2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
50 #include "kernelutil_x86_avx_128_fma_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_128_fma_single
54 * Electrostatics interaction: CubicSplineTable
55 * VdW interaction: LennardJones
56 * Geometry: Water4-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_128_fma_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
84 __m128 fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
90 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
92 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
93 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
94 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
95 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
96 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
97 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
98 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
99 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
102 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
105 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
106 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
108 __m128i ifour = _mm_set1_epi32(4);
109 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
111 __m128 dummy_mask,cutoff_mask;
112 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
113 __m128 one = _mm_set1_ps(1.0);
114 __m128 two = _mm_set1_ps(2.0);
120 jindex = nlist->jindex;
122 shiftidx = nlist->shift;
124 shiftvec = fr->shift_vec[0];
125 fshift = fr->fshift[0];
126 facel = _mm_set1_ps(fr->epsfac);
127 charge = mdatoms->chargeA;
128 nvdwtype = fr->ntype;
130 vdwtype = mdatoms->typeA;
132 vftab = kernel_data->table_elec->data;
133 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
135 /* Setup water-specific parameters */
136 inr = nlist->iinr[0];
137 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
138 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
139 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
140 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
142 /* Avoid stupid compiler warnings */
143 jnrA = jnrB = jnrC = jnrD = 0;
152 for(iidx=0;iidx<4*DIM;iidx++)
157 /* Start outer loop over neighborlists */
158 for(iidx=0; iidx<nri; iidx++)
160 /* Load shift vector for this list */
161 i_shift_offset = DIM*shiftidx[iidx];
163 /* Load limits for loop over neighbors */
164 j_index_start = jindex[iidx];
165 j_index_end = jindex[iidx+1];
167 /* Get outer coordinate index */
169 i_coord_offset = DIM*inr;
171 /* Load i particle coords and add shift vector */
172 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
173 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
175 fix0 = _mm_setzero_ps();
176 fiy0 = _mm_setzero_ps();
177 fiz0 = _mm_setzero_ps();
178 fix1 = _mm_setzero_ps();
179 fiy1 = _mm_setzero_ps();
180 fiz1 = _mm_setzero_ps();
181 fix2 = _mm_setzero_ps();
182 fiy2 = _mm_setzero_ps();
183 fiz2 = _mm_setzero_ps();
184 fix3 = _mm_setzero_ps();
185 fiy3 = _mm_setzero_ps();
186 fiz3 = _mm_setzero_ps();
188 /* Reset potential sums */
189 velecsum = _mm_setzero_ps();
190 vvdwsum = _mm_setzero_ps();
192 /* Start inner kernel loop */
193 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
196 /* Get j neighbor index, and coordinate index */
201 j_coord_offsetA = DIM*jnrA;
202 j_coord_offsetB = DIM*jnrB;
203 j_coord_offsetC = DIM*jnrC;
204 j_coord_offsetD = DIM*jnrD;
206 /* load j atom coordinates */
207 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
208 x+j_coord_offsetC,x+j_coord_offsetD,
211 /* Calculate displacement vector */
212 dx00 = _mm_sub_ps(ix0,jx0);
213 dy00 = _mm_sub_ps(iy0,jy0);
214 dz00 = _mm_sub_ps(iz0,jz0);
215 dx10 = _mm_sub_ps(ix1,jx0);
216 dy10 = _mm_sub_ps(iy1,jy0);
217 dz10 = _mm_sub_ps(iz1,jz0);
218 dx20 = _mm_sub_ps(ix2,jx0);
219 dy20 = _mm_sub_ps(iy2,jy0);
220 dz20 = _mm_sub_ps(iz2,jz0);
221 dx30 = _mm_sub_ps(ix3,jx0);
222 dy30 = _mm_sub_ps(iy3,jy0);
223 dz30 = _mm_sub_ps(iz3,jz0);
225 /* Calculate squared distance and things based on it */
226 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
227 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
228 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
229 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
231 rinv10 = gmx_mm_invsqrt_ps(rsq10);
232 rinv20 = gmx_mm_invsqrt_ps(rsq20);
233 rinv30 = gmx_mm_invsqrt_ps(rsq30);
235 rinvsq00 = gmx_mm_inv_ps(rsq00);
237 /* Load parameters for j particles */
238 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
239 charge+jnrC+0,charge+jnrD+0);
240 vdwjidx0A = 2*vdwtype[jnrA+0];
241 vdwjidx0B = 2*vdwtype[jnrB+0];
242 vdwjidx0C = 2*vdwtype[jnrC+0];
243 vdwjidx0D = 2*vdwtype[jnrD+0];
245 fjx0 = _mm_setzero_ps();
246 fjy0 = _mm_setzero_ps();
247 fjz0 = _mm_setzero_ps();
249 /**************************
250 * CALCULATE INTERACTIONS *
251 **************************/
253 /* Compute parameters for interactions between i and j atoms */
254 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
255 vdwparam+vdwioffset0+vdwjidx0B,
256 vdwparam+vdwioffset0+vdwjidx0C,
257 vdwparam+vdwioffset0+vdwjidx0D,
260 /* LENNARD-JONES DISPERSION/REPULSION */
262 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
263 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
264 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
265 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
266 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
268 /* Update potential sum for this i atom from the interaction with this j atom. */
269 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
273 /* Update vectorial force */
274 fix0 = _mm_macc_ps(dx00,fscal,fix0);
275 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
276 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
278 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
279 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
280 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
282 /**************************
283 * CALCULATE INTERACTIONS *
284 **************************/
286 r10 = _mm_mul_ps(rsq10,rinv10);
288 /* Compute parameters for interactions between i and j atoms */
289 qq10 = _mm_mul_ps(iq1,jq0);
291 /* Calculate table index by multiplying r with table scale and truncate to integer */
292 rt = _mm_mul_ps(r10,vftabscale);
293 vfitab = _mm_cvttps_epi32(rt);
295 vfeps = _mm_frcz_ps(rt);
297 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
299 twovfeps = _mm_add_ps(vfeps,vfeps);
300 vfitab = _mm_slli_epi32(vfitab,2);
302 /* CUBIC SPLINE TABLE ELECTROSTATICS */
303 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
304 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
305 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
306 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
307 _MM_TRANSPOSE4_PS(Y,F,G,H);
308 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
309 VV = _mm_macc_ps(vfeps,Fp,Y);
310 velec = _mm_mul_ps(qq10,VV);
311 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
312 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
314 /* Update potential sum for this i atom from the interaction with this j atom. */
315 velecsum = _mm_add_ps(velecsum,velec);
319 /* Update vectorial force */
320 fix1 = _mm_macc_ps(dx10,fscal,fix1);
321 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
322 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
324 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
325 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
326 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
328 /**************************
329 * CALCULATE INTERACTIONS *
330 **************************/
332 r20 = _mm_mul_ps(rsq20,rinv20);
334 /* Compute parameters for interactions between i and j atoms */
335 qq20 = _mm_mul_ps(iq2,jq0);
337 /* Calculate table index by multiplying r with table scale and truncate to integer */
338 rt = _mm_mul_ps(r20,vftabscale);
339 vfitab = _mm_cvttps_epi32(rt);
341 vfeps = _mm_frcz_ps(rt);
343 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
345 twovfeps = _mm_add_ps(vfeps,vfeps);
346 vfitab = _mm_slli_epi32(vfitab,2);
348 /* CUBIC SPLINE TABLE ELECTROSTATICS */
349 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
350 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
351 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
352 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
353 _MM_TRANSPOSE4_PS(Y,F,G,H);
354 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
355 VV = _mm_macc_ps(vfeps,Fp,Y);
356 velec = _mm_mul_ps(qq20,VV);
357 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
358 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
360 /* Update potential sum for this i atom from the interaction with this j atom. */
361 velecsum = _mm_add_ps(velecsum,velec);
365 /* Update vectorial force */
366 fix2 = _mm_macc_ps(dx20,fscal,fix2);
367 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
368 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
370 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
371 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
372 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
374 /**************************
375 * CALCULATE INTERACTIONS *
376 **************************/
378 r30 = _mm_mul_ps(rsq30,rinv30);
380 /* Compute parameters for interactions between i and j atoms */
381 qq30 = _mm_mul_ps(iq3,jq0);
383 /* Calculate table index by multiplying r with table scale and truncate to integer */
384 rt = _mm_mul_ps(r30,vftabscale);
385 vfitab = _mm_cvttps_epi32(rt);
387 vfeps = _mm_frcz_ps(rt);
389 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
391 twovfeps = _mm_add_ps(vfeps,vfeps);
392 vfitab = _mm_slli_epi32(vfitab,2);
394 /* CUBIC SPLINE TABLE ELECTROSTATICS */
395 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
396 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
397 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
398 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
399 _MM_TRANSPOSE4_PS(Y,F,G,H);
400 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
401 VV = _mm_macc_ps(vfeps,Fp,Y);
402 velec = _mm_mul_ps(qq30,VV);
403 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
404 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
406 /* Update potential sum for this i atom from the interaction with this j atom. */
407 velecsum = _mm_add_ps(velecsum,velec);
411 /* Update vectorial force */
412 fix3 = _mm_macc_ps(dx30,fscal,fix3);
413 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
414 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
416 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
417 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
418 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
420 fjptrA = f+j_coord_offsetA;
421 fjptrB = f+j_coord_offsetB;
422 fjptrC = f+j_coord_offsetC;
423 fjptrD = f+j_coord_offsetD;
425 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
427 /* Inner loop uses 173 flops */
433 /* Get j neighbor index, and coordinate index */
434 jnrlistA = jjnr[jidx];
435 jnrlistB = jjnr[jidx+1];
436 jnrlistC = jjnr[jidx+2];
437 jnrlistD = jjnr[jidx+3];
438 /* Sign of each element will be negative for non-real atoms.
439 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
440 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
442 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
443 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
444 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
445 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
446 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
447 j_coord_offsetA = DIM*jnrA;
448 j_coord_offsetB = DIM*jnrB;
449 j_coord_offsetC = DIM*jnrC;
450 j_coord_offsetD = DIM*jnrD;
452 /* load j atom coordinates */
453 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
454 x+j_coord_offsetC,x+j_coord_offsetD,
457 /* Calculate displacement vector */
458 dx00 = _mm_sub_ps(ix0,jx0);
459 dy00 = _mm_sub_ps(iy0,jy0);
460 dz00 = _mm_sub_ps(iz0,jz0);
461 dx10 = _mm_sub_ps(ix1,jx0);
462 dy10 = _mm_sub_ps(iy1,jy0);
463 dz10 = _mm_sub_ps(iz1,jz0);
464 dx20 = _mm_sub_ps(ix2,jx0);
465 dy20 = _mm_sub_ps(iy2,jy0);
466 dz20 = _mm_sub_ps(iz2,jz0);
467 dx30 = _mm_sub_ps(ix3,jx0);
468 dy30 = _mm_sub_ps(iy3,jy0);
469 dz30 = _mm_sub_ps(iz3,jz0);
471 /* Calculate squared distance and things based on it */
472 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
473 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
474 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
475 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
477 rinv10 = gmx_mm_invsqrt_ps(rsq10);
478 rinv20 = gmx_mm_invsqrt_ps(rsq20);
479 rinv30 = gmx_mm_invsqrt_ps(rsq30);
481 rinvsq00 = gmx_mm_inv_ps(rsq00);
483 /* Load parameters for j particles */
484 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
485 charge+jnrC+0,charge+jnrD+0);
486 vdwjidx0A = 2*vdwtype[jnrA+0];
487 vdwjidx0B = 2*vdwtype[jnrB+0];
488 vdwjidx0C = 2*vdwtype[jnrC+0];
489 vdwjidx0D = 2*vdwtype[jnrD+0];
491 fjx0 = _mm_setzero_ps();
492 fjy0 = _mm_setzero_ps();
493 fjz0 = _mm_setzero_ps();
495 /**************************
496 * CALCULATE INTERACTIONS *
497 **************************/
499 /* Compute parameters for interactions between i and j atoms */
500 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
501 vdwparam+vdwioffset0+vdwjidx0B,
502 vdwparam+vdwioffset0+vdwjidx0C,
503 vdwparam+vdwioffset0+vdwjidx0D,
506 /* LENNARD-JONES DISPERSION/REPULSION */
508 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
509 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
510 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
511 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
512 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
514 /* Update potential sum for this i atom from the interaction with this j atom. */
515 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
516 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
520 fscal = _mm_andnot_ps(dummy_mask,fscal);
522 /* Update vectorial force */
523 fix0 = _mm_macc_ps(dx00,fscal,fix0);
524 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
525 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
527 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
528 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
529 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
531 /**************************
532 * CALCULATE INTERACTIONS *
533 **************************/
535 r10 = _mm_mul_ps(rsq10,rinv10);
536 r10 = _mm_andnot_ps(dummy_mask,r10);
538 /* Compute parameters for interactions between i and j atoms */
539 qq10 = _mm_mul_ps(iq1,jq0);
541 /* Calculate table index by multiplying r with table scale and truncate to integer */
542 rt = _mm_mul_ps(r10,vftabscale);
543 vfitab = _mm_cvttps_epi32(rt);
545 vfeps = _mm_frcz_ps(rt);
547 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
549 twovfeps = _mm_add_ps(vfeps,vfeps);
550 vfitab = _mm_slli_epi32(vfitab,2);
552 /* CUBIC SPLINE TABLE ELECTROSTATICS */
553 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
554 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
555 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
556 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
557 _MM_TRANSPOSE4_PS(Y,F,G,H);
558 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
559 VV = _mm_macc_ps(vfeps,Fp,Y);
560 velec = _mm_mul_ps(qq10,VV);
561 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
562 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
564 /* Update potential sum for this i atom from the interaction with this j atom. */
565 velec = _mm_andnot_ps(dummy_mask,velec);
566 velecsum = _mm_add_ps(velecsum,velec);
570 fscal = _mm_andnot_ps(dummy_mask,fscal);
572 /* Update vectorial force */
573 fix1 = _mm_macc_ps(dx10,fscal,fix1);
574 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
575 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
577 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
578 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
579 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
581 /**************************
582 * CALCULATE INTERACTIONS *
583 **************************/
585 r20 = _mm_mul_ps(rsq20,rinv20);
586 r20 = _mm_andnot_ps(dummy_mask,r20);
588 /* Compute parameters for interactions between i and j atoms */
589 qq20 = _mm_mul_ps(iq2,jq0);
591 /* Calculate table index by multiplying r with table scale and truncate to integer */
592 rt = _mm_mul_ps(r20,vftabscale);
593 vfitab = _mm_cvttps_epi32(rt);
595 vfeps = _mm_frcz_ps(rt);
597 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
599 twovfeps = _mm_add_ps(vfeps,vfeps);
600 vfitab = _mm_slli_epi32(vfitab,2);
602 /* CUBIC SPLINE TABLE ELECTROSTATICS */
603 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
604 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
605 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
606 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
607 _MM_TRANSPOSE4_PS(Y,F,G,H);
608 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
609 VV = _mm_macc_ps(vfeps,Fp,Y);
610 velec = _mm_mul_ps(qq20,VV);
611 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
612 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
614 /* Update potential sum for this i atom from the interaction with this j atom. */
615 velec = _mm_andnot_ps(dummy_mask,velec);
616 velecsum = _mm_add_ps(velecsum,velec);
620 fscal = _mm_andnot_ps(dummy_mask,fscal);
622 /* Update vectorial force */
623 fix2 = _mm_macc_ps(dx20,fscal,fix2);
624 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
625 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
627 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
628 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
629 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
631 /**************************
632 * CALCULATE INTERACTIONS *
633 **************************/
635 r30 = _mm_mul_ps(rsq30,rinv30);
636 r30 = _mm_andnot_ps(dummy_mask,r30);
638 /* Compute parameters for interactions between i and j atoms */
639 qq30 = _mm_mul_ps(iq3,jq0);
641 /* Calculate table index by multiplying r with table scale and truncate to integer */
642 rt = _mm_mul_ps(r30,vftabscale);
643 vfitab = _mm_cvttps_epi32(rt);
645 vfeps = _mm_frcz_ps(rt);
647 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
649 twovfeps = _mm_add_ps(vfeps,vfeps);
650 vfitab = _mm_slli_epi32(vfitab,2);
652 /* CUBIC SPLINE TABLE ELECTROSTATICS */
653 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
654 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
655 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
656 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
657 _MM_TRANSPOSE4_PS(Y,F,G,H);
658 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
659 VV = _mm_macc_ps(vfeps,Fp,Y);
660 velec = _mm_mul_ps(qq30,VV);
661 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
662 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
664 /* Update potential sum for this i atom from the interaction with this j atom. */
665 velec = _mm_andnot_ps(dummy_mask,velec);
666 velecsum = _mm_add_ps(velecsum,velec);
670 fscal = _mm_andnot_ps(dummy_mask,fscal);
672 /* Update vectorial force */
673 fix3 = _mm_macc_ps(dx30,fscal,fix3);
674 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
675 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
677 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
678 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
679 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
681 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
682 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
683 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
684 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
686 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
688 /* Inner loop uses 176 flops */
691 /* End of innermost loop */
693 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
694 f+i_coord_offset,fshift+i_shift_offset);
697 /* Update potential energies */
698 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
699 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
701 /* Increment number of inner iterations */
702 inneriter += j_index_end - j_index_start;
704 /* Outer loop uses 26 flops */
707 /* Increment number of outer iterations */
710 /* Update outer/inner flops */
712 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*176);
715 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_128_fma_single
716 * Electrostatics interaction: CubicSplineTable
717 * VdW interaction: LennardJones
718 * Geometry: Water4-Particle
719 * Calculate force/pot: Force
722 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_128_fma_single
723 (t_nblist * gmx_restrict nlist,
724 rvec * gmx_restrict xx,
725 rvec * gmx_restrict ff,
726 t_forcerec * gmx_restrict fr,
727 t_mdatoms * gmx_restrict mdatoms,
728 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
729 t_nrnb * gmx_restrict nrnb)
731 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
732 * just 0 for non-waters.
733 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
734 * jnr indices corresponding to data put in the four positions in the SIMD register.
736 int i_shift_offset,i_coord_offset,outeriter,inneriter;
737 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
738 int jnrA,jnrB,jnrC,jnrD;
739 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
740 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
741 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
743 real *shiftvec,*fshift,*x,*f;
744 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
746 __m128 fscal,rcutoff,rcutoff2,jidxall;
748 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
750 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
752 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
754 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
755 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
756 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
757 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
758 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
759 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
760 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
761 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
764 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
767 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
768 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
770 __m128i ifour = _mm_set1_epi32(4);
771 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
773 __m128 dummy_mask,cutoff_mask;
774 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
775 __m128 one = _mm_set1_ps(1.0);
776 __m128 two = _mm_set1_ps(2.0);
782 jindex = nlist->jindex;
784 shiftidx = nlist->shift;
786 shiftvec = fr->shift_vec[0];
787 fshift = fr->fshift[0];
788 facel = _mm_set1_ps(fr->epsfac);
789 charge = mdatoms->chargeA;
790 nvdwtype = fr->ntype;
792 vdwtype = mdatoms->typeA;
794 vftab = kernel_data->table_elec->data;
795 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
797 /* Setup water-specific parameters */
798 inr = nlist->iinr[0];
799 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
800 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
801 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
802 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
804 /* Avoid stupid compiler warnings */
805 jnrA = jnrB = jnrC = jnrD = 0;
814 for(iidx=0;iidx<4*DIM;iidx++)
819 /* Start outer loop over neighborlists */
820 for(iidx=0; iidx<nri; iidx++)
822 /* Load shift vector for this list */
823 i_shift_offset = DIM*shiftidx[iidx];
825 /* Load limits for loop over neighbors */
826 j_index_start = jindex[iidx];
827 j_index_end = jindex[iidx+1];
829 /* Get outer coordinate index */
831 i_coord_offset = DIM*inr;
833 /* Load i particle coords and add shift vector */
834 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
835 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
837 fix0 = _mm_setzero_ps();
838 fiy0 = _mm_setzero_ps();
839 fiz0 = _mm_setzero_ps();
840 fix1 = _mm_setzero_ps();
841 fiy1 = _mm_setzero_ps();
842 fiz1 = _mm_setzero_ps();
843 fix2 = _mm_setzero_ps();
844 fiy2 = _mm_setzero_ps();
845 fiz2 = _mm_setzero_ps();
846 fix3 = _mm_setzero_ps();
847 fiy3 = _mm_setzero_ps();
848 fiz3 = _mm_setzero_ps();
850 /* Start inner kernel loop */
851 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
854 /* Get j neighbor index, and coordinate index */
859 j_coord_offsetA = DIM*jnrA;
860 j_coord_offsetB = DIM*jnrB;
861 j_coord_offsetC = DIM*jnrC;
862 j_coord_offsetD = DIM*jnrD;
864 /* load j atom coordinates */
865 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
866 x+j_coord_offsetC,x+j_coord_offsetD,
869 /* Calculate displacement vector */
870 dx00 = _mm_sub_ps(ix0,jx0);
871 dy00 = _mm_sub_ps(iy0,jy0);
872 dz00 = _mm_sub_ps(iz0,jz0);
873 dx10 = _mm_sub_ps(ix1,jx0);
874 dy10 = _mm_sub_ps(iy1,jy0);
875 dz10 = _mm_sub_ps(iz1,jz0);
876 dx20 = _mm_sub_ps(ix2,jx0);
877 dy20 = _mm_sub_ps(iy2,jy0);
878 dz20 = _mm_sub_ps(iz2,jz0);
879 dx30 = _mm_sub_ps(ix3,jx0);
880 dy30 = _mm_sub_ps(iy3,jy0);
881 dz30 = _mm_sub_ps(iz3,jz0);
883 /* Calculate squared distance and things based on it */
884 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
885 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
886 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
887 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
889 rinv10 = gmx_mm_invsqrt_ps(rsq10);
890 rinv20 = gmx_mm_invsqrt_ps(rsq20);
891 rinv30 = gmx_mm_invsqrt_ps(rsq30);
893 rinvsq00 = gmx_mm_inv_ps(rsq00);
895 /* Load parameters for j particles */
896 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
897 charge+jnrC+0,charge+jnrD+0);
898 vdwjidx0A = 2*vdwtype[jnrA+0];
899 vdwjidx0B = 2*vdwtype[jnrB+0];
900 vdwjidx0C = 2*vdwtype[jnrC+0];
901 vdwjidx0D = 2*vdwtype[jnrD+0];
903 fjx0 = _mm_setzero_ps();
904 fjy0 = _mm_setzero_ps();
905 fjz0 = _mm_setzero_ps();
907 /**************************
908 * CALCULATE INTERACTIONS *
909 **************************/
911 /* Compute parameters for interactions between i and j atoms */
912 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
913 vdwparam+vdwioffset0+vdwjidx0B,
914 vdwparam+vdwioffset0+vdwjidx0C,
915 vdwparam+vdwioffset0+vdwjidx0D,
918 /* LENNARD-JONES DISPERSION/REPULSION */
920 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
921 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
925 /* Update vectorial force */
926 fix0 = _mm_macc_ps(dx00,fscal,fix0);
927 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
928 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
930 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
931 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
932 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
934 /**************************
935 * CALCULATE INTERACTIONS *
936 **************************/
938 r10 = _mm_mul_ps(rsq10,rinv10);
940 /* Compute parameters for interactions between i and j atoms */
941 qq10 = _mm_mul_ps(iq1,jq0);
943 /* Calculate table index by multiplying r with table scale and truncate to integer */
944 rt = _mm_mul_ps(r10,vftabscale);
945 vfitab = _mm_cvttps_epi32(rt);
947 vfeps = _mm_frcz_ps(rt);
949 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
951 twovfeps = _mm_add_ps(vfeps,vfeps);
952 vfitab = _mm_slli_epi32(vfitab,2);
954 /* CUBIC SPLINE TABLE ELECTROSTATICS */
955 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
956 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
957 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
958 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
959 _MM_TRANSPOSE4_PS(Y,F,G,H);
960 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
961 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
962 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
966 /* Update vectorial force */
967 fix1 = _mm_macc_ps(dx10,fscal,fix1);
968 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
969 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
971 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
972 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
973 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
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);
988 vfeps = _mm_frcz_ps(rt);
990 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
992 twovfeps = _mm_add_ps(vfeps,vfeps);
993 vfitab = _mm_slli_epi32(vfitab,2);
995 /* CUBIC SPLINE TABLE ELECTROSTATICS */
996 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
997 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
998 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
999 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1000 _MM_TRANSPOSE4_PS(Y,F,G,H);
1001 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1002 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1003 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1007 /* Update vectorial force */
1008 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1009 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1010 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1012 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1013 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1014 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1016 /**************************
1017 * CALCULATE INTERACTIONS *
1018 **************************/
1020 r30 = _mm_mul_ps(rsq30,rinv30);
1022 /* Compute parameters for interactions between i and j atoms */
1023 qq30 = _mm_mul_ps(iq3,jq0);
1025 /* Calculate table index by multiplying r with table scale and truncate to integer */
1026 rt = _mm_mul_ps(r30,vftabscale);
1027 vfitab = _mm_cvttps_epi32(rt);
1029 vfeps = _mm_frcz_ps(rt);
1031 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1033 twovfeps = _mm_add_ps(vfeps,vfeps);
1034 vfitab = _mm_slli_epi32(vfitab,2);
1036 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1037 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1038 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1039 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1040 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1041 _MM_TRANSPOSE4_PS(Y,F,G,H);
1042 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1043 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1044 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1048 /* Update vectorial force */
1049 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1050 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1051 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1053 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1054 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1055 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1057 fjptrA = f+j_coord_offsetA;
1058 fjptrB = f+j_coord_offsetB;
1059 fjptrC = f+j_coord_offsetC;
1060 fjptrD = f+j_coord_offsetD;
1062 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1064 /* Inner loop uses 156 flops */
1067 if(jidx<j_index_end)
1070 /* Get j neighbor index, and coordinate index */
1071 jnrlistA = jjnr[jidx];
1072 jnrlistB = jjnr[jidx+1];
1073 jnrlistC = jjnr[jidx+2];
1074 jnrlistD = jjnr[jidx+3];
1075 /* Sign of each element will be negative for non-real atoms.
1076 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1077 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1079 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1080 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1081 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1082 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1083 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1084 j_coord_offsetA = DIM*jnrA;
1085 j_coord_offsetB = DIM*jnrB;
1086 j_coord_offsetC = DIM*jnrC;
1087 j_coord_offsetD = DIM*jnrD;
1089 /* load j atom coordinates */
1090 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1091 x+j_coord_offsetC,x+j_coord_offsetD,
1094 /* Calculate displacement vector */
1095 dx00 = _mm_sub_ps(ix0,jx0);
1096 dy00 = _mm_sub_ps(iy0,jy0);
1097 dz00 = _mm_sub_ps(iz0,jz0);
1098 dx10 = _mm_sub_ps(ix1,jx0);
1099 dy10 = _mm_sub_ps(iy1,jy0);
1100 dz10 = _mm_sub_ps(iz1,jz0);
1101 dx20 = _mm_sub_ps(ix2,jx0);
1102 dy20 = _mm_sub_ps(iy2,jy0);
1103 dz20 = _mm_sub_ps(iz2,jz0);
1104 dx30 = _mm_sub_ps(ix3,jx0);
1105 dy30 = _mm_sub_ps(iy3,jy0);
1106 dz30 = _mm_sub_ps(iz3,jz0);
1108 /* Calculate squared distance and things based on it */
1109 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1110 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1111 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1112 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1114 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1115 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1116 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1118 rinvsq00 = gmx_mm_inv_ps(rsq00);
1120 /* Load parameters for j particles */
1121 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1122 charge+jnrC+0,charge+jnrD+0);
1123 vdwjidx0A = 2*vdwtype[jnrA+0];
1124 vdwjidx0B = 2*vdwtype[jnrB+0];
1125 vdwjidx0C = 2*vdwtype[jnrC+0];
1126 vdwjidx0D = 2*vdwtype[jnrD+0];
1128 fjx0 = _mm_setzero_ps();
1129 fjy0 = _mm_setzero_ps();
1130 fjz0 = _mm_setzero_ps();
1132 /**************************
1133 * CALCULATE INTERACTIONS *
1134 **************************/
1136 /* Compute parameters for interactions between i and j atoms */
1137 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1138 vdwparam+vdwioffset0+vdwjidx0B,
1139 vdwparam+vdwioffset0+vdwjidx0C,
1140 vdwparam+vdwioffset0+vdwjidx0D,
1143 /* LENNARD-JONES DISPERSION/REPULSION */
1145 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1146 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1150 fscal = _mm_andnot_ps(dummy_mask,fscal);
1152 /* Update vectorial force */
1153 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1154 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1155 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1157 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1158 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1159 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1161 /**************************
1162 * CALCULATE INTERACTIONS *
1163 **************************/
1165 r10 = _mm_mul_ps(rsq10,rinv10);
1166 r10 = _mm_andnot_ps(dummy_mask,r10);
1168 /* Compute parameters for interactions between i and j atoms */
1169 qq10 = _mm_mul_ps(iq1,jq0);
1171 /* Calculate table index by multiplying r with table scale and truncate to integer */
1172 rt = _mm_mul_ps(r10,vftabscale);
1173 vfitab = _mm_cvttps_epi32(rt);
1175 vfeps = _mm_frcz_ps(rt);
1177 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1179 twovfeps = _mm_add_ps(vfeps,vfeps);
1180 vfitab = _mm_slli_epi32(vfitab,2);
1182 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1183 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1184 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1185 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1186 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1187 _MM_TRANSPOSE4_PS(Y,F,G,H);
1188 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1189 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1190 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1194 fscal = _mm_andnot_ps(dummy_mask,fscal);
1196 /* Update vectorial force */
1197 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1198 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1199 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1201 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1202 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1203 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1205 /**************************
1206 * CALCULATE INTERACTIONS *
1207 **************************/
1209 r20 = _mm_mul_ps(rsq20,rinv20);
1210 r20 = _mm_andnot_ps(dummy_mask,r20);
1212 /* Compute parameters for interactions between i and j atoms */
1213 qq20 = _mm_mul_ps(iq2,jq0);
1215 /* Calculate table index by multiplying r with table scale and truncate to integer */
1216 rt = _mm_mul_ps(r20,vftabscale);
1217 vfitab = _mm_cvttps_epi32(rt);
1219 vfeps = _mm_frcz_ps(rt);
1221 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1223 twovfeps = _mm_add_ps(vfeps,vfeps);
1224 vfitab = _mm_slli_epi32(vfitab,2);
1226 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1227 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1228 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1229 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1230 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1231 _MM_TRANSPOSE4_PS(Y,F,G,H);
1232 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1233 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1234 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1238 fscal = _mm_andnot_ps(dummy_mask,fscal);
1240 /* Update vectorial force */
1241 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1242 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1243 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1245 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1246 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1247 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1249 /**************************
1250 * CALCULATE INTERACTIONS *
1251 **************************/
1253 r30 = _mm_mul_ps(rsq30,rinv30);
1254 r30 = _mm_andnot_ps(dummy_mask,r30);
1256 /* Compute parameters for interactions between i and j atoms */
1257 qq30 = _mm_mul_ps(iq3,jq0);
1259 /* Calculate table index by multiplying r with table scale and truncate to integer */
1260 rt = _mm_mul_ps(r30,vftabscale);
1261 vfitab = _mm_cvttps_epi32(rt);
1263 vfeps = _mm_frcz_ps(rt);
1265 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1267 twovfeps = _mm_add_ps(vfeps,vfeps);
1268 vfitab = _mm_slli_epi32(vfitab,2);
1270 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1271 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1272 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1273 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1274 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1275 _MM_TRANSPOSE4_PS(Y,F,G,H);
1276 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1277 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1278 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1282 fscal = _mm_andnot_ps(dummy_mask,fscal);
1284 /* Update vectorial force */
1285 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1286 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1287 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1289 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1290 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1291 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1293 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1294 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1295 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1296 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1298 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1300 /* Inner loop uses 159 flops */
1303 /* End of innermost loop */
1305 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1306 f+i_coord_offset,fshift+i_shift_offset);
1308 /* Increment number of inner iterations */
1309 inneriter += j_index_end - j_index_start;
1311 /* Outer loop uses 24 flops */
1314 /* Increment number of outer iterations */
1317 /* Update outer/inner flops */
1319 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*159);