2 * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse4_1_single.h"
34 #include "kernelutil_x86_sse4_1_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW4W4_VF_sse4_1_single
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: CubicSplineTable
40 * Geometry: Water4-Water4
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_VdwCSTab_GeomW4W4_VF_sse4_1_single
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
63 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
65 real *shiftvec,*fshift,*x,*f;
66 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
68 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
72 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
74 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
76 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
77 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
78 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
79 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
80 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
81 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
82 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
83 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
84 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
85 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
86 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
87 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
88 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
89 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
90 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
91 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
92 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
93 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
94 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
95 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
98 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
101 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
102 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
104 __m128i ifour = _mm_set1_epi32(4);
105 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
107 __m128 dummy_mask,cutoff_mask;
108 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
109 __m128 one = _mm_set1_ps(1.0);
110 __m128 two = _mm_set1_ps(2.0);
116 jindex = nlist->jindex;
118 shiftidx = nlist->shift;
120 shiftvec = fr->shift_vec[0];
121 fshift = fr->fshift[0];
122 facel = _mm_set1_ps(fr->epsfac);
123 charge = mdatoms->chargeA;
124 nvdwtype = fr->ntype;
126 vdwtype = mdatoms->typeA;
128 vftab = kernel_data->table_vdw->data;
129 vftabscale = _mm_set1_ps(kernel_data->table_vdw->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]));
136 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
138 jq1 = _mm_set1_ps(charge[inr+1]);
139 jq2 = _mm_set1_ps(charge[inr+2]);
140 jq3 = _mm_set1_ps(charge[inr+3]);
141 vdwjidx0A = 2*vdwtype[inr+0];
142 c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
143 c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
144 qq11 = _mm_mul_ps(iq1,jq1);
145 qq12 = _mm_mul_ps(iq1,jq2);
146 qq13 = _mm_mul_ps(iq1,jq3);
147 qq21 = _mm_mul_ps(iq2,jq1);
148 qq22 = _mm_mul_ps(iq2,jq2);
149 qq23 = _mm_mul_ps(iq2,jq3);
150 qq31 = _mm_mul_ps(iq3,jq1);
151 qq32 = _mm_mul_ps(iq3,jq2);
152 qq33 = _mm_mul_ps(iq3,jq3);
154 /* Avoid stupid compiler warnings */
155 jnrA = jnrB = jnrC = jnrD = 0;
164 for(iidx=0;iidx<4*DIM;iidx++)
169 /* Start outer loop over neighborlists */
170 for(iidx=0; iidx<nri; iidx++)
172 /* Load shift vector for this list */
173 i_shift_offset = DIM*shiftidx[iidx];
175 /* Load limits for loop over neighbors */
176 j_index_start = jindex[iidx];
177 j_index_end = jindex[iidx+1];
179 /* Get outer coordinate index */
181 i_coord_offset = DIM*inr;
183 /* Load i particle coords and add shift vector */
184 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
185 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
187 fix0 = _mm_setzero_ps();
188 fiy0 = _mm_setzero_ps();
189 fiz0 = _mm_setzero_ps();
190 fix1 = _mm_setzero_ps();
191 fiy1 = _mm_setzero_ps();
192 fiz1 = _mm_setzero_ps();
193 fix2 = _mm_setzero_ps();
194 fiy2 = _mm_setzero_ps();
195 fiz2 = _mm_setzero_ps();
196 fix3 = _mm_setzero_ps();
197 fiy3 = _mm_setzero_ps();
198 fiz3 = _mm_setzero_ps();
200 /* Reset potential sums */
201 velecsum = _mm_setzero_ps();
202 vvdwsum = _mm_setzero_ps();
204 /* Start inner kernel loop */
205 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
208 /* Get j neighbor index, and coordinate index */
213 j_coord_offsetA = DIM*jnrA;
214 j_coord_offsetB = DIM*jnrB;
215 j_coord_offsetC = DIM*jnrC;
216 j_coord_offsetD = DIM*jnrD;
218 /* load j atom coordinates */
219 gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
220 x+j_coord_offsetC,x+j_coord_offsetD,
221 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
222 &jy2,&jz2,&jx3,&jy3,&jz3);
224 /* Calculate displacement vector */
225 dx00 = _mm_sub_ps(ix0,jx0);
226 dy00 = _mm_sub_ps(iy0,jy0);
227 dz00 = _mm_sub_ps(iz0,jz0);
228 dx11 = _mm_sub_ps(ix1,jx1);
229 dy11 = _mm_sub_ps(iy1,jy1);
230 dz11 = _mm_sub_ps(iz1,jz1);
231 dx12 = _mm_sub_ps(ix1,jx2);
232 dy12 = _mm_sub_ps(iy1,jy2);
233 dz12 = _mm_sub_ps(iz1,jz2);
234 dx13 = _mm_sub_ps(ix1,jx3);
235 dy13 = _mm_sub_ps(iy1,jy3);
236 dz13 = _mm_sub_ps(iz1,jz3);
237 dx21 = _mm_sub_ps(ix2,jx1);
238 dy21 = _mm_sub_ps(iy2,jy1);
239 dz21 = _mm_sub_ps(iz2,jz1);
240 dx22 = _mm_sub_ps(ix2,jx2);
241 dy22 = _mm_sub_ps(iy2,jy2);
242 dz22 = _mm_sub_ps(iz2,jz2);
243 dx23 = _mm_sub_ps(ix2,jx3);
244 dy23 = _mm_sub_ps(iy2,jy3);
245 dz23 = _mm_sub_ps(iz2,jz3);
246 dx31 = _mm_sub_ps(ix3,jx1);
247 dy31 = _mm_sub_ps(iy3,jy1);
248 dz31 = _mm_sub_ps(iz3,jz1);
249 dx32 = _mm_sub_ps(ix3,jx2);
250 dy32 = _mm_sub_ps(iy3,jy2);
251 dz32 = _mm_sub_ps(iz3,jz2);
252 dx33 = _mm_sub_ps(ix3,jx3);
253 dy33 = _mm_sub_ps(iy3,jy3);
254 dz33 = _mm_sub_ps(iz3,jz3);
256 /* Calculate squared distance and things based on it */
257 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
258 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
259 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
260 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
261 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
262 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
263 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
264 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
265 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
266 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
268 rinv00 = gmx_mm_invsqrt_ps(rsq00);
269 rinv11 = gmx_mm_invsqrt_ps(rsq11);
270 rinv12 = gmx_mm_invsqrt_ps(rsq12);
271 rinv13 = gmx_mm_invsqrt_ps(rsq13);
272 rinv21 = gmx_mm_invsqrt_ps(rsq21);
273 rinv22 = gmx_mm_invsqrt_ps(rsq22);
274 rinv23 = gmx_mm_invsqrt_ps(rsq23);
275 rinv31 = gmx_mm_invsqrt_ps(rsq31);
276 rinv32 = gmx_mm_invsqrt_ps(rsq32);
277 rinv33 = gmx_mm_invsqrt_ps(rsq33);
279 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
280 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
281 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
282 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
283 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
284 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
285 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
286 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
287 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
289 fjx0 = _mm_setzero_ps();
290 fjy0 = _mm_setzero_ps();
291 fjz0 = _mm_setzero_ps();
292 fjx1 = _mm_setzero_ps();
293 fjy1 = _mm_setzero_ps();
294 fjz1 = _mm_setzero_ps();
295 fjx2 = _mm_setzero_ps();
296 fjy2 = _mm_setzero_ps();
297 fjz2 = _mm_setzero_ps();
298 fjx3 = _mm_setzero_ps();
299 fjy3 = _mm_setzero_ps();
300 fjz3 = _mm_setzero_ps();
302 /**************************
303 * CALCULATE INTERACTIONS *
304 **************************/
306 r00 = _mm_mul_ps(rsq00,rinv00);
308 /* Calculate table index by multiplying r with table scale and truncate to integer */
309 rt = _mm_mul_ps(r00,vftabscale);
310 vfitab = _mm_cvttps_epi32(rt);
311 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
312 vfitab = _mm_slli_epi32(vfitab,3);
314 /* CUBIC SPLINE TABLE DISPERSION */
315 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
316 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
317 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
318 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
319 _MM_TRANSPOSE4_PS(Y,F,G,H);
320 Heps = _mm_mul_ps(vfeps,H);
321 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
322 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
323 vvdw6 = _mm_mul_ps(c6_00,VV);
324 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
325 fvdw6 = _mm_mul_ps(c6_00,FF);
327 /* CUBIC SPLINE TABLE REPULSION */
328 vfitab = _mm_add_epi32(vfitab,ifour);
329 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
330 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
331 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
332 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
333 _MM_TRANSPOSE4_PS(Y,F,G,H);
334 Heps = _mm_mul_ps(vfeps,H);
335 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
336 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
337 vvdw12 = _mm_mul_ps(c12_00,VV);
338 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
339 fvdw12 = _mm_mul_ps(c12_00,FF);
340 vvdw = _mm_add_ps(vvdw12,vvdw6);
341 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
343 /* Update potential sum for this i atom from the interaction with this j atom. */
344 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
348 /* Calculate temporary vectorial force */
349 tx = _mm_mul_ps(fscal,dx00);
350 ty = _mm_mul_ps(fscal,dy00);
351 tz = _mm_mul_ps(fscal,dz00);
353 /* Update vectorial force */
354 fix0 = _mm_add_ps(fix0,tx);
355 fiy0 = _mm_add_ps(fiy0,ty);
356 fiz0 = _mm_add_ps(fiz0,tz);
358 fjx0 = _mm_add_ps(fjx0,tx);
359 fjy0 = _mm_add_ps(fjy0,ty);
360 fjz0 = _mm_add_ps(fjz0,tz);
362 /**************************
363 * CALCULATE INTERACTIONS *
364 **************************/
366 /* COULOMB ELECTROSTATICS */
367 velec = _mm_mul_ps(qq11,rinv11);
368 felec = _mm_mul_ps(velec,rinvsq11);
370 /* Update potential sum for this i atom from the interaction with this j atom. */
371 velecsum = _mm_add_ps(velecsum,velec);
375 /* Calculate temporary vectorial force */
376 tx = _mm_mul_ps(fscal,dx11);
377 ty = _mm_mul_ps(fscal,dy11);
378 tz = _mm_mul_ps(fscal,dz11);
380 /* Update vectorial force */
381 fix1 = _mm_add_ps(fix1,tx);
382 fiy1 = _mm_add_ps(fiy1,ty);
383 fiz1 = _mm_add_ps(fiz1,tz);
385 fjx1 = _mm_add_ps(fjx1,tx);
386 fjy1 = _mm_add_ps(fjy1,ty);
387 fjz1 = _mm_add_ps(fjz1,tz);
389 /**************************
390 * CALCULATE INTERACTIONS *
391 **************************/
393 /* COULOMB ELECTROSTATICS */
394 velec = _mm_mul_ps(qq12,rinv12);
395 felec = _mm_mul_ps(velec,rinvsq12);
397 /* Update potential sum for this i atom from the interaction with this j atom. */
398 velecsum = _mm_add_ps(velecsum,velec);
402 /* Calculate temporary vectorial force */
403 tx = _mm_mul_ps(fscal,dx12);
404 ty = _mm_mul_ps(fscal,dy12);
405 tz = _mm_mul_ps(fscal,dz12);
407 /* Update vectorial force */
408 fix1 = _mm_add_ps(fix1,tx);
409 fiy1 = _mm_add_ps(fiy1,ty);
410 fiz1 = _mm_add_ps(fiz1,tz);
412 fjx2 = _mm_add_ps(fjx2,tx);
413 fjy2 = _mm_add_ps(fjy2,ty);
414 fjz2 = _mm_add_ps(fjz2,tz);
416 /**************************
417 * CALCULATE INTERACTIONS *
418 **************************/
420 /* COULOMB ELECTROSTATICS */
421 velec = _mm_mul_ps(qq13,rinv13);
422 felec = _mm_mul_ps(velec,rinvsq13);
424 /* Update potential sum for this i atom from the interaction with this j atom. */
425 velecsum = _mm_add_ps(velecsum,velec);
429 /* Calculate temporary vectorial force */
430 tx = _mm_mul_ps(fscal,dx13);
431 ty = _mm_mul_ps(fscal,dy13);
432 tz = _mm_mul_ps(fscal,dz13);
434 /* Update vectorial force */
435 fix1 = _mm_add_ps(fix1,tx);
436 fiy1 = _mm_add_ps(fiy1,ty);
437 fiz1 = _mm_add_ps(fiz1,tz);
439 fjx3 = _mm_add_ps(fjx3,tx);
440 fjy3 = _mm_add_ps(fjy3,ty);
441 fjz3 = _mm_add_ps(fjz3,tz);
443 /**************************
444 * CALCULATE INTERACTIONS *
445 **************************/
447 /* COULOMB ELECTROSTATICS */
448 velec = _mm_mul_ps(qq21,rinv21);
449 felec = _mm_mul_ps(velec,rinvsq21);
451 /* Update potential sum for this i atom from the interaction with this j atom. */
452 velecsum = _mm_add_ps(velecsum,velec);
456 /* Calculate temporary vectorial force */
457 tx = _mm_mul_ps(fscal,dx21);
458 ty = _mm_mul_ps(fscal,dy21);
459 tz = _mm_mul_ps(fscal,dz21);
461 /* Update vectorial force */
462 fix2 = _mm_add_ps(fix2,tx);
463 fiy2 = _mm_add_ps(fiy2,ty);
464 fiz2 = _mm_add_ps(fiz2,tz);
466 fjx1 = _mm_add_ps(fjx1,tx);
467 fjy1 = _mm_add_ps(fjy1,ty);
468 fjz1 = _mm_add_ps(fjz1,tz);
470 /**************************
471 * CALCULATE INTERACTIONS *
472 **************************/
474 /* COULOMB ELECTROSTATICS */
475 velec = _mm_mul_ps(qq22,rinv22);
476 felec = _mm_mul_ps(velec,rinvsq22);
478 /* Update potential sum for this i atom from the interaction with this j atom. */
479 velecsum = _mm_add_ps(velecsum,velec);
483 /* Calculate temporary vectorial force */
484 tx = _mm_mul_ps(fscal,dx22);
485 ty = _mm_mul_ps(fscal,dy22);
486 tz = _mm_mul_ps(fscal,dz22);
488 /* Update vectorial force */
489 fix2 = _mm_add_ps(fix2,tx);
490 fiy2 = _mm_add_ps(fiy2,ty);
491 fiz2 = _mm_add_ps(fiz2,tz);
493 fjx2 = _mm_add_ps(fjx2,tx);
494 fjy2 = _mm_add_ps(fjy2,ty);
495 fjz2 = _mm_add_ps(fjz2,tz);
497 /**************************
498 * CALCULATE INTERACTIONS *
499 **************************/
501 /* COULOMB ELECTROSTATICS */
502 velec = _mm_mul_ps(qq23,rinv23);
503 felec = _mm_mul_ps(velec,rinvsq23);
505 /* Update potential sum for this i atom from the interaction with this j atom. */
506 velecsum = _mm_add_ps(velecsum,velec);
510 /* Calculate temporary vectorial force */
511 tx = _mm_mul_ps(fscal,dx23);
512 ty = _mm_mul_ps(fscal,dy23);
513 tz = _mm_mul_ps(fscal,dz23);
515 /* Update vectorial force */
516 fix2 = _mm_add_ps(fix2,tx);
517 fiy2 = _mm_add_ps(fiy2,ty);
518 fiz2 = _mm_add_ps(fiz2,tz);
520 fjx3 = _mm_add_ps(fjx3,tx);
521 fjy3 = _mm_add_ps(fjy3,ty);
522 fjz3 = _mm_add_ps(fjz3,tz);
524 /**************************
525 * CALCULATE INTERACTIONS *
526 **************************/
528 /* COULOMB ELECTROSTATICS */
529 velec = _mm_mul_ps(qq31,rinv31);
530 felec = _mm_mul_ps(velec,rinvsq31);
532 /* Update potential sum for this i atom from the interaction with this j atom. */
533 velecsum = _mm_add_ps(velecsum,velec);
537 /* Calculate temporary vectorial force */
538 tx = _mm_mul_ps(fscal,dx31);
539 ty = _mm_mul_ps(fscal,dy31);
540 tz = _mm_mul_ps(fscal,dz31);
542 /* Update vectorial force */
543 fix3 = _mm_add_ps(fix3,tx);
544 fiy3 = _mm_add_ps(fiy3,ty);
545 fiz3 = _mm_add_ps(fiz3,tz);
547 fjx1 = _mm_add_ps(fjx1,tx);
548 fjy1 = _mm_add_ps(fjy1,ty);
549 fjz1 = _mm_add_ps(fjz1,tz);
551 /**************************
552 * CALCULATE INTERACTIONS *
553 **************************/
555 /* COULOMB ELECTROSTATICS */
556 velec = _mm_mul_ps(qq32,rinv32);
557 felec = _mm_mul_ps(velec,rinvsq32);
559 /* Update potential sum for this i atom from the interaction with this j atom. */
560 velecsum = _mm_add_ps(velecsum,velec);
564 /* Calculate temporary vectorial force */
565 tx = _mm_mul_ps(fscal,dx32);
566 ty = _mm_mul_ps(fscal,dy32);
567 tz = _mm_mul_ps(fscal,dz32);
569 /* Update vectorial force */
570 fix3 = _mm_add_ps(fix3,tx);
571 fiy3 = _mm_add_ps(fiy3,ty);
572 fiz3 = _mm_add_ps(fiz3,tz);
574 fjx2 = _mm_add_ps(fjx2,tx);
575 fjy2 = _mm_add_ps(fjy2,ty);
576 fjz2 = _mm_add_ps(fjz2,tz);
578 /**************************
579 * CALCULATE INTERACTIONS *
580 **************************/
582 /* COULOMB ELECTROSTATICS */
583 velec = _mm_mul_ps(qq33,rinv33);
584 felec = _mm_mul_ps(velec,rinvsq33);
586 /* Update potential sum for this i atom from the interaction with this j atom. */
587 velecsum = _mm_add_ps(velecsum,velec);
591 /* Calculate temporary vectorial force */
592 tx = _mm_mul_ps(fscal,dx33);
593 ty = _mm_mul_ps(fscal,dy33);
594 tz = _mm_mul_ps(fscal,dz33);
596 /* Update vectorial force */
597 fix3 = _mm_add_ps(fix3,tx);
598 fiy3 = _mm_add_ps(fiy3,ty);
599 fiz3 = _mm_add_ps(fiz3,tz);
601 fjx3 = _mm_add_ps(fjx3,tx);
602 fjy3 = _mm_add_ps(fjy3,ty);
603 fjz3 = _mm_add_ps(fjz3,tz);
605 fjptrA = f+j_coord_offsetA;
606 fjptrB = f+j_coord_offsetB;
607 fjptrC = f+j_coord_offsetC;
608 fjptrD = f+j_coord_offsetD;
610 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
611 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
612 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
614 /* Inner loop uses 311 flops */
620 /* Get j neighbor index, and coordinate index */
621 jnrlistA = jjnr[jidx];
622 jnrlistB = jjnr[jidx+1];
623 jnrlistC = jjnr[jidx+2];
624 jnrlistD = jjnr[jidx+3];
625 /* Sign of each element will be negative for non-real atoms.
626 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
627 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
629 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
630 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
631 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
632 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
633 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
634 j_coord_offsetA = DIM*jnrA;
635 j_coord_offsetB = DIM*jnrB;
636 j_coord_offsetC = DIM*jnrC;
637 j_coord_offsetD = DIM*jnrD;
639 /* load j atom coordinates */
640 gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
641 x+j_coord_offsetC,x+j_coord_offsetD,
642 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
643 &jy2,&jz2,&jx3,&jy3,&jz3);
645 /* Calculate displacement vector */
646 dx00 = _mm_sub_ps(ix0,jx0);
647 dy00 = _mm_sub_ps(iy0,jy0);
648 dz00 = _mm_sub_ps(iz0,jz0);
649 dx11 = _mm_sub_ps(ix1,jx1);
650 dy11 = _mm_sub_ps(iy1,jy1);
651 dz11 = _mm_sub_ps(iz1,jz1);
652 dx12 = _mm_sub_ps(ix1,jx2);
653 dy12 = _mm_sub_ps(iy1,jy2);
654 dz12 = _mm_sub_ps(iz1,jz2);
655 dx13 = _mm_sub_ps(ix1,jx3);
656 dy13 = _mm_sub_ps(iy1,jy3);
657 dz13 = _mm_sub_ps(iz1,jz3);
658 dx21 = _mm_sub_ps(ix2,jx1);
659 dy21 = _mm_sub_ps(iy2,jy1);
660 dz21 = _mm_sub_ps(iz2,jz1);
661 dx22 = _mm_sub_ps(ix2,jx2);
662 dy22 = _mm_sub_ps(iy2,jy2);
663 dz22 = _mm_sub_ps(iz2,jz2);
664 dx23 = _mm_sub_ps(ix2,jx3);
665 dy23 = _mm_sub_ps(iy2,jy3);
666 dz23 = _mm_sub_ps(iz2,jz3);
667 dx31 = _mm_sub_ps(ix3,jx1);
668 dy31 = _mm_sub_ps(iy3,jy1);
669 dz31 = _mm_sub_ps(iz3,jz1);
670 dx32 = _mm_sub_ps(ix3,jx2);
671 dy32 = _mm_sub_ps(iy3,jy2);
672 dz32 = _mm_sub_ps(iz3,jz2);
673 dx33 = _mm_sub_ps(ix3,jx3);
674 dy33 = _mm_sub_ps(iy3,jy3);
675 dz33 = _mm_sub_ps(iz3,jz3);
677 /* Calculate squared distance and things based on it */
678 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
679 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
680 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
681 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
682 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
683 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
684 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
685 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
686 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
687 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
689 rinv00 = gmx_mm_invsqrt_ps(rsq00);
690 rinv11 = gmx_mm_invsqrt_ps(rsq11);
691 rinv12 = gmx_mm_invsqrt_ps(rsq12);
692 rinv13 = gmx_mm_invsqrt_ps(rsq13);
693 rinv21 = gmx_mm_invsqrt_ps(rsq21);
694 rinv22 = gmx_mm_invsqrt_ps(rsq22);
695 rinv23 = gmx_mm_invsqrt_ps(rsq23);
696 rinv31 = gmx_mm_invsqrt_ps(rsq31);
697 rinv32 = gmx_mm_invsqrt_ps(rsq32);
698 rinv33 = gmx_mm_invsqrt_ps(rsq33);
700 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
701 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
702 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
703 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
704 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
705 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
706 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
707 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
708 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
710 fjx0 = _mm_setzero_ps();
711 fjy0 = _mm_setzero_ps();
712 fjz0 = _mm_setzero_ps();
713 fjx1 = _mm_setzero_ps();
714 fjy1 = _mm_setzero_ps();
715 fjz1 = _mm_setzero_ps();
716 fjx2 = _mm_setzero_ps();
717 fjy2 = _mm_setzero_ps();
718 fjz2 = _mm_setzero_ps();
719 fjx3 = _mm_setzero_ps();
720 fjy3 = _mm_setzero_ps();
721 fjz3 = _mm_setzero_ps();
723 /**************************
724 * CALCULATE INTERACTIONS *
725 **************************/
727 r00 = _mm_mul_ps(rsq00,rinv00);
728 r00 = _mm_andnot_ps(dummy_mask,r00);
730 /* Calculate table index by multiplying r with table scale and truncate to integer */
731 rt = _mm_mul_ps(r00,vftabscale);
732 vfitab = _mm_cvttps_epi32(rt);
733 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
734 vfitab = _mm_slli_epi32(vfitab,3);
736 /* CUBIC SPLINE TABLE DISPERSION */
737 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
738 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
739 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
740 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
741 _MM_TRANSPOSE4_PS(Y,F,G,H);
742 Heps = _mm_mul_ps(vfeps,H);
743 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
744 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
745 vvdw6 = _mm_mul_ps(c6_00,VV);
746 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
747 fvdw6 = _mm_mul_ps(c6_00,FF);
749 /* CUBIC SPLINE TABLE REPULSION */
750 vfitab = _mm_add_epi32(vfitab,ifour);
751 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
752 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
753 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
754 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
755 _MM_TRANSPOSE4_PS(Y,F,G,H);
756 Heps = _mm_mul_ps(vfeps,H);
757 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
758 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
759 vvdw12 = _mm_mul_ps(c12_00,VV);
760 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
761 fvdw12 = _mm_mul_ps(c12_00,FF);
762 vvdw = _mm_add_ps(vvdw12,vvdw6);
763 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
765 /* Update potential sum for this i atom from the interaction with this j atom. */
766 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
767 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
771 fscal = _mm_andnot_ps(dummy_mask,fscal);
773 /* Calculate temporary vectorial force */
774 tx = _mm_mul_ps(fscal,dx00);
775 ty = _mm_mul_ps(fscal,dy00);
776 tz = _mm_mul_ps(fscal,dz00);
778 /* Update vectorial force */
779 fix0 = _mm_add_ps(fix0,tx);
780 fiy0 = _mm_add_ps(fiy0,ty);
781 fiz0 = _mm_add_ps(fiz0,tz);
783 fjx0 = _mm_add_ps(fjx0,tx);
784 fjy0 = _mm_add_ps(fjy0,ty);
785 fjz0 = _mm_add_ps(fjz0,tz);
787 /**************************
788 * CALCULATE INTERACTIONS *
789 **************************/
791 /* COULOMB ELECTROSTATICS */
792 velec = _mm_mul_ps(qq11,rinv11);
793 felec = _mm_mul_ps(velec,rinvsq11);
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 /* COULOMB ELECTROSTATICS */
822 velec = _mm_mul_ps(qq12,rinv12);
823 felec = _mm_mul_ps(velec,rinvsq12);
825 /* Update potential sum for this i atom from the interaction with this j atom. */
826 velec = _mm_andnot_ps(dummy_mask,velec);
827 velecsum = _mm_add_ps(velecsum,velec);
831 fscal = _mm_andnot_ps(dummy_mask,fscal);
833 /* Calculate temporary vectorial force */
834 tx = _mm_mul_ps(fscal,dx12);
835 ty = _mm_mul_ps(fscal,dy12);
836 tz = _mm_mul_ps(fscal,dz12);
838 /* Update vectorial force */
839 fix1 = _mm_add_ps(fix1,tx);
840 fiy1 = _mm_add_ps(fiy1,ty);
841 fiz1 = _mm_add_ps(fiz1,tz);
843 fjx2 = _mm_add_ps(fjx2,tx);
844 fjy2 = _mm_add_ps(fjy2,ty);
845 fjz2 = _mm_add_ps(fjz2,tz);
847 /**************************
848 * CALCULATE INTERACTIONS *
849 **************************/
851 /* COULOMB ELECTROSTATICS */
852 velec = _mm_mul_ps(qq13,rinv13);
853 felec = _mm_mul_ps(velec,rinvsq13);
855 /* Update potential sum for this i atom from the interaction with this j atom. */
856 velec = _mm_andnot_ps(dummy_mask,velec);
857 velecsum = _mm_add_ps(velecsum,velec);
861 fscal = _mm_andnot_ps(dummy_mask,fscal);
863 /* Calculate temporary vectorial force */
864 tx = _mm_mul_ps(fscal,dx13);
865 ty = _mm_mul_ps(fscal,dy13);
866 tz = _mm_mul_ps(fscal,dz13);
868 /* Update vectorial force */
869 fix1 = _mm_add_ps(fix1,tx);
870 fiy1 = _mm_add_ps(fiy1,ty);
871 fiz1 = _mm_add_ps(fiz1,tz);
873 fjx3 = _mm_add_ps(fjx3,tx);
874 fjy3 = _mm_add_ps(fjy3,ty);
875 fjz3 = _mm_add_ps(fjz3,tz);
877 /**************************
878 * CALCULATE INTERACTIONS *
879 **************************/
881 /* COULOMB ELECTROSTATICS */
882 velec = _mm_mul_ps(qq21,rinv21);
883 felec = _mm_mul_ps(velec,rinvsq21);
885 /* Update potential sum for this i atom from the interaction with this j atom. */
886 velec = _mm_andnot_ps(dummy_mask,velec);
887 velecsum = _mm_add_ps(velecsum,velec);
891 fscal = _mm_andnot_ps(dummy_mask,fscal);
893 /* Calculate temporary vectorial force */
894 tx = _mm_mul_ps(fscal,dx21);
895 ty = _mm_mul_ps(fscal,dy21);
896 tz = _mm_mul_ps(fscal,dz21);
898 /* Update vectorial force */
899 fix2 = _mm_add_ps(fix2,tx);
900 fiy2 = _mm_add_ps(fiy2,ty);
901 fiz2 = _mm_add_ps(fiz2,tz);
903 fjx1 = _mm_add_ps(fjx1,tx);
904 fjy1 = _mm_add_ps(fjy1,ty);
905 fjz1 = _mm_add_ps(fjz1,tz);
907 /**************************
908 * CALCULATE INTERACTIONS *
909 **************************/
911 /* COULOMB ELECTROSTATICS */
912 velec = _mm_mul_ps(qq22,rinv22);
913 felec = _mm_mul_ps(velec,rinvsq22);
915 /* Update potential sum for this i atom from the interaction with this j atom. */
916 velec = _mm_andnot_ps(dummy_mask,velec);
917 velecsum = _mm_add_ps(velecsum,velec);
921 fscal = _mm_andnot_ps(dummy_mask,fscal);
923 /* Calculate temporary vectorial force */
924 tx = _mm_mul_ps(fscal,dx22);
925 ty = _mm_mul_ps(fscal,dy22);
926 tz = _mm_mul_ps(fscal,dz22);
928 /* Update vectorial force */
929 fix2 = _mm_add_ps(fix2,tx);
930 fiy2 = _mm_add_ps(fiy2,ty);
931 fiz2 = _mm_add_ps(fiz2,tz);
933 fjx2 = _mm_add_ps(fjx2,tx);
934 fjy2 = _mm_add_ps(fjy2,ty);
935 fjz2 = _mm_add_ps(fjz2,tz);
937 /**************************
938 * CALCULATE INTERACTIONS *
939 **************************/
941 /* COULOMB ELECTROSTATICS */
942 velec = _mm_mul_ps(qq23,rinv23);
943 felec = _mm_mul_ps(velec,rinvsq23);
945 /* Update potential sum for this i atom from the interaction with this j atom. */
946 velec = _mm_andnot_ps(dummy_mask,velec);
947 velecsum = _mm_add_ps(velecsum,velec);
951 fscal = _mm_andnot_ps(dummy_mask,fscal);
953 /* Calculate temporary vectorial force */
954 tx = _mm_mul_ps(fscal,dx23);
955 ty = _mm_mul_ps(fscal,dy23);
956 tz = _mm_mul_ps(fscal,dz23);
958 /* Update vectorial force */
959 fix2 = _mm_add_ps(fix2,tx);
960 fiy2 = _mm_add_ps(fiy2,ty);
961 fiz2 = _mm_add_ps(fiz2,tz);
963 fjx3 = _mm_add_ps(fjx3,tx);
964 fjy3 = _mm_add_ps(fjy3,ty);
965 fjz3 = _mm_add_ps(fjz3,tz);
967 /**************************
968 * CALCULATE INTERACTIONS *
969 **************************/
971 /* COULOMB ELECTROSTATICS */
972 velec = _mm_mul_ps(qq31,rinv31);
973 felec = _mm_mul_ps(velec,rinvsq31);
975 /* Update potential sum for this i atom from the interaction with this j atom. */
976 velec = _mm_andnot_ps(dummy_mask,velec);
977 velecsum = _mm_add_ps(velecsum,velec);
981 fscal = _mm_andnot_ps(dummy_mask,fscal);
983 /* Calculate temporary vectorial force */
984 tx = _mm_mul_ps(fscal,dx31);
985 ty = _mm_mul_ps(fscal,dy31);
986 tz = _mm_mul_ps(fscal,dz31);
988 /* Update vectorial force */
989 fix3 = _mm_add_ps(fix3,tx);
990 fiy3 = _mm_add_ps(fiy3,ty);
991 fiz3 = _mm_add_ps(fiz3,tz);
993 fjx1 = _mm_add_ps(fjx1,tx);
994 fjy1 = _mm_add_ps(fjy1,ty);
995 fjz1 = _mm_add_ps(fjz1,tz);
997 /**************************
998 * CALCULATE INTERACTIONS *
999 **************************/
1001 /* COULOMB ELECTROSTATICS */
1002 velec = _mm_mul_ps(qq32,rinv32);
1003 felec = _mm_mul_ps(velec,rinvsq32);
1005 /* Update potential sum for this i atom from the interaction with this j atom. */
1006 velec = _mm_andnot_ps(dummy_mask,velec);
1007 velecsum = _mm_add_ps(velecsum,velec);
1011 fscal = _mm_andnot_ps(dummy_mask,fscal);
1013 /* Calculate temporary vectorial force */
1014 tx = _mm_mul_ps(fscal,dx32);
1015 ty = _mm_mul_ps(fscal,dy32);
1016 tz = _mm_mul_ps(fscal,dz32);
1018 /* Update vectorial force */
1019 fix3 = _mm_add_ps(fix3,tx);
1020 fiy3 = _mm_add_ps(fiy3,ty);
1021 fiz3 = _mm_add_ps(fiz3,tz);
1023 fjx2 = _mm_add_ps(fjx2,tx);
1024 fjy2 = _mm_add_ps(fjy2,ty);
1025 fjz2 = _mm_add_ps(fjz2,tz);
1027 /**************************
1028 * CALCULATE INTERACTIONS *
1029 **************************/
1031 /* COULOMB ELECTROSTATICS */
1032 velec = _mm_mul_ps(qq33,rinv33);
1033 felec = _mm_mul_ps(velec,rinvsq33);
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,dx33);
1045 ty = _mm_mul_ps(fscal,dy33);
1046 tz = _mm_mul_ps(fscal,dz33);
1048 /* Update vectorial force */
1049 fix3 = _mm_add_ps(fix3,tx);
1050 fiy3 = _mm_add_ps(fiy3,ty);
1051 fiz3 = _mm_add_ps(fiz3,tz);
1053 fjx3 = _mm_add_ps(fjx3,tx);
1054 fjy3 = _mm_add_ps(fjy3,ty);
1055 fjz3 = _mm_add_ps(fjz3,tz);
1057 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1058 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1059 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1060 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1062 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1063 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1064 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1066 /* Inner loop uses 312 flops */
1069 /* End of innermost loop */
1071 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1072 f+i_coord_offset,fshift+i_shift_offset);
1075 /* Update potential energies */
1076 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1077 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1079 /* Increment number of inner iterations */
1080 inneriter += j_index_end - j_index_start;
1082 /* Outer loop uses 26 flops */
1085 /* Increment number of outer iterations */
1088 /* Update outer/inner flops */
1090 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*312);
1093 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW4W4_F_sse4_1_single
1094 * Electrostatics interaction: Coulomb
1095 * VdW interaction: CubicSplineTable
1096 * Geometry: Water4-Water4
1097 * Calculate force/pot: Force
1100 nb_kernel_ElecCoul_VdwCSTab_GeomW4W4_F_sse4_1_single
1101 (t_nblist * gmx_restrict nlist,
1102 rvec * gmx_restrict xx,
1103 rvec * gmx_restrict ff,
1104 t_forcerec * gmx_restrict fr,
1105 t_mdatoms * gmx_restrict mdatoms,
1106 nb_kernel_data_t * gmx_restrict kernel_data,
1107 t_nrnb * gmx_restrict nrnb)
1109 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1110 * just 0 for non-waters.
1111 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1112 * jnr indices corresponding to data put in the four positions in the SIMD register.
1114 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1115 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1116 int jnrA,jnrB,jnrC,jnrD;
1117 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1118 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1119 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1120 real rcutoff_scalar;
1121 real *shiftvec,*fshift,*x,*f;
1122 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1123 real scratch[4*DIM];
1124 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1126 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1128 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1130 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1132 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1133 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1134 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1135 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1136 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1137 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1138 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1139 int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1140 __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1141 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1142 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1143 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1144 __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1145 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1146 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1147 __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1148 __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1149 __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1150 __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1151 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
1154 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1157 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
1158 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
1160 __m128i ifour = _mm_set1_epi32(4);
1161 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1163 __m128 dummy_mask,cutoff_mask;
1164 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1165 __m128 one = _mm_set1_ps(1.0);
1166 __m128 two = _mm_set1_ps(2.0);
1172 jindex = nlist->jindex;
1174 shiftidx = nlist->shift;
1176 shiftvec = fr->shift_vec[0];
1177 fshift = fr->fshift[0];
1178 facel = _mm_set1_ps(fr->epsfac);
1179 charge = mdatoms->chargeA;
1180 nvdwtype = fr->ntype;
1181 vdwparam = fr->nbfp;
1182 vdwtype = mdatoms->typeA;
1184 vftab = kernel_data->table_vdw->data;
1185 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
1187 /* Setup water-specific parameters */
1188 inr = nlist->iinr[0];
1189 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1190 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1191 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1192 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
1194 jq1 = _mm_set1_ps(charge[inr+1]);
1195 jq2 = _mm_set1_ps(charge[inr+2]);
1196 jq3 = _mm_set1_ps(charge[inr+3]);
1197 vdwjidx0A = 2*vdwtype[inr+0];
1198 c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1199 c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1200 qq11 = _mm_mul_ps(iq1,jq1);
1201 qq12 = _mm_mul_ps(iq1,jq2);
1202 qq13 = _mm_mul_ps(iq1,jq3);
1203 qq21 = _mm_mul_ps(iq2,jq1);
1204 qq22 = _mm_mul_ps(iq2,jq2);
1205 qq23 = _mm_mul_ps(iq2,jq3);
1206 qq31 = _mm_mul_ps(iq3,jq1);
1207 qq32 = _mm_mul_ps(iq3,jq2);
1208 qq33 = _mm_mul_ps(iq3,jq3);
1210 /* Avoid stupid compiler warnings */
1211 jnrA = jnrB = jnrC = jnrD = 0;
1212 j_coord_offsetA = 0;
1213 j_coord_offsetB = 0;
1214 j_coord_offsetC = 0;
1215 j_coord_offsetD = 0;
1220 for(iidx=0;iidx<4*DIM;iidx++)
1222 scratch[iidx] = 0.0;
1225 /* Start outer loop over neighborlists */
1226 for(iidx=0; iidx<nri; iidx++)
1228 /* Load shift vector for this list */
1229 i_shift_offset = DIM*shiftidx[iidx];
1231 /* Load limits for loop over neighbors */
1232 j_index_start = jindex[iidx];
1233 j_index_end = jindex[iidx+1];
1235 /* Get outer coordinate index */
1237 i_coord_offset = DIM*inr;
1239 /* Load i particle coords and add shift vector */
1240 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1241 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1243 fix0 = _mm_setzero_ps();
1244 fiy0 = _mm_setzero_ps();
1245 fiz0 = _mm_setzero_ps();
1246 fix1 = _mm_setzero_ps();
1247 fiy1 = _mm_setzero_ps();
1248 fiz1 = _mm_setzero_ps();
1249 fix2 = _mm_setzero_ps();
1250 fiy2 = _mm_setzero_ps();
1251 fiz2 = _mm_setzero_ps();
1252 fix3 = _mm_setzero_ps();
1253 fiy3 = _mm_setzero_ps();
1254 fiz3 = _mm_setzero_ps();
1256 /* Start inner kernel loop */
1257 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1260 /* Get j neighbor index, and coordinate index */
1262 jnrB = jjnr[jidx+1];
1263 jnrC = jjnr[jidx+2];
1264 jnrD = jjnr[jidx+3];
1265 j_coord_offsetA = DIM*jnrA;
1266 j_coord_offsetB = DIM*jnrB;
1267 j_coord_offsetC = DIM*jnrC;
1268 j_coord_offsetD = DIM*jnrD;
1270 /* load j atom coordinates */
1271 gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1272 x+j_coord_offsetC,x+j_coord_offsetD,
1273 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1274 &jy2,&jz2,&jx3,&jy3,&jz3);
1276 /* Calculate displacement vector */
1277 dx00 = _mm_sub_ps(ix0,jx0);
1278 dy00 = _mm_sub_ps(iy0,jy0);
1279 dz00 = _mm_sub_ps(iz0,jz0);
1280 dx11 = _mm_sub_ps(ix1,jx1);
1281 dy11 = _mm_sub_ps(iy1,jy1);
1282 dz11 = _mm_sub_ps(iz1,jz1);
1283 dx12 = _mm_sub_ps(ix1,jx2);
1284 dy12 = _mm_sub_ps(iy1,jy2);
1285 dz12 = _mm_sub_ps(iz1,jz2);
1286 dx13 = _mm_sub_ps(ix1,jx3);
1287 dy13 = _mm_sub_ps(iy1,jy3);
1288 dz13 = _mm_sub_ps(iz1,jz3);
1289 dx21 = _mm_sub_ps(ix2,jx1);
1290 dy21 = _mm_sub_ps(iy2,jy1);
1291 dz21 = _mm_sub_ps(iz2,jz1);
1292 dx22 = _mm_sub_ps(ix2,jx2);
1293 dy22 = _mm_sub_ps(iy2,jy2);
1294 dz22 = _mm_sub_ps(iz2,jz2);
1295 dx23 = _mm_sub_ps(ix2,jx3);
1296 dy23 = _mm_sub_ps(iy2,jy3);
1297 dz23 = _mm_sub_ps(iz2,jz3);
1298 dx31 = _mm_sub_ps(ix3,jx1);
1299 dy31 = _mm_sub_ps(iy3,jy1);
1300 dz31 = _mm_sub_ps(iz3,jz1);
1301 dx32 = _mm_sub_ps(ix3,jx2);
1302 dy32 = _mm_sub_ps(iy3,jy2);
1303 dz32 = _mm_sub_ps(iz3,jz2);
1304 dx33 = _mm_sub_ps(ix3,jx3);
1305 dy33 = _mm_sub_ps(iy3,jy3);
1306 dz33 = _mm_sub_ps(iz3,jz3);
1308 /* Calculate squared distance and things based on it */
1309 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1310 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1311 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1312 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1313 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1314 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1315 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1316 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1317 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1318 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1320 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1321 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1322 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1323 rinv13 = gmx_mm_invsqrt_ps(rsq13);
1324 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1325 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1326 rinv23 = gmx_mm_invsqrt_ps(rsq23);
1327 rinv31 = gmx_mm_invsqrt_ps(rsq31);
1328 rinv32 = gmx_mm_invsqrt_ps(rsq32);
1329 rinv33 = gmx_mm_invsqrt_ps(rsq33);
1331 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
1332 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
1333 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
1334 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
1335 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
1336 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
1337 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
1338 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
1339 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
1341 fjx0 = _mm_setzero_ps();
1342 fjy0 = _mm_setzero_ps();
1343 fjz0 = _mm_setzero_ps();
1344 fjx1 = _mm_setzero_ps();
1345 fjy1 = _mm_setzero_ps();
1346 fjz1 = _mm_setzero_ps();
1347 fjx2 = _mm_setzero_ps();
1348 fjy2 = _mm_setzero_ps();
1349 fjz2 = _mm_setzero_ps();
1350 fjx3 = _mm_setzero_ps();
1351 fjy3 = _mm_setzero_ps();
1352 fjz3 = _mm_setzero_ps();
1354 /**************************
1355 * CALCULATE INTERACTIONS *
1356 **************************/
1358 r00 = _mm_mul_ps(rsq00,rinv00);
1360 /* Calculate table index by multiplying r with table scale and truncate to integer */
1361 rt = _mm_mul_ps(r00,vftabscale);
1362 vfitab = _mm_cvttps_epi32(rt);
1363 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1364 vfitab = _mm_slli_epi32(vfitab,3);
1366 /* CUBIC SPLINE TABLE DISPERSION */
1367 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1368 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1369 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1370 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1371 _MM_TRANSPOSE4_PS(Y,F,G,H);
1372 Heps = _mm_mul_ps(vfeps,H);
1373 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1374 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1375 fvdw6 = _mm_mul_ps(c6_00,FF);
1377 /* CUBIC SPLINE TABLE REPULSION */
1378 vfitab = _mm_add_epi32(vfitab,ifour);
1379 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1380 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1381 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1382 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1383 _MM_TRANSPOSE4_PS(Y,F,G,H);
1384 Heps = _mm_mul_ps(vfeps,H);
1385 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1386 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1387 fvdw12 = _mm_mul_ps(c12_00,FF);
1388 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1392 /* Calculate temporary vectorial force */
1393 tx = _mm_mul_ps(fscal,dx00);
1394 ty = _mm_mul_ps(fscal,dy00);
1395 tz = _mm_mul_ps(fscal,dz00);
1397 /* Update vectorial force */
1398 fix0 = _mm_add_ps(fix0,tx);
1399 fiy0 = _mm_add_ps(fiy0,ty);
1400 fiz0 = _mm_add_ps(fiz0,tz);
1402 fjx0 = _mm_add_ps(fjx0,tx);
1403 fjy0 = _mm_add_ps(fjy0,ty);
1404 fjz0 = _mm_add_ps(fjz0,tz);
1406 /**************************
1407 * CALCULATE INTERACTIONS *
1408 **************************/
1410 /* COULOMB ELECTROSTATICS */
1411 velec = _mm_mul_ps(qq11,rinv11);
1412 felec = _mm_mul_ps(velec,rinvsq11);
1416 /* Calculate temporary vectorial force */
1417 tx = _mm_mul_ps(fscal,dx11);
1418 ty = _mm_mul_ps(fscal,dy11);
1419 tz = _mm_mul_ps(fscal,dz11);
1421 /* Update vectorial force */
1422 fix1 = _mm_add_ps(fix1,tx);
1423 fiy1 = _mm_add_ps(fiy1,ty);
1424 fiz1 = _mm_add_ps(fiz1,tz);
1426 fjx1 = _mm_add_ps(fjx1,tx);
1427 fjy1 = _mm_add_ps(fjy1,ty);
1428 fjz1 = _mm_add_ps(fjz1,tz);
1430 /**************************
1431 * CALCULATE INTERACTIONS *
1432 **************************/
1434 /* COULOMB ELECTROSTATICS */
1435 velec = _mm_mul_ps(qq12,rinv12);
1436 felec = _mm_mul_ps(velec,rinvsq12);
1440 /* Calculate temporary vectorial force */
1441 tx = _mm_mul_ps(fscal,dx12);
1442 ty = _mm_mul_ps(fscal,dy12);
1443 tz = _mm_mul_ps(fscal,dz12);
1445 /* Update vectorial force */
1446 fix1 = _mm_add_ps(fix1,tx);
1447 fiy1 = _mm_add_ps(fiy1,ty);
1448 fiz1 = _mm_add_ps(fiz1,tz);
1450 fjx2 = _mm_add_ps(fjx2,tx);
1451 fjy2 = _mm_add_ps(fjy2,ty);
1452 fjz2 = _mm_add_ps(fjz2,tz);
1454 /**************************
1455 * CALCULATE INTERACTIONS *
1456 **************************/
1458 /* COULOMB ELECTROSTATICS */
1459 velec = _mm_mul_ps(qq13,rinv13);
1460 felec = _mm_mul_ps(velec,rinvsq13);
1464 /* Calculate temporary vectorial force */
1465 tx = _mm_mul_ps(fscal,dx13);
1466 ty = _mm_mul_ps(fscal,dy13);
1467 tz = _mm_mul_ps(fscal,dz13);
1469 /* Update vectorial force */
1470 fix1 = _mm_add_ps(fix1,tx);
1471 fiy1 = _mm_add_ps(fiy1,ty);
1472 fiz1 = _mm_add_ps(fiz1,tz);
1474 fjx3 = _mm_add_ps(fjx3,tx);
1475 fjy3 = _mm_add_ps(fjy3,ty);
1476 fjz3 = _mm_add_ps(fjz3,tz);
1478 /**************************
1479 * CALCULATE INTERACTIONS *
1480 **************************/
1482 /* COULOMB ELECTROSTATICS */
1483 velec = _mm_mul_ps(qq21,rinv21);
1484 felec = _mm_mul_ps(velec,rinvsq21);
1488 /* Calculate temporary vectorial force */
1489 tx = _mm_mul_ps(fscal,dx21);
1490 ty = _mm_mul_ps(fscal,dy21);
1491 tz = _mm_mul_ps(fscal,dz21);
1493 /* Update vectorial force */
1494 fix2 = _mm_add_ps(fix2,tx);
1495 fiy2 = _mm_add_ps(fiy2,ty);
1496 fiz2 = _mm_add_ps(fiz2,tz);
1498 fjx1 = _mm_add_ps(fjx1,tx);
1499 fjy1 = _mm_add_ps(fjy1,ty);
1500 fjz1 = _mm_add_ps(fjz1,tz);
1502 /**************************
1503 * CALCULATE INTERACTIONS *
1504 **************************/
1506 /* COULOMB ELECTROSTATICS */
1507 velec = _mm_mul_ps(qq22,rinv22);
1508 felec = _mm_mul_ps(velec,rinvsq22);
1512 /* Calculate temporary vectorial force */
1513 tx = _mm_mul_ps(fscal,dx22);
1514 ty = _mm_mul_ps(fscal,dy22);
1515 tz = _mm_mul_ps(fscal,dz22);
1517 /* Update vectorial force */
1518 fix2 = _mm_add_ps(fix2,tx);
1519 fiy2 = _mm_add_ps(fiy2,ty);
1520 fiz2 = _mm_add_ps(fiz2,tz);
1522 fjx2 = _mm_add_ps(fjx2,tx);
1523 fjy2 = _mm_add_ps(fjy2,ty);
1524 fjz2 = _mm_add_ps(fjz2,tz);
1526 /**************************
1527 * CALCULATE INTERACTIONS *
1528 **************************/
1530 /* COULOMB ELECTROSTATICS */
1531 velec = _mm_mul_ps(qq23,rinv23);
1532 felec = _mm_mul_ps(velec,rinvsq23);
1536 /* Calculate temporary vectorial force */
1537 tx = _mm_mul_ps(fscal,dx23);
1538 ty = _mm_mul_ps(fscal,dy23);
1539 tz = _mm_mul_ps(fscal,dz23);
1541 /* Update vectorial force */
1542 fix2 = _mm_add_ps(fix2,tx);
1543 fiy2 = _mm_add_ps(fiy2,ty);
1544 fiz2 = _mm_add_ps(fiz2,tz);
1546 fjx3 = _mm_add_ps(fjx3,tx);
1547 fjy3 = _mm_add_ps(fjy3,ty);
1548 fjz3 = _mm_add_ps(fjz3,tz);
1550 /**************************
1551 * CALCULATE INTERACTIONS *
1552 **************************/
1554 /* COULOMB ELECTROSTATICS */
1555 velec = _mm_mul_ps(qq31,rinv31);
1556 felec = _mm_mul_ps(velec,rinvsq31);
1560 /* Calculate temporary vectorial force */
1561 tx = _mm_mul_ps(fscal,dx31);
1562 ty = _mm_mul_ps(fscal,dy31);
1563 tz = _mm_mul_ps(fscal,dz31);
1565 /* Update vectorial force */
1566 fix3 = _mm_add_ps(fix3,tx);
1567 fiy3 = _mm_add_ps(fiy3,ty);
1568 fiz3 = _mm_add_ps(fiz3,tz);
1570 fjx1 = _mm_add_ps(fjx1,tx);
1571 fjy1 = _mm_add_ps(fjy1,ty);
1572 fjz1 = _mm_add_ps(fjz1,tz);
1574 /**************************
1575 * CALCULATE INTERACTIONS *
1576 **************************/
1578 /* COULOMB ELECTROSTATICS */
1579 velec = _mm_mul_ps(qq32,rinv32);
1580 felec = _mm_mul_ps(velec,rinvsq32);
1584 /* Calculate temporary vectorial force */
1585 tx = _mm_mul_ps(fscal,dx32);
1586 ty = _mm_mul_ps(fscal,dy32);
1587 tz = _mm_mul_ps(fscal,dz32);
1589 /* Update vectorial force */
1590 fix3 = _mm_add_ps(fix3,tx);
1591 fiy3 = _mm_add_ps(fiy3,ty);
1592 fiz3 = _mm_add_ps(fiz3,tz);
1594 fjx2 = _mm_add_ps(fjx2,tx);
1595 fjy2 = _mm_add_ps(fjy2,ty);
1596 fjz2 = _mm_add_ps(fjz2,tz);
1598 /**************************
1599 * CALCULATE INTERACTIONS *
1600 **************************/
1602 /* COULOMB ELECTROSTATICS */
1603 velec = _mm_mul_ps(qq33,rinv33);
1604 felec = _mm_mul_ps(velec,rinvsq33);
1608 /* Calculate temporary vectorial force */
1609 tx = _mm_mul_ps(fscal,dx33);
1610 ty = _mm_mul_ps(fscal,dy33);
1611 tz = _mm_mul_ps(fscal,dz33);
1613 /* Update vectorial force */
1614 fix3 = _mm_add_ps(fix3,tx);
1615 fiy3 = _mm_add_ps(fiy3,ty);
1616 fiz3 = _mm_add_ps(fiz3,tz);
1618 fjx3 = _mm_add_ps(fjx3,tx);
1619 fjy3 = _mm_add_ps(fjy3,ty);
1620 fjz3 = _mm_add_ps(fjz3,tz);
1622 fjptrA = f+j_coord_offsetA;
1623 fjptrB = f+j_coord_offsetB;
1624 fjptrC = f+j_coord_offsetC;
1625 fjptrD = f+j_coord_offsetD;
1627 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1628 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1629 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1631 /* Inner loop uses 294 flops */
1634 if(jidx<j_index_end)
1637 /* Get j neighbor index, and coordinate index */
1638 jnrlistA = jjnr[jidx];
1639 jnrlistB = jjnr[jidx+1];
1640 jnrlistC = jjnr[jidx+2];
1641 jnrlistD = jjnr[jidx+3];
1642 /* Sign of each element will be negative for non-real atoms.
1643 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1644 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1646 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1647 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1648 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1649 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1650 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1651 j_coord_offsetA = DIM*jnrA;
1652 j_coord_offsetB = DIM*jnrB;
1653 j_coord_offsetC = DIM*jnrC;
1654 j_coord_offsetD = DIM*jnrD;
1656 /* load j atom coordinates */
1657 gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1658 x+j_coord_offsetC,x+j_coord_offsetD,
1659 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1660 &jy2,&jz2,&jx3,&jy3,&jz3);
1662 /* Calculate displacement vector */
1663 dx00 = _mm_sub_ps(ix0,jx0);
1664 dy00 = _mm_sub_ps(iy0,jy0);
1665 dz00 = _mm_sub_ps(iz0,jz0);
1666 dx11 = _mm_sub_ps(ix1,jx1);
1667 dy11 = _mm_sub_ps(iy1,jy1);
1668 dz11 = _mm_sub_ps(iz1,jz1);
1669 dx12 = _mm_sub_ps(ix1,jx2);
1670 dy12 = _mm_sub_ps(iy1,jy2);
1671 dz12 = _mm_sub_ps(iz1,jz2);
1672 dx13 = _mm_sub_ps(ix1,jx3);
1673 dy13 = _mm_sub_ps(iy1,jy3);
1674 dz13 = _mm_sub_ps(iz1,jz3);
1675 dx21 = _mm_sub_ps(ix2,jx1);
1676 dy21 = _mm_sub_ps(iy2,jy1);
1677 dz21 = _mm_sub_ps(iz2,jz1);
1678 dx22 = _mm_sub_ps(ix2,jx2);
1679 dy22 = _mm_sub_ps(iy2,jy2);
1680 dz22 = _mm_sub_ps(iz2,jz2);
1681 dx23 = _mm_sub_ps(ix2,jx3);
1682 dy23 = _mm_sub_ps(iy2,jy3);
1683 dz23 = _mm_sub_ps(iz2,jz3);
1684 dx31 = _mm_sub_ps(ix3,jx1);
1685 dy31 = _mm_sub_ps(iy3,jy1);
1686 dz31 = _mm_sub_ps(iz3,jz1);
1687 dx32 = _mm_sub_ps(ix3,jx2);
1688 dy32 = _mm_sub_ps(iy3,jy2);
1689 dz32 = _mm_sub_ps(iz3,jz2);
1690 dx33 = _mm_sub_ps(ix3,jx3);
1691 dy33 = _mm_sub_ps(iy3,jy3);
1692 dz33 = _mm_sub_ps(iz3,jz3);
1694 /* Calculate squared distance and things based on it */
1695 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1696 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1697 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1698 rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1699 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1700 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1701 rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1702 rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1703 rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1704 rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1706 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1707 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1708 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1709 rinv13 = gmx_mm_invsqrt_ps(rsq13);
1710 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1711 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1712 rinv23 = gmx_mm_invsqrt_ps(rsq23);
1713 rinv31 = gmx_mm_invsqrt_ps(rsq31);
1714 rinv32 = gmx_mm_invsqrt_ps(rsq32);
1715 rinv33 = gmx_mm_invsqrt_ps(rsq33);
1717 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
1718 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
1719 rinvsq13 = _mm_mul_ps(rinv13,rinv13);
1720 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
1721 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
1722 rinvsq23 = _mm_mul_ps(rinv23,rinv23);
1723 rinvsq31 = _mm_mul_ps(rinv31,rinv31);
1724 rinvsq32 = _mm_mul_ps(rinv32,rinv32);
1725 rinvsq33 = _mm_mul_ps(rinv33,rinv33);
1727 fjx0 = _mm_setzero_ps();
1728 fjy0 = _mm_setzero_ps();
1729 fjz0 = _mm_setzero_ps();
1730 fjx1 = _mm_setzero_ps();
1731 fjy1 = _mm_setzero_ps();
1732 fjz1 = _mm_setzero_ps();
1733 fjx2 = _mm_setzero_ps();
1734 fjy2 = _mm_setzero_ps();
1735 fjz2 = _mm_setzero_ps();
1736 fjx3 = _mm_setzero_ps();
1737 fjy3 = _mm_setzero_ps();
1738 fjz3 = _mm_setzero_ps();
1740 /**************************
1741 * CALCULATE INTERACTIONS *
1742 **************************/
1744 r00 = _mm_mul_ps(rsq00,rinv00);
1745 r00 = _mm_andnot_ps(dummy_mask,r00);
1747 /* Calculate table index by multiplying r with table scale and truncate to integer */
1748 rt = _mm_mul_ps(r00,vftabscale);
1749 vfitab = _mm_cvttps_epi32(rt);
1750 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1751 vfitab = _mm_slli_epi32(vfitab,3);
1753 /* CUBIC SPLINE TABLE DISPERSION */
1754 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1755 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1756 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1757 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1758 _MM_TRANSPOSE4_PS(Y,F,G,H);
1759 Heps = _mm_mul_ps(vfeps,H);
1760 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1761 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1762 fvdw6 = _mm_mul_ps(c6_00,FF);
1764 /* CUBIC SPLINE TABLE REPULSION */
1765 vfitab = _mm_add_epi32(vfitab,ifour);
1766 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1767 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1768 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1769 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1770 _MM_TRANSPOSE4_PS(Y,F,G,H);
1771 Heps = _mm_mul_ps(vfeps,H);
1772 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1773 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1774 fvdw12 = _mm_mul_ps(c12_00,FF);
1775 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1779 fscal = _mm_andnot_ps(dummy_mask,fscal);
1781 /* Calculate temporary vectorial force */
1782 tx = _mm_mul_ps(fscal,dx00);
1783 ty = _mm_mul_ps(fscal,dy00);
1784 tz = _mm_mul_ps(fscal,dz00);
1786 /* Update vectorial force */
1787 fix0 = _mm_add_ps(fix0,tx);
1788 fiy0 = _mm_add_ps(fiy0,ty);
1789 fiz0 = _mm_add_ps(fiz0,tz);
1791 fjx0 = _mm_add_ps(fjx0,tx);
1792 fjy0 = _mm_add_ps(fjy0,ty);
1793 fjz0 = _mm_add_ps(fjz0,tz);
1795 /**************************
1796 * CALCULATE INTERACTIONS *
1797 **************************/
1799 /* COULOMB ELECTROSTATICS */
1800 velec = _mm_mul_ps(qq11,rinv11);
1801 felec = _mm_mul_ps(velec,rinvsq11);
1805 fscal = _mm_andnot_ps(dummy_mask,fscal);
1807 /* Calculate temporary vectorial force */
1808 tx = _mm_mul_ps(fscal,dx11);
1809 ty = _mm_mul_ps(fscal,dy11);
1810 tz = _mm_mul_ps(fscal,dz11);
1812 /* Update vectorial force */
1813 fix1 = _mm_add_ps(fix1,tx);
1814 fiy1 = _mm_add_ps(fiy1,ty);
1815 fiz1 = _mm_add_ps(fiz1,tz);
1817 fjx1 = _mm_add_ps(fjx1,tx);
1818 fjy1 = _mm_add_ps(fjy1,ty);
1819 fjz1 = _mm_add_ps(fjz1,tz);
1821 /**************************
1822 * CALCULATE INTERACTIONS *
1823 **************************/
1825 /* COULOMB ELECTROSTATICS */
1826 velec = _mm_mul_ps(qq12,rinv12);
1827 felec = _mm_mul_ps(velec,rinvsq12);
1831 fscal = _mm_andnot_ps(dummy_mask,fscal);
1833 /* Calculate temporary vectorial force */
1834 tx = _mm_mul_ps(fscal,dx12);
1835 ty = _mm_mul_ps(fscal,dy12);
1836 tz = _mm_mul_ps(fscal,dz12);
1838 /* Update vectorial force */
1839 fix1 = _mm_add_ps(fix1,tx);
1840 fiy1 = _mm_add_ps(fiy1,ty);
1841 fiz1 = _mm_add_ps(fiz1,tz);
1843 fjx2 = _mm_add_ps(fjx2,tx);
1844 fjy2 = _mm_add_ps(fjy2,ty);
1845 fjz2 = _mm_add_ps(fjz2,tz);
1847 /**************************
1848 * CALCULATE INTERACTIONS *
1849 **************************/
1851 /* COULOMB ELECTROSTATICS */
1852 velec = _mm_mul_ps(qq13,rinv13);
1853 felec = _mm_mul_ps(velec,rinvsq13);
1857 fscal = _mm_andnot_ps(dummy_mask,fscal);
1859 /* Calculate temporary vectorial force */
1860 tx = _mm_mul_ps(fscal,dx13);
1861 ty = _mm_mul_ps(fscal,dy13);
1862 tz = _mm_mul_ps(fscal,dz13);
1864 /* Update vectorial force */
1865 fix1 = _mm_add_ps(fix1,tx);
1866 fiy1 = _mm_add_ps(fiy1,ty);
1867 fiz1 = _mm_add_ps(fiz1,tz);
1869 fjx3 = _mm_add_ps(fjx3,tx);
1870 fjy3 = _mm_add_ps(fjy3,ty);
1871 fjz3 = _mm_add_ps(fjz3,tz);
1873 /**************************
1874 * CALCULATE INTERACTIONS *
1875 **************************/
1877 /* COULOMB ELECTROSTATICS */
1878 velec = _mm_mul_ps(qq21,rinv21);
1879 felec = _mm_mul_ps(velec,rinvsq21);
1883 fscal = _mm_andnot_ps(dummy_mask,fscal);
1885 /* Calculate temporary vectorial force */
1886 tx = _mm_mul_ps(fscal,dx21);
1887 ty = _mm_mul_ps(fscal,dy21);
1888 tz = _mm_mul_ps(fscal,dz21);
1890 /* Update vectorial force */
1891 fix2 = _mm_add_ps(fix2,tx);
1892 fiy2 = _mm_add_ps(fiy2,ty);
1893 fiz2 = _mm_add_ps(fiz2,tz);
1895 fjx1 = _mm_add_ps(fjx1,tx);
1896 fjy1 = _mm_add_ps(fjy1,ty);
1897 fjz1 = _mm_add_ps(fjz1,tz);
1899 /**************************
1900 * CALCULATE INTERACTIONS *
1901 **************************/
1903 /* COULOMB ELECTROSTATICS */
1904 velec = _mm_mul_ps(qq22,rinv22);
1905 felec = _mm_mul_ps(velec,rinvsq22);
1909 fscal = _mm_andnot_ps(dummy_mask,fscal);
1911 /* Calculate temporary vectorial force */
1912 tx = _mm_mul_ps(fscal,dx22);
1913 ty = _mm_mul_ps(fscal,dy22);
1914 tz = _mm_mul_ps(fscal,dz22);
1916 /* Update vectorial force */
1917 fix2 = _mm_add_ps(fix2,tx);
1918 fiy2 = _mm_add_ps(fiy2,ty);
1919 fiz2 = _mm_add_ps(fiz2,tz);
1921 fjx2 = _mm_add_ps(fjx2,tx);
1922 fjy2 = _mm_add_ps(fjy2,ty);
1923 fjz2 = _mm_add_ps(fjz2,tz);
1925 /**************************
1926 * CALCULATE INTERACTIONS *
1927 **************************/
1929 /* COULOMB ELECTROSTATICS */
1930 velec = _mm_mul_ps(qq23,rinv23);
1931 felec = _mm_mul_ps(velec,rinvsq23);
1935 fscal = _mm_andnot_ps(dummy_mask,fscal);
1937 /* Calculate temporary vectorial force */
1938 tx = _mm_mul_ps(fscal,dx23);
1939 ty = _mm_mul_ps(fscal,dy23);
1940 tz = _mm_mul_ps(fscal,dz23);
1942 /* Update vectorial force */
1943 fix2 = _mm_add_ps(fix2,tx);
1944 fiy2 = _mm_add_ps(fiy2,ty);
1945 fiz2 = _mm_add_ps(fiz2,tz);
1947 fjx3 = _mm_add_ps(fjx3,tx);
1948 fjy3 = _mm_add_ps(fjy3,ty);
1949 fjz3 = _mm_add_ps(fjz3,tz);
1951 /**************************
1952 * CALCULATE INTERACTIONS *
1953 **************************/
1955 /* COULOMB ELECTROSTATICS */
1956 velec = _mm_mul_ps(qq31,rinv31);
1957 felec = _mm_mul_ps(velec,rinvsq31);
1961 fscal = _mm_andnot_ps(dummy_mask,fscal);
1963 /* Calculate temporary vectorial force */
1964 tx = _mm_mul_ps(fscal,dx31);
1965 ty = _mm_mul_ps(fscal,dy31);
1966 tz = _mm_mul_ps(fscal,dz31);
1968 /* Update vectorial force */
1969 fix3 = _mm_add_ps(fix3,tx);
1970 fiy3 = _mm_add_ps(fiy3,ty);
1971 fiz3 = _mm_add_ps(fiz3,tz);
1973 fjx1 = _mm_add_ps(fjx1,tx);
1974 fjy1 = _mm_add_ps(fjy1,ty);
1975 fjz1 = _mm_add_ps(fjz1,tz);
1977 /**************************
1978 * CALCULATE INTERACTIONS *
1979 **************************/
1981 /* COULOMB ELECTROSTATICS */
1982 velec = _mm_mul_ps(qq32,rinv32);
1983 felec = _mm_mul_ps(velec,rinvsq32);
1987 fscal = _mm_andnot_ps(dummy_mask,fscal);
1989 /* Calculate temporary vectorial force */
1990 tx = _mm_mul_ps(fscal,dx32);
1991 ty = _mm_mul_ps(fscal,dy32);
1992 tz = _mm_mul_ps(fscal,dz32);
1994 /* Update vectorial force */
1995 fix3 = _mm_add_ps(fix3,tx);
1996 fiy3 = _mm_add_ps(fiy3,ty);
1997 fiz3 = _mm_add_ps(fiz3,tz);
1999 fjx2 = _mm_add_ps(fjx2,tx);
2000 fjy2 = _mm_add_ps(fjy2,ty);
2001 fjz2 = _mm_add_ps(fjz2,tz);
2003 /**************************
2004 * CALCULATE INTERACTIONS *
2005 **************************/
2007 /* COULOMB ELECTROSTATICS */
2008 velec = _mm_mul_ps(qq33,rinv33);
2009 felec = _mm_mul_ps(velec,rinvsq33);
2013 fscal = _mm_andnot_ps(dummy_mask,fscal);
2015 /* Calculate temporary vectorial force */
2016 tx = _mm_mul_ps(fscal,dx33);
2017 ty = _mm_mul_ps(fscal,dy33);
2018 tz = _mm_mul_ps(fscal,dz33);
2020 /* Update vectorial force */
2021 fix3 = _mm_add_ps(fix3,tx);
2022 fiy3 = _mm_add_ps(fiy3,ty);
2023 fiz3 = _mm_add_ps(fiz3,tz);
2025 fjx3 = _mm_add_ps(fjx3,tx);
2026 fjy3 = _mm_add_ps(fjy3,ty);
2027 fjz3 = _mm_add_ps(fjz3,tz);
2029 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2030 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2031 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2032 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2034 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2035 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2036 fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2038 /* Inner loop uses 295 flops */
2041 /* End of innermost loop */
2043 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2044 f+i_coord_offset,fshift+i_shift_offset);
2046 /* Increment number of inner iterations */
2047 inneriter += j_index_end - j_index_start;
2049 /* Outer loop uses 24 flops */
2052 /* Increment number of outer iterations */
2055 /* Update outer/inner flops */
2057 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*295);