
1774 /* Calculate table index by multiplying r with table scale and truncate to integer */
1775 rt = _mm_mul_ps(r33,vftabscale);
1776 vfitab = _mm_cvttps_epi32(rt);
1777 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1778 vfitab = _mm_slli_epi32(vfitab,2);
1780 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1781 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1782 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1783 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1784 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1785 _MM_TRANSPOSE4_PS(Y,F,G,H);
1786 Heps = _mm_mul_ps(vfeps,H);
1787 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1788 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1789 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
1793 /* Calculate temporary vectorial force */
1794 tx = _mm_mul_ps(fscal,dx33);
1795 ty = _mm_mul_ps(fscal,dy33);
1796 tz = _mm_mul_ps(fscal,dz33);
1798 /* Update vectorial force */
1799 fix3 = _mm_add_ps(fix3,tx);
1800 fiy3 = _mm_add_ps(fiy3,ty);
1801 fiz3 = _mm_add_ps(fiz3,tz);
1803 fjx3 = _mm_add_ps(fjx3,tx);
1804 fjy3 = _mm_add_ps(fjy3,ty);
1805 fjz3 = _mm_add_ps(fjz3,tz);
1807 fjptrA = f+j_coord_offsetA;
1808 fjptrB = f+j_coord_offsetB;
1809 fjptrC = f+j_coord_offsetC;
1810 fjptrD = f+j_coord_offsetD;
1812 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1813 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1815 /* Inner loop uses 351 flops */
1818 if(jidx<j_index_end)
1821 /* Get j neighbor index, and coordinate index */
1822 jnrlistA = jjnr[jidx];
1823 jnrlistB = jjnr[jidx+1];
1824 jnrlistC = jjnr[jidx+2];
1825 jnrlistD = jjnr[jidx+3];
1826 /* Sign of each element will be negative for non-real atoms.
1827 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1828 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1830 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1831 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1832 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1833 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1834 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1835 j_coord_offsetA = DIM*jnrA;
1836 j_coord_offsetB = DIM*jnrB;
1837 j_coord_offsetC = DIM*jnrC;
1838 j_coord_offsetD = DIM*jnrD;
1840 /* load j atom coordinates */
1841 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1842 x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1843 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1845 /* Calculate displacement vector */
1846 dx11 = _mm_sub_ps(ix1,jx1);
1847 dy11 = _mm_sub_ps(iy1,jy1);
1848 dz11 = _mm_sub_ps(iz1,jz1);
1849 dx12 = _mm_sub_ps(ix1,jx2);
1850 dy12 = _mm_sub_ps(iy1,jy2);
1851 dz12 = _mm_sub_ps(iz1,jz2);
1852 dx13 = _mm_sub_ps(ix1,jx3);
1853 dy13 = _mm_sub_ps(iy1,jy3);
1854 dz13 = _mm_sub_ps(iz1,jz3);
1855 dx21 = _mm_sub_ps(ix2,jx1);
1856 dy21 = _mm_sub_ps(iy2,jy1);
1857 dz21 = _mm_sub_ps(iz2,jz1);
1858 dx22 = _mm_sub_ps(ix2,jx2);
1859 dy22 = _mm_sub_ps(iy2,jy2);
1860 dz22 = _mm_sub_ps(iz2,jz2);
1861 dx23 = _mm_sub_ps(ix2,jx3);
1862 dy23 = _mm_sub_ps(iy2,jy3);
1863 dz23 = _mm_sub_ps(iz2,jz3);
1864 dx31 = _mm_sub_ps(ix3,jx1);
1865 dy31 = _mm_sub_ps(iy3,jy1);
1866 dz31 = _mm_sub_ps(iz3,jz1);
1867 dx32 = _mm_sub_ps(ix3,jx2);
1868 dy32 = _mm_sub_ps(iy3,jy2);
1869 dz32 = _mm_sub_ps(iz3,jz2);
1870 dx33 = _mm_sub_ps(ix3,jx3);
1871 dy33 = _mm_sub_ps(iy3,jy3);
1872 dz33 = _mm_sub_ps(iz3,jz3);
1874 /* Calculate squared distance and things based on it */
1875 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1876 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1877 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1878 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1879 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1880 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1881 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1882 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1883 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1885 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1886 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1887 rinv13 = gmx_mm_invsqrt_ps(rsq13);
1888 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1889 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1890 rinv23 = gmx_mm_invsqrt_ps(rsq23);
1891 rinv31 = gmx_mm_invsqrt_ps(rsq31);
1892 rinv32 = gmx_mm_invsqrt_ps(rsq32);
1893 rinv33 = gmx_mm_invsqrt_ps(rsq33);
1895 fjx1 = _mm_setzero_ps();
1896 fjy1 = _mm_setzero_ps();
1897 fjz1 = _mm_setzero_ps();
1898 fjx2 = _mm_setzero_ps();
1899 fjy2 = _mm_setzero_ps();
1900 fjz2 = _mm_setzero_ps();
1901 fjx3 = _mm_setzero_ps();
1902 fjy3 = _mm_setzero_ps();
1903 fjz3 = _mm_setzero_ps();
1905 /**************************
1906 * CALCULATE INTERACTIONS *
1907 **************************/
1909 r11 = _mm_mul_ps(rsq11,rinv11);
1910 r11 = _mm_andnot_ps(dummy_mask,r11);
1912 /* Calculate table index by multiplying r with table scale and truncate to integer */
1913 rt = _mm_mul_ps(r11,vftabscale);
1914 vfitab = _mm_cvttps_epi32(rt);
1915 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1916 vfitab = _mm_slli_epi32(vfitab,2);
1918 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1919 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1920 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1921 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1922 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1923 _MM_TRANSPOSE4_PS(Y,F,G,H);
1924 Heps = _mm_mul_ps(vfeps,H);
1925 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1926 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1927 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1931 fscal = _mm_andnot_ps(dummy_mask,fscal);
1933 /* Calculate temporary vectorial force */
1934 tx = _mm_mul_ps(fscal,dx11);
1935 ty = _mm_mul_ps(fscal,dy11);
1936 tz = _mm_mul_ps(fscal,dz11);
1938 /* Update vectorial force */
1939 fix1 = _mm_add_ps(fix1,tx);
1940 fiy1 = _mm_add_ps(fiy1,ty);
1941 fiz1 = _mm_add_ps(fiz1,tz);
1943 fjx1 = _mm_add_ps(fjx1,tx);
1944 fjy1 = _mm_add_ps(fjy1,ty);
1945 fjz1 = _mm_add_ps(fjz1,tz);
1947 /**************************
1948 * CALCULATE INTERACTIONS *
1949 **************************/
1951 r12 = _mm_mul_ps(rsq12,rinv12);
1952 r12 = _mm_andnot_ps(dummy_mask,r12);
1954 /* Calculate table index by multiplying r with table scale and truncate to integer */
1955 rt = _mm_mul_ps(r12,vftabscale);
1956 vfitab = _mm_cvttps_epi32(rt);
1957 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1958 vfitab = _mm_slli_epi32(vfitab,2);
1960 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1961 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1962 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1963 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1964 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1965 _MM_TRANSPOSE4_PS(Y,F,G,H);
1966 Heps = _mm_mul_ps(vfeps,H);
1967 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1968 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1969 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1973 fscal = _mm_andnot_ps(dummy_mask,fscal);
1975 /* Calculate temporary vectorial force */
1976 tx = _mm_mul_ps(fscal,dx12);
1977 ty = _mm_mul_ps(fscal,dy12);
1978 tz = _mm_mul_ps(fscal,dz12);
1980 /* Update vectorial force */
1981 fix1 = _mm_add_ps(fix1,tx);
1982 fiy1 = _mm_add_ps(fiy1,ty);
1983 fiz1 = _mm_add_ps(fiz1,tz);
1985 fjx2 = _mm_add_ps(fjx2,tx);
1986 fjy2 = _mm_add_ps(fjy2,ty);
1987 fjz2 = _mm_add_ps(fjz2,tz);
1989 /**************************
1990 * CALCULATE INTERACTIONS *
1991 **************************/
1993 r13 = _mm_mul_ps(rsq13,rinv13);
1994 r13 = _mm_andnot_ps(dummy_mask,r13);
1996 /* Calculate table index by multiplying r with table scale and truncate to integer */
1997 rt = _mm_mul_ps(r13,vftabscale);
1998 vfitab = _mm_cvttps_epi32(rt);
1999 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2000 vfitab = _mm_slli_epi32(vfitab,2);
2002 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2003 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2004 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2005 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2006 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2007 _MM_TRANSPOSE4_PS(Y,F,G,H);
2008 Heps = _mm_mul_ps(vfeps,H);
2009 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2010 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2011 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq13,FF),_mm_mul_ps(vftabscale,rinv13)));
2015 fscal = _mm_andnot_ps(dummy_mask,fscal);
2017 /* Calculate temporary vectorial force */
2018 tx = _mm_mul_ps(fscal,dx13);
2019 ty = _mm_mul_ps(fscal,dy13);
2020 tz = _mm_mul_ps(fscal,dz13);
2022 /* Update vectorial force */
2023 fix1 = _mm_add_ps(fix1,tx);
2024 fiy1 = _mm_add_ps(fiy1,ty);
2025 fiz1 = _mm_add_ps(fiz1,tz);
2027 fjx3 = _mm_add_ps(fjx3,tx);
2028 fjy3 = _mm_add_ps(fjy3,ty);
2029 fjz3 = _mm_add_ps(fjz3,tz);
2031 /**************************
2032 * CALCULATE INTERACTIONS *
2033 **************************/
2035 r21 = _mm_mul_ps(rsq21,rinv21);
2036 r21 = _mm_andnot_ps(dummy_mask,r21);
2038 /* Calculate table index by multiplying r with table scale and truncate to integer */
2039 rt = _mm_mul_ps(r21,vftabscale);
2040 vfitab = _mm_cvttps_epi32(rt);
2041 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2042 vfitab = _mm_slli_epi32(vfitab,2);
2044 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2045 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2046 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2047 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2048 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2049 _MM_TRANSPOSE4_PS(Y,F,G,H);
2050 Heps = _mm_mul_ps(vfeps,H);
2051 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2052 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2053 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
2057 fscal = _mm_andnot_ps(dummy_mask,fscal);
2059 /* Calculate temporary vectorial force */
2060 tx = _mm_mul_ps(fscal,dx21);
2061 ty = _mm_mul_ps(fscal,dy21);
2062 tz = _mm_mul_ps(fscal,dz21);
2064 /* Update vectorial force */
2065 fix2 = _mm_add_ps(fix2,tx);
2066 fiy2 = _mm_add_ps(fiy2,ty);
2067 fiz2 = _mm_add_ps(fiz2,tz);
2069 fjx1 = _mm_add_ps(fjx1,tx);
2070 fjy1 = _mm_add_ps(fjy1,ty);
2071 fjz1 = _mm_add_ps(fjz1,tz);
2073 /**************************
2074 * CALCULATE INTERACTIONS *
2075 **************************/
2077 r22 = _mm_mul_ps(rsq22,rinv22);
2078 r22 = _mm_andnot_ps(dummy_mask,r22);
2080 /* Calculate table index by multiplying r with table scale and truncate to integer */
2081 rt = _mm_mul_ps(r22,vftabscale);
2082 vfitab = _mm_cvttps_epi32(rt);
2083 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2084 vfitab = _mm_slli_epi32(vfitab,2);
2086 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2087 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2088 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2089 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2090 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2091 _MM_TRANSPOSE4_PS(Y,F,G,H);
2092 Heps = _mm_mul_ps(vfeps,H);
2093 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2094 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2095 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
2099 fscal = _mm_andnot_ps(dummy_mask,fscal);
2101 /* Calculate temporary vectorial force */
2102 tx = _mm_mul_ps(fscal,dx22);
2103 ty = _mm_mul_ps(fscal,dy22);
2104 tz = _mm_mul_ps(fscal,dz22);
2106 /* Update vectorial force */
2107 fix2 = _mm_add_ps(fix2,tx);
2108 fiy2 = _mm_add_ps(fiy2,ty);
2109 fiz2 = _mm_add_ps(fiz2,tz);
2111 fjx2 = _mm_add_ps(fjx2,tx);
2112 fjy2 = _mm_add_ps(fjy2,ty);
2113 fjz2 = _mm_add_ps(fjz2,tz);
2115 /**************************
2116 * CALCULATE INTERACTIONS *
2117 **************************/
2119 r23 = _mm_mul_ps(rsq23,rinv23);
2120 r23 = _mm_andnot_ps(dummy_mask,r23);
2122 /* Calculate table index by multiplying r with table scale and truncate to integer */
2123 rt = _mm_mul_ps(r23,vftabscale);
2124 vfitab = _mm_cvttps_epi32(rt);
2125 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2126 vfitab = _mm_slli_epi32(vfitab,2);
2128 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2129 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2130 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2131 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2132 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2133 _MM_TRANSPOSE4_PS(Y,F,G,H);
2134 Heps = _mm_mul_ps(vfeps,H);
2135 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2136 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2137 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq23,FF),_mm_mul_ps(vftabscale,rinv23)));
2141 fscal = _mm_andnot_ps(dummy_mask,fscal);
2143 /* Calculate temporary vectorial force */
2144 tx = _mm_mul_ps(fscal,dx23);
2145 ty = _mm_mul_ps(fscal,dy23);
2146 tz = _mm_mul_ps(fscal,dz23);
2148 /* Update vectorial force */
2149 fix2 = _mm_add_ps(fix2,tx);
2150 fiy2 = _mm_add_ps(fiy2,ty);
2151 fiz2 = _mm_add_ps(fiz2,tz);
2153 fjx3 = _mm_add_ps(fjx3,tx);
2154 fjy3 = _mm_add_ps(fjy3,ty);
2155 fjz3 = _mm_add_ps(fjz3,tz);
2157 /**************************
2158 * CALCULATE INTERACTIONS *
2159 **************************/
2161 r31 = _mm_mul_ps(rsq31,rinv31);
2162 r31 = _mm_andnot_ps(dummy_mask,r31);
2164 /* Calculate table index by multiplying r with table scale and truncate to integer */
2165 rt = _mm_mul_ps(r31,vftabscale);
2166 vfitab = _mm_cvttps_epi32(rt);
2167 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2168 vfitab = _mm_slli_epi32(vfitab,2);
2170 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2171 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2172 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2173 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2174 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2175 _MM_TRANSPOSE4_PS(Y,F,G,H);
2176 Heps = _mm_mul_ps(vfeps,H);
2177 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2178 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2179 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq31,FF),_mm_mul_ps(vftabscale,rinv31)));
2183 fscal = _mm_andnot_ps(dummy_mask,fscal);
2185 /* Calculate temporary vectorial force */
2186 tx = _mm_mul_ps(fscal,dx31);
2187 ty = _mm_mul_ps(fscal,dy31);
2188 tz = _mm_mul_ps(fscal,dz31);
2190 /* Update vectorial force */
2191 fix3 = _mm_add_ps(fix3,tx);
2192 fiy3 = _mm_add_ps(fiy3,ty);
2193 fiz3 = _mm_add_ps(fiz3,tz);
2195 fjx1 = _mm_add_ps(fjx1,tx);
2196 fjy1 = _mm_add_ps(fjy1,ty);
2197 fjz1 = _mm_add_ps(fjz1,tz);
2199 /**************************
2200 * CALCULATE INTERACTIONS *
2201 **************************/
2203 r32 = _mm_mul_ps(rsq32,rinv32);
2204 r32 = _mm_andnot_ps(dummy_mask,r32);
2206 /* Calculate table index by multiplying r with table scale and truncate to integer */
2207 rt = _mm_mul_ps(r32,vftabscale);
2208 vfitab = _mm_cvttps_epi32(rt);
2209 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2210 vfitab = _mm_slli_epi32(vfitab,2);
2212 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2213 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2214 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2215 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2216 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2217 _MM_TRANSPOSE4_PS(Y,F,G,H);
2218 Heps = _mm_mul_ps(vfeps,H);
2219 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2220 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2221 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq32,FF),_mm_mul_ps(vftabscale,rinv32)));
2225 fscal = _mm_andnot_ps(dummy_mask,fscal);
2227 /* Calculate temporary vectorial force */
2228 tx = _mm_mul_ps(fscal,dx32);
2229 ty = _mm_mul_ps(fscal,dy32);
2230 tz = _mm_mul_ps(fscal,dz32);
2232 /* Update vectorial force */
2233 fix3 = _mm_add_ps(fix3,tx);
2234 fiy3 = _mm_add_ps(fiy3,ty);
2235 fiz3 = _mm_add_ps(fiz3,tz);
2237 fjx2 = _mm_add_ps(fjx2,tx);
2238 fjy2 = _mm_add_ps(fjy2,ty);
2239 fjz2 = _mm_add_ps(fjz2,tz);
2241 /**************************
2242 * CALCULATE INTERACTIONS *
2243 **************************/
2245 r33 = _mm_mul_ps(rsq33,rinv33);
2246 r33 = _mm_andnot_ps(dummy_mask,r33);
2248 /* Calculate table index by multiplying r with table scale and truncate to integer */
2249 rt = _mm_mul_ps(r33,vftabscale);
2250 vfitab = _mm_cvttps_epi32(rt);
2251 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2252 vfitab = _mm_slli_epi32(vfitab,2);
2254 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2255 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2256 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2257 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2258 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2259 _MM_TRANSPOSE4_PS(Y,F,G,H);
2260 Heps = _mm_mul_ps(vfeps,H);
2261 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2262 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2263 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq33,FF),_mm_mul_ps(vftabscale,rinv33)));
2267 fscal = _mm_andnot_ps(dummy_mask,fscal);
2269 /* Calculate temporary vectorial force */
2270 tx = _mm_mul_ps(fscal,dx33);
2271 ty = _mm_mul_ps(fscal,dy33);
2272 tz = _mm_mul_ps(fscal,dz33);
2274 /* Update vectorial force */
2275 fix3 = _mm_add_ps(fix3,tx);
2276 fiy3 = _mm_add_ps(fiy3,ty);
2277 fiz3 = _mm_add_ps(fiz3,tz);
2279 fjx3 = _mm_add_ps(fjx3,tx);
2280 fjy3 = _mm_add_ps(fjy3,ty);
2281 fjz3 = _mm_add_ps(fjz3,tz);
2283 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2284 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2285 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2286 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2288 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2289 fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2291 /* Inner loop uses 360 flops */
2294 /* End of innermost loop */
2296 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2297 f+i_coord_offset+DIM,fshift+i_shift_offset);
2299 /* Increment number of inner iterations */
2300 inneriter += j_index_end - j_index_start;
2302 /* Outer loop uses 18 flops */
2305 /* Increment number of outer iterations */
2308 /* Update outer/inner flops */
2310 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*360);