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_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_single
38 * Electrostatics interaction: CubicSplineTable
39 * VdW interaction: CubicSplineTable
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_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;
75 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
76 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
77 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
78 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
79 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
80 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
83 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
86 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
87 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
89 __m128i ifour = _mm_set1_epi32(4);
90 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
92 __m128 dummy_mask,cutoff_mask;
93 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
94 __m128 one = _mm_set1_ps(1.0);
95 __m128 two = _mm_set1_ps(2.0);
101 jindex = nlist->jindex;
103 shiftidx = nlist->shift;
105 shiftvec = fr->shift_vec[0];
106 fshift = fr->fshift[0];
107 facel = _mm_set1_ps(fr->epsfac);
108 charge = mdatoms->chargeA;
109 nvdwtype = fr->ntype;
111 vdwtype = mdatoms->typeA;
113 vftab = kernel_data->table_elec_vdw->data;
114 vftabscale = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
116 /* Setup water-specific parameters */
117 inr = nlist->iinr[0];
118 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
119 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
120 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
121 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
123 /* Avoid stupid compiler warnings */
124 jnrA = jnrB = jnrC = jnrD = 0;
133 for(iidx=0;iidx<4*DIM;iidx++)
138 /* Start outer loop over neighborlists */
139 for(iidx=0; iidx<nri; iidx++)
141 /* Load shift vector for this list */
142 i_shift_offset = DIM*shiftidx[iidx];
144 /* Load limits for loop over neighbors */
145 j_index_start = jindex[iidx];
146 j_index_end = jindex[iidx+1];
148 /* Get outer coordinate index */
150 i_coord_offset = DIM*inr;
152 /* Load i particle coords and add shift vector */
153 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
154 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
156 fix0 = _mm_setzero_ps();
157 fiy0 = _mm_setzero_ps();
158 fiz0 = _mm_setzero_ps();
159 fix1 = _mm_setzero_ps();
160 fiy1 = _mm_setzero_ps();
161 fiz1 = _mm_setzero_ps();
162 fix2 = _mm_setzero_ps();
163 fiy2 = _mm_setzero_ps();
164 fiz2 = _mm_setzero_ps();
166 /* Reset potential sums */
167 velecsum = _mm_setzero_ps();
168 vvdwsum = _mm_setzero_ps();
170 /* Start inner kernel loop */
171 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
174 /* Get j neighbor index, and coordinate index */
179 j_coord_offsetA = DIM*jnrA;
180 j_coord_offsetB = DIM*jnrB;
181 j_coord_offsetC = DIM*jnrC;
182 j_coord_offsetD = DIM*jnrD;
184 /* load j atom coordinates */
185 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
186 x+j_coord_offsetC,x+j_coord_offsetD,
189 /* Calculate displacement vector */
190 dx00 = _mm_sub_ps(ix0,jx0);
191 dy00 = _mm_sub_ps(iy0,jy0);
192 dz00 = _mm_sub_ps(iz0,jz0);
193 dx10 = _mm_sub_ps(ix1,jx0);
194 dy10 = _mm_sub_ps(iy1,jy0);
195 dz10 = _mm_sub_ps(iz1,jz0);
196 dx20 = _mm_sub_ps(ix2,jx0);
197 dy20 = _mm_sub_ps(iy2,jy0);
198 dz20 = _mm_sub_ps(iz2,jz0);
200 /* Calculate squared distance and things based on it */
201 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
202 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
203 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
205 rinv00 = gmx_mm_invsqrt_ps(rsq00);
206 rinv10 = gmx_mm_invsqrt_ps(rsq10);
207 rinv20 = gmx_mm_invsqrt_ps(rsq20);
209 /* Load parameters for j particles */
210 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
211 charge+jnrC+0,charge+jnrD+0);
212 vdwjidx0A = 2*vdwtype[jnrA+0];
213 vdwjidx0B = 2*vdwtype[jnrB+0];
214 vdwjidx0C = 2*vdwtype[jnrC+0];
215 vdwjidx0D = 2*vdwtype[jnrD+0];
217 /**************************
218 * CALCULATE INTERACTIONS *
219 **************************/
221 r00 = _mm_mul_ps(rsq00,rinv00);
223 /* Compute parameters for interactions between i and j atoms */
224 qq00 = _mm_mul_ps(iq0,jq0);
225 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
226 vdwparam+vdwioffset0+vdwjidx0B,
227 vdwparam+vdwioffset0+vdwjidx0C,
228 vdwparam+vdwioffset0+vdwjidx0D,
231 /* Calculate table index by multiplying r with table scale and truncate to integer */
232 rt = _mm_mul_ps(r00,vftabscale);
233 vfitab = _mm_cvttps_epi32(rt);
234 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
235 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
237 /* CUBIC SPLINE TABLE ELECTROSTATICS */
238 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
239 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
240 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
241 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
242 _MM_TRANSPOSE4_PS(Y,F,G,H);
243 Heps = _mm_mul_ps(vfeps,H);
244 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
245 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
246 velec = _mm_mul_ps(qq00,VV);
247 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
248 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
250 /* CUBIC SPLINE TABLE DISPERSION */
251 vfitab = _mm_add_epi32(vfitab,ifour);
252 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
253 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
254 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
255 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
256 _MM_TRANSPOSE4_PS(Y,F,G,H);
257 Heps = _mm_mul_ps(vfeps,H);
258 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
259 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
260 vvdw6 = _mm_mul_ps(c6_00,VV);
261 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
262 fvdw6 = _mm_mul_ps(c6_00,FF);
264 /* CUBIC SPLINE TABLE REPULSION */
265 vfitab = _mm_add_epi32(vfitab,ifour);
266 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
267 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
268 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
269 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
270 _MM_TRANSPOSE4_PS(Y,F,G,H);
271 Heps = _mm_mul_ps(vfeps,H);
272 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
273 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
274 vvdw12 = _mm_mul_ps(c12_00,VV);
275 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
276 fvdw12 = _mm_mul_ps(c12_00,FF);
277 vvdw = _mm_add_ps(vvdw12,vvdw6);
278 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
280 /* Update potential sum for this i atom from the interaction with this j atom. */
281 velecsum = _mm_add_ps(velecsum,velec);
282 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
284 fscal = _mm_add_ps(felec,fvdw);
286 /* Calculate temporary vectorial force */
287 tx = _mm_mul_ps(fscal,dx00);
288 ty = _mm_mul_ps(fscal,dy00);
289 tz = _mm_mul_ps(fscal,dz00);
291 /* Update vectorial force */
292 fix0 = _mm_add_ps(fix0,tx);
293 fiy0 = _mm_add_ps(fiy0,ty);
294 fiz0 = _mm_add_ps(fiz0,tz);
296 fjptrA = f+j_coord_offsetA;
297 fjptrB = f+j_coord_offsetB;
298 fjptrC = f+j_coord_offsetC;
299 fjptrD = f+j_coord_offsetD;
300 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
302 /**************************
303 * CALCULATE INTERACTIONS *
304 **************************/
306 r10 = _mm_mul_ps(rsq10,rinv10);
308 /* Compute parameters for interactions between i and j atoms */
309 qq10 = _mm_mul_ps(iq1,jq0);
311 /* Calculate table index by multiplying r with table scale and truncate to integer */
312 rt = _mm_mul_ps(r10,vftabscale);
313 vfitab = _mm_cvttps_epi32(rt);
314 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
315 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
317 /* CUBIC SPLINE TABLE ELECTROSTATICS */
318 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
319 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
320 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
321 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
322 _MM_TRANSPOSE4_PS(Y,F,G,H);
323 Heps = _mm_mul_ps(vfeps,H);
324 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
325 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
326 velec = _mm_mul_ps(qq10,VV);
327 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
328 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
330 /* Update potential sum for this i atom from the interaction with this j atom. */
331 velecsum = _mm_add_ps(velecsum,velec);
335 /* Calculate temporary vectorial force */
336 tx = _mm_mul_ps(fscal,dx10);
337 ty = _mm_mul_ps(fscal,dy10);
338 tz = _mm_mul_ps(fscal,dz10);
340 /* Update vectorial force */
341 fix1 = _mm_add_ps(fix1,tx);
342 fiy1 = _mm_add_ps(fiy1,ty);
343 fiz1 = _mm_add_ps(fiz1,tz);
345 fjptrA = f+j_coord_offsetA;
346 fjptrB = f+j_coord_offsetB;
347 fjptrC = f+j_coord_offsetC;
348 fjptrD = f+j_coord_offsetD;
349 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
351 /**************************
352 * CALCULATE INTERACTIONS *
353 **************************/
355 r20 = _mm_mul_ps(rsq20,rinv20);
357 /* Compute parameters for interactions between i and j atoms */
358 qq20 = _mm_mul_ps(iq2,jq0);
360 /* Calculate table index by multiplying r with table scale and truncate to integer */
361 rt = _mm_mul_ps(r20,vftabscale);
362 vfitab = _mm_cvttps_epi32(rt);
363 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
364 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
366 /* CUBIC SPLINE TABLE ELECTROSTATICS */
367 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
368 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
369 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
370 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
371 _MM_TRANSPOSE4_PS(Y,F,G,H);
372 Heps = _mm_mul_ps(vfeps,H);
373 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
374 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
375 velec = _mm_mul_ps(qq20,VV);
376 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
377 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
379 /* Update potential sum for this i atom from the interaction with this j atom. */
380 velecsum = _mm_add_ps(velecsum,velec);
384 /* Calculate temporary vectorial force */
385 tx = _mm_mul_ps(fscal,dx20);
386 ty = _mm_mul_ps(fscal,dy20);
387 tz = _mm_mul_ps(fscal,dz20);
389 /* Update vectorial force */
390 fix2 = _mm_add_ps(fix2,tx);
391 fiy2 = _mm_add_ps(fiy2,ty);
392 fiz2 = _mm_add_ps(fiz2,tz);
394 fjptrA = f+j_coord_offsetA;
395 fjptrB = f+j_coord_offsetB;
396 fjptrC = f+j_coord_offsetC;
397 fjptrD = f+j_coord_offsetD;
398 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
400 /* Inner loop uses 159 flops */
406 /* Get j neighbor index, and coordinate index */
407 jnrlistA = jjnr[jidx];
408 jnrlistB = jjnr[jidx+1];
409 jnrlistC = jjnr[jidx+2];
410 jnrlistD = jjnr[jidx+3];
411 /* Sign of each element will be negative for non-real atoms.
412 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
413 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
415 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
416 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
417 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
418 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
419 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
420 j_coord_offsetA = DIM*jnrA;
421 j_coord_offsetB = DIM*jnrB;
422 j_coord_offsetC = DIM*jnrC;
423 j_coord_offsetD = DIM*jnrD;
425 /* load j atom coordinates */
426 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
427 x+j_coord_offsetC,x+j_coord_offsetD,
430 /* Calculate displacement vector */
431 dx00 = _mm_sub_ps(ix0,jx0);
432 dy00 = _mm_sub_ps(iy0,jy0);
433 dz00 = _mm_sub_ps(iz0,jz0);
434 dx10 = _mm_sub_ps(ix1,jx0);
435 dy10 = _mm_sub_ps(iy1,jy0);
436 dz10 = _mm_sub_ps(iz1,jz0);
437 dx20 = _mm_sub_ps(ix2,jx0);
438 dy20 = _mm_sub_ps(iy2,jy0);
439 dz20 = _mm_sub_ps(iz2,jz0);
441 /* Calculate squared distance and things based on it */
442 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
443 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
444 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
446 rinv00 = gmx_mm_invsqrt_ps(rsq00);
447 rinv10 = gmx_mm_invsqrt_ps(rsq10);
448 rinv20 = gmx_mm_invsqrt_ps(rsq20);
450 /* Load parameters for j particles */
451 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
452 charge+jnrC+0,charge+jnrD+0);
453 vdwjidx0A = 2*vdwtype[jnrA+0];
454 vdwjidx0B = 2*vdwtype[jnrB+0];
455 vdwjidx0C = 2*vdwtype[jnrC+0];
456 vdwjidx0D = 2*vdwtype[jnrD+0];
458 /**************************
459 * CALCULATE INTERACTIONS *
460 **************************/
462 r00 = _mm_mul_ps(rsq00,rinv00);
463 r00 = _mm_andnot_ps(dummy_mask,r00);
465 /* Compute parameters for interactions between i and j atoms */
466 qq00 = _mm_mul_ps(iq0,jq0);
467 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
468 vdwparam+vdwioffset0+vdwjidx0B,
469 vdwparam+vdwioffset0+vdwjidx0C,
470 vdwparam+vdwioffset0+vdwjidx0D,
473 /* Calculate table index by multiplying r with table scale and truncate to integer */
474 rt = _mm_mul_ps(r00,vftabscale);
475 vfitab = _mm_cvttps_epi32(rt);
476 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
477 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
479 /* CUBIC SPLINE TABLE ELECTROSTATICS */
480 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
481 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
482 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
483 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
484 _MM_TRANSPOSE4_PS(Y,F,G,H);
485 Heps = _mm_mul_ps(vfeps,H);
486 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
487 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
488 velec = _mm_mul_ps(qq00,VV);
489 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
490 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
492 /* CUBIC SPLINE TABLE DISPERSION */
493 vfitab = _mm_add_epi32(vfitab,ifour);
494 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
495 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
496 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
497 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
498 _MM_TRANSPOSE4_PS(Y,F,G,H);
499 Heps = _mm_mul_ps(vfeps,H);
500 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
501 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
502 vvdw6 = _mm_mul_ps(c6_00,VV);
503 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
504 fvdw6 = _mm_mul_ps(c6_00,FF);
506 /* CUBIC SPLINE TABLE REPULSION */
507 vfitab = _mm_add_epi32(vfitab,ifour);
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 vvdw12 = _mm_mul_ps(c12_00,VV);
517 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
518 fvdw12 = _mm_mul_ps(c12_00,FF);
519 vvdw = _mm_add_ps(vvdw12,vvdw6);
520 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
522 /* Update potential sum for this i atom from the interaction with this j atom. */
523 velec = _mm_andnot_ps(dummy_mask,velec);
524 velecsum = _mm_add_ps(velecsum,velec);
525 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
526 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
528 fscal = _mm_add_ps(felec,fvdw);
530 fscal = _mm_andnot_ps(dummy_mask,fscal);
532 /* Calculate temporary vectorial force */
533 tx = _mm_mul_ps(fscal,dx00);
534 ty = _mm_mul_ps(fscal,dy00);
535 tz = _mm_mul_ps(fscal,dz00);
537 /* Update vectorial force */
538 fix0 = _mm_add_ps(fix0,tx);
539 fiy0 = _mm_add_ps(fiy0,ty);
540 fiz0 = _mm_add_ps(fiz0,tz);
542 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
543 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
544 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
545 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
546 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
548 /**************************
549 * CALCULATE INTERACTIONS *
550 **************************/
552 r10 = _mm_mul_ps(rsq10,rinv10);
553 r10 = _mm_andnot_ps(dummy_mask,r10);
555 /* Compute parameters for interactions between i and j atoms */
556 qq10 = _mm_mul_ps(iq1,jq0);
558 /* Calculate table index by multiplying r with table scale and truncate to integer */
559 rt = _mm_mul_ps(r10,vftabscale);
560 vfitab = _mm_cvttps_epi32(rt);
561 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
562 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
564 /* CUBIC SPLINE TABLE ELECTROSTATICS */
565 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
566 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
567 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
568 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
569 _MM_TRANSPOSE4_PS(Y,F,G,H);
570 Heps = _mm_mul_ps(vfeps,H);
571 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
572 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
573 velec = _mm_mul_ps(qq10,VV);
574 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
575 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
577 /* Update potential sum for this i atom from the interaction with this j atom. */
578 velec = _mm_andnot_ps(dummy_mask,velec);
579 velecsum = _mm_add_ps(velecsum,velec);
583 fscal = _mm_andnot_ps(dummy_mask,fscal);
585 /* Calculate temporary vectorial force */
586 tx = _mm_mul_ps(fscal,dx10);
587 ty = _mm_mul_ps(fscal,dy10);
588 tz = _mm_mul_ps(fscal,dz10);
590 /* Update vectorial force */
591 fix1 = _mm_add_ps(fix1,tx);
592 fiy1 = _mm_add_ps(fiy1,ty);
593 fiz1 = _mm_add_ps(fiz1,tz);
595 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
596 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
597 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
598 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
599 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
601 /**************************
602 * CALCULATE INTERACTIONS *
603 **************************/
605 r20 = _mm_mul_ps(rsq20,rinv20);
606 r20 = _mm_andnot_ps(dummy_mask,r20);
608 /* Compute parameters for interactions between i and j atoms */
609 qq20 = _mm_mul_ps(iq2,jq0);
611 /* Calculate table index by multiplying r with table scale and truncate to integer */
612 rt = _mm_mul_ps(r20,vftabscale);
613 vfitab = _mm_cvttps_epi32(rt);
614 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
615 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
617 /* CUBIC SPLINE TABLE ELECTROSTATICS */
618 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
619 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
620 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
621 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
622 _MM_TRANSPOSE4_PS(Y,F,G,H);
623 Heps = _mm_mul_ps(vfeps,H);
624 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
625 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
626 velec = _mm_mul_ps(qq20,VV);
627 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
628 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
630 /* Update potential sum for this i atom from the interaction with this j atom. */
631 velec = _mm_andnot_ps(dummy_mask,velec);
632 velecsum = _mm_add_ps(velecsum,velec);
636 fscal = _mm_andnot_ps(dummy_mask,fscal);
638 /* Calculate temporary vectorial force */
639 tx = _mm_mul_ps(fscal,dx20);
640 ty = _mm_mul_ps(fscal,dy20);
641 tz = _mm_mul_ps(fscal,dz20);
643 /* Update vectorial force */
644 fix2 = _mm_add_ps(fix2,tx);
645 fiy2 = _mm_add_ps(fiy2,ty);
646 fiz2 = _mm_add_ps(fiz2,tz);
648 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
649 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
650 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
651 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
652 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
654 /* Inner loop uses 162 flops */
657 /* End of innermost loop */
659 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
660 f+i_coord_offset,fshift+i_shift_offset);
663 /* Update potential energies */
664 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
665 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
667 /* Increment number of inner iterations */
668 inneriter += j_index_end - j_index_start;
670 /* Outer loop uses 20 flops */
673 /* Increment number of outer iterations */
676 /* Update outer/inner flops */
678 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*162);
681 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_single
682 * Electrostatics interaction: CubicSplineTable
683 * VdW interaction: CubicSplineTable
684 * Geometry: Water3-Particle
685 * Calculate force/pot: Force
688 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_single
689 (t_nblist * gmx_restrict nlist,
690 rvec * gmx_restrict xx,
691 rvec * gmx_restrict ff,
692 t_forcerec * gmx_restrict fr,
693 t_mdatoms * gmx_restrict mdatoms,
694 nb_kernel_data_t * gmx_restrict kernel_data,
695 t_nrnb * gmx_restrict nrnb)
697 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
698 * just 0 for non-waters.
699 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
700 * jnr indices corresponding to data put in the four positions in the SIMD register.
702 int i_shift_offset,i_coord_offset,outeriter,inneriter;
703 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
704 int jnrA,jnrB,jnrC,jnrD;
705 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
706 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
707 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
709 real *shiftvec,*fshift,*x,*f;
710 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
712 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
714 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
716 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
718 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
719 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
720 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
721 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
722 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
723 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
724 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
727 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
730 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
731 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
733 __m128i ifour = _mm_set1_epi32(4);
734 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
736 __m128 dummy_mask,cutoff_mask;
737 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
738 __m128 one = _mm_set1_ps(1.0);
739 __m128 two = _mm_set1_ps(2.0);
745 jindex = nlist->jindex;
747 shiftidx = nlist->shift;
749 shiftvec = fr->shift_vec[0];
750 fshift = fr->fshift[0];
751 facel = _mm_set1_ps(fr->epsfac);
752 charge = mdatoms->chargeA;
753 nvdwtype = fr->ntype;
755 vdwtype = mdatoms->typeA;
757 vftab = kernel_data->table_elec_vdw->data;
758 vftabscale = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
760 /* Setup water-specific parameters */
761 inr = nlist->iinr[0];
762 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
763 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
764 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
765 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
767 /* Avoid stupid compiler warnings */
768 jnrA = jnrB = jnrC = jnrD = 0;
777 for(iidx=0;iidx<4*DIM;iidx++)
782 /* Start outer loop over neighborlists */
783 for(iidx=0; iidx<nri; iidx++)
785 /* Load shift vector for this list */
786 i_shift_offset = DIM*shiftidx[iidx];
788 /* Load limits for loop over neighbors */
789 j_index_start = jindex[iidx];
790 j_index_end = jindex[iidx+1];
792 /* Get outer coordinate index */
794 i_coord_offset = DIM*inr;
796 /* Load i particle coords and add shift vector */
797 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
798 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
800 fix0 = _mm_setzero_ps();
801 fiy0 = _mm_setzero_ps();
802 fiz0 = _mm_setzero_ps();
803 fix1 = _mm_setzero_ps();
804 fiy1 = _mm_setzero_ps();
805 fiz1 = _mm_setzero_ps();
806 fix2 = _mm_setzero_ps();
807 fiy2 = _mm_setzero_ps();
808 fiz2 = _mm_setzero_ps();
810 /* Start inner kernel loop */
811 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
814 /* Get j neighbor index, and coordinate index */
819 j_coord_offsetA = DIM*jnrA;
820 j_coord_offsetB = DIM*jnrB;
821 j_coord_offsetC = DIM*jnrC;
822 j_coord_offsetD = DIM*jnrD;
824 /* load j atom coordinates */
825 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
826 x+j_coord_offsetC,x+j_coord_offsetD,
829 /* Calculate displacement vector */
830 dx00 = _mm_sub_ps(ix0,jx0);
831 dy00 = _mm_sub_ps(iy0,jy0);
832 dz00 = _mm_sub_ps(iz0,jz0);
833 dx10 = _mm_sub_ps(ix1,jx0);
834 dy10 = _mm_sub_ps(iy1,jy0);
835 dz10 = _mm_sub_ps(iz1,jz0);
836 dx20 = _mm_sub_ps(ix2,jx0);
837 dy20 = _mm_sub_ps(iy2,jy0);
838 dz20 = _mm_sub_ps(iz2,jz0);
840 /* Calculate squared distance and things based on it */
841 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
842 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
843 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
845 rinv00 = gmx_mm_invsqrt_ps(rsq00);
846 rinv10 = gmx_mm_invsqrt_ps(rsq10);
847 rinv20 = gmx_mm_invsqrt_ps(rsq20);
849 /* Load parameters for j particles */
850 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
851 charge+jnrC+0,charge+jnrD+0);
852 vdwjidx0A = 2*vdwtype[jnrA+0];
853 vdwjidx0B = 2*vdwtype[jnrB+0];
854 vdwjidx0C = 2*vdwtype[jnrC+0];
855 vdwjidx0D = 2*vdwtype[jnrD+0];
857 /**************************
858 * CALCULATE INTERACTIONS *
859 **************************/
861 r00 = _mm_mul_ps(rsq00,rinv00);
863 /* Compute parameters for interactions between i and j atoms */
864 qq00 = _mm_mul_ps(iq0,jq0);
865 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
866 vdwparam+vdwioffset0+vdwjidx0B,
867 vdwparam+vdwioffset0+vdwjidx0C,
868 vdwparam+vdwioffset0+vdwjidx0D,
871 /* Calculate table index by multiplying r with table scale and truncate to integer */
872 rt = _mm_mul_ps(r00,vftabscale);
873 vfitab = _mm_cvttps_epi32(rt);
874 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
875 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
877 /* CUBIC SPLINE TABLE ELECTROSTATICS */
878 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
879 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
880 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
881 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
882 _MM_TRANSPOSE4_PS(Y,F,G,H);
883 Heps = _mm_mul_ps(vfeps,H);
884 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
885 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
886 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
888 /* CUBIC SPLINE TABLE DISPERSION */
889 vfitab = _mm_add_epi32(vfitab,ifour);
890 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
891 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
892 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
893 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
894 _MM_TRANSPOSE4_PS(Y,F,G,H);
895 Heps = _mm_mul_ps(vfeps,H);
896 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
897 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
898 fvdw6 = _mm_mul_ps(c6_00,FF);
900 /* CUBIC SPLINE TABLE REPULSION */
901 vfitab = _mm_add_epi32(vfitab,ifour);
902 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
903 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
904 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
905 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
906 _MM_TRANSPOSE4_PS(Y,F,G,H);
907 Heps = _mm_mul_ps(vfeps,H);
908 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
909 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
910 fvdw12 = _mm_mul_ps(c12_00,FF);
911 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
913 fscal = _mm_add_ps(felec,fvdw);
915 /* Calculate temporary vectorial force */
916 tx = _mm_mul_ps(fscal,dx00);
917 ty = _mm_mul_ps(fscal,dy00);
918 tz = _mm_mul_ps(fscal,dz00);
920 /* Update vectorial force */
921 fix0 = _mm_add_ps(fix0,tx);
922 fiy0 = _mm_add_ps(fiy0,ty);
923 fiz0 = _mm_add_ps(fiz0,tz);
925 fjptrA = f+j_coord_offsetA;
926 fjptrB = f+j_coord_offsetB;
927 fjptrC = f+j_coord_offsetC;
928 fjptrD = f+j_coord_offsetD;
929 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
931 /**************************
932 * CALCULATE INTERACTIONS *
933 **************************/
935 r10 = _mm_mul_ps(rsq10,rinv10);
937 /* Compute parameters for interactions between i and j atoms */
938 qq10 = _mm_mul_ps(iq1,jq0);
940 /* Calculate table index by multiplying r with table scale and truncate to integer */
941 rt = _mm_mul_ps(r10,vftabscale);
942 vfitab = _mm_cvttps_epi32(rt);
943 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
944 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
946 /* CUBIC SPLINE TABLE ELECTROSTATICS */
947 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
948 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
949 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
950 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
951 _MM_TRANSPOSE4_PS(Y,F,G,H);
952 Heps = _mm_mul_ps(vfeps,H);
953 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
954 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
955 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
959 /* Calculate temporary vectorial force */
960 tx = _mm_mul_ps(fscal,dx10);
961 ty = _mm_mul_ps(fscal,dy10);
962 tz = _mm_mul_ps(fscal,dz10);
964 /* Update vectorial force */
965 fix1 = _mm_add_ps(fix1,tx);
966 fiy1 = _mm_add_ps(fiy1,ty);
967 fiz1 = _mm_add_ps(fiz1,tz);
969 fjptrA = f+j_coord_offsetA;
970 fjptrB = f+j_coord_offsetB;
971 fjptrC = f+j_coord_offsetC;
972 fjptrD = f+j_coord_offsetD;
973 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
975 /**************************
976 * CALCULATE INTERACTIONS *
977 **************************/
979 r20 = _mm_mul_ps(rsq20,rinv20);
981 /* Compute parameters for interactions between i and j atoms */
982 qq20 = _mm_mul_ps(iq2,jq0);
984 /* Calculate table index by multiplying r with table scale and truncate to integer */
985 rt = _mm_mul_ps(r20,vftabscale);
986 vfitab = _mm_cvttps_epi32(rt);
987 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
988 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
990 /* CUBIC SPLINE TABLE ELECTROSTATICS */
991 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
992 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
993 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
994 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
995 _MM_TRANSPOSE4_PS(Y,F,G,H);
996 Heps = _mm_mul_ps(vfeps,H);
997 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
998 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
999 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1003 /* Calculate temporary vectorial force */
1004 tx = _mm_mul_ps(fscal,dx20);
1005 ty = _mm_mul_ps(fscal,dy20);
1006 tz = _mm_mul_ps(fscal,dz20);
1008 /* Update vectorial force */
1009 fix2 = _mm_add_ps(fix2,tx);
1010 fiy2 = _mm_add_ps(fiy2,ty);
1011 fiz2 = _mm_add_ps(fiz2,tz);
1013 fjptrA = f+j_coord_offsetA;
1014 fjptrB = f+j_coord_offsetB;
1015 fjptrC = f+j_coord_offsetC;
1016 fjptrD = f+j_coord_offsetD;
1017 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1019 /* Inner loop uses 139 flops */
1022 if(jidx<j_index_end)
1025 /* Get j neighbor index, and coordinate index */
1026 jnrlistA = jjnr[jidx];
1027 jnrlistB = jjnr[jidx+1];
1028 jnrlistC = jjnr[jidx+2];
1029 jnrlistD = jjnr[jidx+3];
1030 /* Sign of each element will be negative for non-real atoms.
1031 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1032 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1034 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1035 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1036 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1037 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1038 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1039 j_coord_offsetA = DIM*jnrA;
1040 j_coord_offsetB = DIM*jnrB;
1041 j_coord_offsetC = DIM*jnrC;
1042 j_coord_offsetD = DIM*jnrD;
1044 /* load j atom coordinates */
1045 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1046 x+j_coord_offsetC,x+j_coord_offsetD,
1049 /* Calculate displacement vector */
1050 dx00 = _mm_sub_ps(ix0,jx0);
1051 dy00 = _mm_sub_ps(iy0,jy0);
1052 dz00 = _mm_sub_ps(iz0,jz0);
1053 dx10 = _mm_sub_ps(ix1,jx0);
1054 dy10 = _mm_sub_ps(iy1,jy0);
1055 dz10 = _mm_sub_ps(iz1,jz0);
1056 dx20 = _mm_sub_ps(ix2,jx0);
1057 dy20 = _mm_sub_ps(iy2,jy0);
1058 dz20 = _mm_sub_ps(iz2,jz0);
1060 /* Calculate squared distance and things based on it */
1061 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1062 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1063 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1065 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1066 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1067 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1069 /* Load parameters for j particles */
1070 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1071 charge+jnrC+0,charge+jnrD+0);
1072 vdwjidx0A = 2*vdwtype[jnrA+0];
1073 vdwjidx0B = 2*vdwtype[jnrB+0];
1074 vdwjidx0C = 2*vdwtype[jnrC+0];
1075 vdwjidx0D = 2*vdwtype[jnrD+0];
1077 /**************************
1078 * CALCULATE INTERACTIONS *
1079 **************************/
1081 r00 = _mm_mul_ps(rsq00,rinv00);
1082 r00 = _mm_andnot_ps(dummy_mask,r00);
1084 /* Compute parameters for interactions between i and j atoms */
1085 qq00 = _mm_mul_ps(iq0,jq0);
1086 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1087 vdwparam+vdwioffset0+vdwjidx0B,
1088 vdwparam+vdwioffset0+vdwjidx0C,
1089 vdwparam+vdwioffset0+vdwjidx0D,
1092 /* Calculate table index by multiplying r with table scale and truncate to integer */
1093 rt = _mm_mul_ps(r00,vftabscale);
1094 vfitab = _mm_cvttps_epi32(rt);
1095 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1096 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1098 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1099 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1100 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1101 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1102 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1103 _MM_TRANSPOSE4_PS(Y,F,G,H);
1104 Heps = _mm_mul_ps(vfeps,H);
1105 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1106 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1107 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1109 /* CUBIC SPLINE TABLE DISPERSION */
1110 vfitab = _mm_add_epi32(vfitab,ifour);
1111 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1112 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1113 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1114 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1115 _MM_TRANSPOSE4_PS(Y,F,G,H);
1116 Heps = _mm_mul_ps(vfeps,H);
1117 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1118 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1119 fvdw6 = _mm_mul_ps(c6_00,FF);
1121 /* CUBIC SPLINE TABLE REPULSION */
1122 vfitab = _mm_add_epi32(vfitab,ifour);
1123 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1124 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1125 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1126 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1127 _MM_TRANSPOSE4_PS(Y,F,G,H);
1128 Heps = _mm_mul_ps(vfeps,H);
1129 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1130 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1131 fvdw12 = _mm_mul_ps(c12_00,FF);
1132 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1134 fscal = _mm_add_ps(felec,fvdw);
1136 fscal = _mm_andnot_ps(dummy_mask,fscal);
1138 /* Calculate temporary vectorial force */
1139 tx = _mm_mul_ps(fscal,dx00);
1140 ty = _mm_mul_ps(fscal,dy00);
1141 tz = _mm_mul_ps(fscal,dz00);
1143 /* Update vectorial force */
1144 fix0 = _mm_add_ps(fix0,tx);
1145 fiy0 = _mm_add_ps(fiy0,ty);
1146 fiz0 = _mm_add_ps(fiz0,tz);
1148 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1149 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1150 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1151 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1152 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1154 /**************************
1155 * CALCULATE INTERACTIONS *
1156 **************************/
1158 r10 = _mm_mul_ps(rsq10,rinv10);
1159 r10 = _mm_andnot_ps(dummy_mask,r10);
1161 /* Compute parameters for interactions between i and j atoms */
1162 qq10 = _mm_mul_ps(iq1,jq0);
1164 /* Calculate table index by multiplying r with table scale and truncate to integer */
1165 rt = _mm_mul_ps(r10,vftabscale);
1166 vfitab = _mm_cvttps_epi32(rt);
1167 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1168 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1170 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1171 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1172 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1173 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1174 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1175 _MM_TRANSPOSE4_PS(Y,F,G,H);
1176 Heps = _mm_mul_ps(vfeps,H);
1177 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1178 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1179 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1183 fscal = _mm_andnot_ps(dummy_mask,fscal);
1185 /* Calculate temporary vectorial force */
1186 tx = _mm_mul_ps(fscal,dx10);
1187 ty = _mm_mul_ps(fscal,dy10);
1188 tz = _mm_mul_ps(fscal,dz10);
1190 /* Update vectorial force */
1191 fix1 = _mm_add_ps(fix1,tx);
1192 fiy1 = _mm_add_ps(fiy1,ty);
1193 fiz1 = _mm_add_ps(fiz1,tz);
1195 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1196 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1197 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1198 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1199 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1201 /**************************
1202 * CALCULATE INTERACTIONS *
1203 **************************/
1205 r20 = _mm_mul_ps(rsq20,rinv20);
1206 r20 = _mm_andnot_ps(dummy_mask,r20);
1208 /* Compute parameters for interactions between i and j atoms */
1209 qq20 = _mm_mul_ps(iq2,jq0);
1211 /* Calculate table index by multiplying r with table scale and truncate to integer */
1212 rt = _mm_mul_ps(r20,vftabscale);
1213 vfitab = _mm_cvttps_epi32(rt);
1214 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1215 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1217 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1218 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1219 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1220 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1221 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1222 _MM_TRANSPOSE4_PS(Y,F,G,H);
1223 Heps = _mm_mul_ps(vfeps,H);
1224 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1225 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1226 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1230 fscal = _mm_andnot_ps(dummy_mask,fscal);
1232 /* Calculate temporary vectorial force */
1233 tx = _mm_mul_ps(fscal,dx20);
1234 ty = _mm_mul_ps(fscal,dy20);
1235 tz = _mm_mul_ps(fscal,dz20);
1237 /* Update vectorial force */
1238 fix2 = _mm_add_ps(fix2,tx);
1239 fiy2 = _mm_add_ps(fiy2,ty);
1240 fiz2 = _mm_add_ps(fiz2,tz);
1242 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1243 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1244 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1245 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1246 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1248 /* Inner loop uses 142 flops */
1251 /* End of innermost loop */
1253 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1254 f+i_coord_offset,fshift+i_shift_offset);
1256 /* Increment number of inner iterations */
1257 inneriter += j_index_end - j_index_start;
1259 /* Outer loop uses 18 flops */
1262 /* Increment number of outer iterations */
1265 /* Update outer/inner flops */
1267 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);