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.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
48 #include "kernelutil_x86_avx_128_fma_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_128_fma_single
52 * Electrostatics interaction: CubicSplineTable
53 * VdW interaction: LennardJones
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_128_fma_single
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m128 fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
92 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
94 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
95 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
96 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
97 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
100 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
103 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
104 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
106 __m128i ifour = _mm_set1_epi32(4);
107 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
109 __m128 dummy_mask,cutoff_mask;
110 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
111 __m128 one = _mm_set1_ps(1.0);
112 __m128 two = _mm_set1_ps(2.0);
118 jindex = nlist->jindex;
120 shiftidx = nlist->shift;
122 shiftvec = fr->shift_vec[0];
123 fshift = fr->fshift[0];
124 facel = _mm_set1_ps(fr->epsfac);
125 charge = mdatoms->chargeA;
126 nvdwtype = fr->ntype;
128 vdwtype = mdatoms->typeA;
130 vftab = kernel_data->table_elec->data;
131 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
133 /* Setup water-specific parameters */
134 inr = nlist->iinr[0];
135 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
136 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
137 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
138 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
140 /* Avoid stupid compiler warnings */
141 jnrA = jnrB = jnrC = jnrD = 0;
150 for(iidx=0;iidx<4*DIM;iidx++)
155 /* Start outer loop over neighborlists */
156 for(iidx=0; iidx<nri; iidx++)
158 /* Load shift vector for this list */
159 i_shift_offset = DIM*shiftidx[iidx];
161 /* Load limits for loop over neighbors */
162 j_index_start = jindex[iidx];
163 j_index_end = jindex[iidx+1];
165 /* Get outer coordinate index */
167 i_coord_offset = DIM*inr;
169 /* Load i particle coords and add shift vector */
170 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
171 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
173 fix0 = _mm_setzero_ps();
174 fiy0 = _mm_setzero_ps();
175 fiz0 = _mm_setzero_ps();
176 fix1 = _mm_setzero_ps();
177 fiy1 = _mm_setzero_ps();
178 fiz1 = _mm_setzero_ps();
179 fix2 = _mm_setzero_ps();
180 fiy2 = _mm_setzero_ps();
181 fiz2 = _mm_setzero_ps();
182 fix3 = _mm_setzero_ps();
183 fiy3 = _mm_setzero_ps();
184 fiz3 = _mm_setzero_ps();
186 /* Reset potential sums */
187 velecsum = _mm_setzero_ps();
188 vvdwsum = _mm_setzero_ps();
190 /* Start inner kernel loop */
191 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
194 /* Get j neighbor index, and coordinate index */
199 j_coord_offsetA = DIM*jnrA;
200 j_coord_offsetB = DIM*jnrB;
201 j_coord_offsetC = DIM*jnrC;
202 j_coord_offsetD = DIM*jnrD;
204 /* load j atom coordinates */
205 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
206 x+j_coord_offsetC,x+j_coord_offsetD,
209 /* Calculate displacement vector */
210 dx00 = _mm_sub_ps(ix0,jx0);
211 dy00 = _mm_sub_ps(iy0,jy0);
212 dz00 = _mm_sub_ps(iz0,jz0);
213 dx10 = _mm_sub_ps(ix1,jx0);
214 dy10 = _mm_sub_ps(iy1,jy0);
215 dz10 = _mm_sub_ps(iz1,jz0);
216 dx20 = _mm_sub_ps(ix2,jx0);
217 dy20 = _mm_sub_ps(iy2,jy0);
218 dz20 = _mm_sub_ps(iz2,jz0);
219 dx30 = _mm_sub_ps(ix3,jx0);
220 dy30 = _mm_sub_ps(iy3,jy0);
221 dz30 = _mm_sub_ps(iz3,jz0);
223 /* Calculate squared distance and things based on it */
224 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
225 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
226 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
227 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
229 rinv10 = gmx_mm_invsqrt_ps(rsq10);
230 rinv20 = gmx_mm_invsqrt_ps(rsq20);
231 rinv30 = gmx_mm_invsqrt_ps(rsq30);
233 rinvsq00 = gmx_mm_inv_ps(rsq00);
235 /* Load parameters for j particles */
236 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
237 charge+jnrC+0,charge+jnrD+0);
238 vdwjidx0A = 2*vdwtype[jnrA+0];
239 vdwjidx0B = 2*vdwtype[jnrB+0];
240 vdwjidx0C = 2*vdwtype[jnrC+0];
241 vdwjidx0D = 2*vdwtype[jnrD+0];
243 fjx0 = _mm_setzero_ps();
244 fjy0 = _mm_setzero_ps();
245 fjz0 = _mm_setzero_ps();
247 /**************************
248 * CALCULATE INTERACTIONS *
249 **************************/
251 /* Compute parameters for interactions between i and j atoms */
252 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
253 vdwparam+vdwioffset0+vdwjidx0B,
254 vdwparam+vdwioffset0+vdwjidx0C,
255 vdwparam+vdwioffset0+vdwjidx0D,
258 /* LENNARD-JONES DISPERSION/REPULSION */
260 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
261 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
262 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
263 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
264 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
266 /* Update potential sum for this i atom from the interaction with this j atom. */
267 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
271 /* Update vectorial force */
272 fix0 = _mm_macc_ps(dx00,fscal,fix0);
273 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
274 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
276 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
277 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
278 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
280 /**************************
281 * CALCULATE INTERACTIONS *
282 **************************/
284 r10 = _mm_mul_ps(rsq10,rinv10);
286 /* Compute parameters for interactions between i and j atoms */
287 qq10 = _mm_mul_ps(iq1,jq0);
289 /* Calculate table index by multiplying r with table scale and truncate to integer */
290 rt = _mm_mul_ps(r10,vftabscale);
291 vfitab = _mm_cvttps_epi32(rt);
293 vfeps = _mm_frcz_ps(rt);
295 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
297 twovfeps = _mm_add_ps(vfeps,vfeps);
298 vfitab = _mm_slli_epi32(vfitab,2);
300 /* CUBIC SPLINE TABLE ELECTROSTATICS */
301 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
302 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
303 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
304 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
305 _MM_TRANSPOSE4_PS(Y,F,G,H);
306 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
307 VV = _mm_macc_ps(vfeps,Fp,Y);
308 velec = _mm_mul_ps(qq10,VV);
309 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
310 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
312 /* Update potential sum for this i atom from the interaction with this j atom. */
313 velecsum = _mm_add_ps(velecsum,velec);
317 /* Update vectorial force */
318 fix1 = _mm_macc_ps(dx10,fscal,fix1);
319 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
320 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
322 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
323 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
324 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
326 /**************************
327 * CALCULATE INTERACTIONS *
328 **************************/
330 r20 = _mm_mul_ps(rsq20,rinv20);
332 /* Compute parameters for interactions between i and j atoms */
333 qq20 = _mm_mul_ps(iq2,jq0);
335 /* Calculate table index by multiplying r with table scale and truncate to integer */
336 rt = _mm_mul_ps(r20,vftabscale);
337 vfitab = _mm_cvttps_epi32(rt);
339 vfeps = _mm_frcz_ps(rt);
341 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
343 twovfeps = _mm_add_ps(vfeps,vfeps);
344 vfitab = _mm_slli_epi32(vfitab,2);
346 /* CUBIC SPLINE TABLE ELECTROSTATICS */
347 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
348 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
349 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
350 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
351 _MM_TRANSPOSE4_PS(Y,F,G,H);
352 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
353 VV = _mm_macc_ps(vfeps,Fp,Y);
354 velec = _mm_mul_ps(qq20,VV);
355 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
356 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
358 /* Update potential sum for this i atom from the interaction with this j atom. */
359 velecsum = _mm_add_ps(velecsum,velec);
363 /* Update vectorial force */
364 fix2 = _mm_macc_ps(dx20,fscal,fix2);
365 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
366 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
368 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
369 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
370 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
372 /**************************
373 * CALCULATE INTERACTIONS *
374 **************************/
376 r30 = _mm_mul_ps(rsq30,rinv30);
378 /* Compute parameters for interactions between i and j atoms */
379 qq30 = _mm_mul_ps(iq3,jq0);
381 /* Calculate table index by multiplying r with table scale and truncate to integer */
382 rt = _mm_mul_ps(r30,vftabscale);
383 vfitab = _mm_cvttps_epi32(rt);
385 vfeps = _mm_frcz_ps(rt);
387 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
389 twovfeps = _mm_add_ps(vfeps,vfeps);
390 vfitab = _mm_slli_epi32(vfitab,2);
392 /* CUBIC SPLINE TABLE ELECTROSTATICS */
393 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
394 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
395 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
396 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
397 _MM_TRANSPOSE4_PS(Y,F,G,H);
398 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
399 VV = _mm_macc_ps(vfeps,Fp,Y);
400 velec = _mm_mul_ps(qq30,VV);
401 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
402 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
404 /* Update potential sum for this i atom from the interaction with this j atom. */
405 velecsum = _mm_add_ps(velecsum,velec);
409 /* Update vectorial force */
410 fix3 = _mm_macc_ps(dx30,fscal,fix3);
411 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
412 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
414 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
415 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
416 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
418 fjptrA = f+j_coord_offsetA;
419 fjptrB = f+j_coord_offsetB;
420 fjptrC = f+j_coord_offsetC;
421 fjptrD = f+j_coord_offsetD;
423 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
425 /* Inner loop uses 173 flops */
431 /* Get j neighbor index, and coordinate index */
432 jnrlistA = jjnr[jidx];
433 jnrlistB = jjnr[jidx+1];
434 jnrlistC = jjnr[jidx+2];
435 jnrlistD = jjnr[jidx+3];
436 /* Sign of each element will be negative for non-real atoms.
437 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
438 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
440 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
441 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
442 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
443 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
444 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
445 j_coord_offsetA = DIM*jnrA;
446 j_coord_offsetB = DIM*jnrB;
447 j_coord_offsetC = DIM*jnrC;
448 j_coord_offsetD = DIM*jnrD;
450 /* load j atom coordinates */
451 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
452 x+j_coord_offsetC,x+j_coord_offsetD,
455 /* Calculate displacement vector */
456 dx00 = _mm_sub_ps(ix0,jx0);
457 dy00 = _mm_sub_ps(iy0,jy0);
458 dz00 = _mm_sub_ps(iz0,jz0);
459 dx10 = _mm_sub_ps(ix1,jx0);
460 dy10 = _mm_sub_ps(iy1,jy0);
461 dz10 = _mm_sub_ps(iz1,jz0);
462 dx20 = _mm_sub_ps(ix2,jx0);
463 dy20 = _mm_sub_ps(iy2,jy0);
464 dz20 = _mm_sub_ps(iz2,jz0);
465 dx30 = _mm_sub_ps(ix3,jx0);
466 dy30 = _mm_sub_ps(iy3,jy0);
467 dz30 = _mm_sub_ps(iz3,jz0);
469 /* Calculate squared distance and things based on it */
470 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
471 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
472 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
473 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
475 rinv10 = gmx_mm_invsqrt_ps(rsq10);
476 rinv20 = gmx_mm_invsqrt_ps(rsq20);
477 rinv30 = gmx_mm_invsqrt_ps(rsq30);
479 rinvsq00 = gmx_mm_inv_ps(rsq00);
481 /* Load parameters for j particles */
482 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
483 charge+jnrC+0,charge+jnrD+0);
484 vdwjidx0A = 2*vdwtype[jnrA+0];
485 vdwjidx0B = 2*vdwtype[jnrB+0];
486 vdwjidx0C = 2*vdwtype[jnrC+0];
487 vdwjidx0D = 2*vdwtype[jnrD+0];
489 fjx0 = _mm_setzero_ps();
490 fjy0 = _mm_setzero_ps();
491 fjz0 = _mm_setzero_ps();
493 /**************************
494 * CALCULATE INTERACTIONS *
495 **************************/
497 /* Compute parameters for interactions between i and j atoms */
498 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
499 vdwparam+vdwioffset0+vdwjidx0B,
500 vdwparam+vdwioffset0+vdwjidx0C,
501 vdwparam+vdwioffset0+vdwjidx0D,
504 /* LENNARD-JONES DISPERSION/REPULSION */
506 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
507 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
508 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
509 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
510 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
512 /* Update potential sum for this i atom from the interaction with this j atom. */
513 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
514 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
518 fscal = _mm_andnot_ps(dummy_mask,fscal);
520 /* Update vectorial force */
521 fix0 = _mm_macc_ps(dx00,fscal,fix0);
522 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
523 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
525 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
526 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
527 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
529 /**************************
530 * CALCULATE INTERACTIONS *
531 **************************/
533 r10 = _mm_mul_ps(rsq10,rinv10);
534 r10 = _mm_andnot_ps(dummy_mask,r10);
536 /* Compute parameters for interactions between i and j atoms */
537 qq10 = _mm_mul_ps(iq1,jq0);
539 /* Calculate table index by multiplying r with table scale and truncate to integer */
540 rt = _mm_mul_ps(r10,vftabscale);
541 vfitab = _mm_cvttps_epi32(rt);
543 vfeps = _mm_frcz_ps(rt);
545 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
547 twovfeps = _mm_add_ps(vfeps,vfeps);
548 vfitab = _mm_slli_epi32(vfitab,2);
550 /* CUBIC SPLINE TABLE ELECTROSTATICS */
551 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
552 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
553 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
554 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
555 _MM_TRANSPOSE4_PS(Y,F,G,H);
556 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
557 VV = _mm_macc_ps(vfeps,Fp,Y);
558 velec = _mm_mul_ps(qq10,VV);
559 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
560 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
562 /* Update potential sum for this i atom from the interaction with this j atom. */
563 velec = _mm_andnot_ps(dummy_mask,velec);
564 velecsum = _mm_add_ps(velecsum,velec);
568 fscal = _mm_andnot_ps(dummy_mask,fscal);
570 /* Update vectorial force */
571 fix1 = _mm_macc_ps(dx10,fscal,fix1);
572 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
573 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
575 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
576 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
577 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
579 /**************************
580 * CALCULATE INTERACTIONS *
581 **************************/
583 r20 = _mm_mul_ps(rsq20,rinv20);
584 r20 = _mm_andnot_ps(dummy_mask,r20);
586 /* Compute parameters for interactions between i and j atoms */
587 qq20 = _mm_mul_ps(iq2,jq0);
589 /* Calculate table index by multiplying r with table scale and truncate to integer */
590 rt = _mm_mul_ps(r20,vftabscale);
591 vfitab = _mm_cvttps_epi32(rt);
593 vfeps = _mm_frcz_ps(rt);
595 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
597 twovfeps = _mm_add_ps(vfeps,vfeps);
598 vfitab = _mm_slli_epi32(vfitab,2);
600 /* CUBIC SPLINE TABLE ELECTROSTATICS */
601 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
602 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
603 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
604 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
605 _MM_TRANSPOSE4_PS(Y,F,G,H);
606 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
607 VV = _mm_macc_ps(vfeps,Fp,Y);
608 velec = _mm_mul_ps(qq20,VV);
609 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
610 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
612 /* Update potential sum for this i atom from the interaction with this j atom. */
613 velec = _mm_andnot_ps(dummy_mask,velec);
614 velecsum = _mm_add_ps(velecsum,velec);
618 fscal = _mm_andnot_ps(dummy_mask,fscal);
620 /* Update vectorial force */
621 fix2 = _mm_macc_ps(dx20,fscal,fix2);
622 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
623 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
625 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
626 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
627 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
629 /**************************
630 * CALCULATE INTERACTIONS *
631 **************************/
633 r30 = _mm_mul_ps(rsq30,rinv30);
634 r30 = _mm_andnot_ps(dummy_mask,r30);
636 /* Compute parameters for interactions between i and j atoms */
637 qq30 = _mm_mul_ps(iq3,jq0);
639 /* Calculate table index by multiplying r with table scale and truncate to integer */
640 rt = _mm_mul_ps(r30,vftabscale);
641 vfitab = _mm_cvttps_epi32(rt);
643 vfeps = _mm_frcz_ps(rt);
645 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
647 twovfeps = _mm_add_ps(vfeps,vfeps);
648 vfitab = _mm_slli_epi32(vfitab,2);
650 /* CUBIC SPLINE TABLE ELECTROSTATICS */
651 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
652 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
653 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
654 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
655 _MM_TRANSPOSE4_PS(Y,F,G,H);
656 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
657 VV = _mm_macc_ps(vfeps,Fp,Y);
658 velec = _mm_mul_ps(qq30,VV);
659 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
660 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
662 /* Update potential sum for this i atom from the interaction with this j atom. */
663 velec = _mm_andnot_ps(dummy_mask,velec);
664 velecsum = _mm_add_ps(velecsum,velec);
668 fscal = _mm_andnot_ps(dummy_mask,fscal);
670 /* Update vectorial force */
671 fix3 = _mm_macc_ps(dx30,fscal,fix3);
672 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
673 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
675 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
676 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
677 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
679 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
680 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
681 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
682 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
684 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
686 /* Inner loop uses 176 flops */
689 /* End of innermost loop */
691 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
692 f+i_coord_offset,fshift+i_shift_offset);
695 /* Update potential energies */
696 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
697 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
699 /* Increment number of inner iterations */
700 inneriter += j_index_end - j_index_start;
702 /* Outer loop uses 26 flops */
705 /* Increment number of outer iterations */
708 /* Update outer/inner flops */
710 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*176);
713 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_128_fma_single
714 * Electrostatics interaction: CubicSplineTable
715 * VdW interaction: LennardJones
716 * Geometry: Water4-Particle
717 * Calculate force/pot: Force
720 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_128_fma_single
721 (t_nblist * gmx_restrict nlist,
722 rvec * gmx_restrict xx,
723 rvec * gmx_restrict ff,
724 t_forcerec * gmx_restrict fr,
725 t_mdatoms * gmx_restrict mdatoms,
726 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
727 t_nrnb * gmx_restrict nrnb)
729 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
730 * just 0 for non-waters.
731 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
732 * jnr indices corresponding to data put in the four positions in the SIMD register.
734 int i_shift_offset,i_coord_offset,outeriter,inneriter;
735 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
736 int jnrA,jnrB,jnrC,jnrD;
737 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
738 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
739 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
741 real *shiftvec,*fshift,*x,*f;
742 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
744 __m128 fscal,rcutoff,rcutoff2,jidxall;
746 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
748 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
750 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
752 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
753 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
754 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
755 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
756 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
757 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
758 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
759 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
762 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
765 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
766 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
768 __m128i ifour = _mm_set1_epi32(4);
769 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
771 __m128 dummy_mask,cutoff_mask;
772 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
773 __m128 one = _mm_set1_ps(1.0);
774 __m128 two = _mm_set1_ps(2.0);
780 jindex = nlist->jindex;
782 shiftidx = nlist->shift;
784 shiftvec = fr->shift_vec[0];
785 fshift = fr->fshift[0];
786 facel = _mm_set1_ps(fr->epsfac);
787 charge = mdatoms->chargeA;
788 nvdwtype = fr->ntype;
790 vdwtype = mdatoms->typeA;
792 vftab = kernel_data->table_elec->data;
793 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
795 /* Setup water-specific parameters */
796 inr = nlist->iinr[0];
797 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
798 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
799 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
800 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
802 /* Avoid stupid compiler warnings */
803 jnrA = jnrB = jnrC = jnrD = 0;
812 for(iidx=0;iidx<4*DIM;iidx++)
817 /* Start outer loop over neighborlists */
818 for(iidx=0; iidx<nri; iidx++)
820 /* Load shift vector for this list */
821 i_shift_offset = DIM*shiftidx[iidx];
823 /* Load limits for loop over neighbors */
824 j_index_start = jindex[iidx];
825 j_index_end = jindex[iidx+1];
827 /* Get outer coordinate index */
829 i_coord_offset = DIM*inr;
831 /* Load i particle coords and add shift vector */
832 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
833 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
835 fix0 = _mm_setzero_ps();
836 fiy0 = _mm_setzero_ps();
837 fiz0 = _mm_setzero_ps();
838 fix1 = _mm_setzero_ps();
839 fiy1 = _mm_setzero_ps();
840 fiz1 = _mm_setzero_ps();
841 fix2 = _mm_setzero_ps();
842 fiy2 = _mm_setzero_ps();
843 fiz2 = _mm_setzero_ps();
844 fix3 = _mm_setzero_ps();
845 fiy3 = _mm_setzero_ps();
846 fiz3 = _mm_setzero_ps();
848 /* Start inner kernel loop */
849 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
852 /* Get j neighbor index, and coordinate index */
857 j_coord_offsetA = DIM*jnrA;
858 j_coord_offsetB = DIM*jnrB;
859 j_coord_offsetC = DIM*jnrC;
860 j_coord_offsetD = DIM*jnrD;
862 /* load j atom coordinates */
863 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
864 x+j_coord_offsetC,x+j_coord_offsetD,
867 /* Calculate displacement vector */
868 dx00 = _mm_sub_ps(ix0,jx0);
869 dy00 = _mm_sub_ps(iy0,jy0);
870 dz00 = _mm_sub_ps(iz0,jz0);
871 dx10 = _mm_sub_ps(ix1,jx0);
872 dy10 = _mm_sub_ps(iy1,jy0);
873 dz10 = _mm_sub_ps(iz1,jz0);
874 dx20 = _mm_sub_ps(ix2,jx0);
875 dy20 = _mm_sub_ps(iy2,jy0);
876 dz20 = _mm_sub_ps(iz2,jz0);
877 dx30 = _mm_sub_ps(ix3,jx0);
878 dy30 = _mm_sub_ps(iy3,jy0);
879 dz30 = _mm_sub_ps(iz3,jz0);
881 /* Calculate squared distance and things based on it */
882 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
883 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
884 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
885 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
887 rinv10 = gmx_mm_invsqrt_ps(rsq10);
888 rinv20 = gmx_mm_invsqrt_ps(rsq20);
889 rinv30 = gmx_mm_invsqrt_ps(rsq30);
891 rinvsq00 = gmx_mm_inv_ps(rsq00);
893 /* Load parameters for j particles */
894 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
895 charge+jnrC+0,charge+jnrD+0);
896 vdwjidx0A = 2*vdwtype[jnrA+0];
897 vdwjidx0B = 2*vdwtype[jnrB+0];
898 vdwjidx0C = 2*vdwtype[jnrC+0];
899 vdwjidx0D = 2*vdwtype[jnrD+0];
901 fjx0 = _mm_setzero_ps();
902 fjy0 = _mm_setzero_ps();
903 fjz0 = _mm_setzero_ps();
905 /**************************
906 * CALCULATE INTERACTIONS *
907 **************************/
909 /* Compute parameters for interactions between i and j atoms */
910 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
911 vdwparam+vdwioffset0+vdwjidx0B,
912 vdwparam+vdwioffset0+vdwjidx0C,
913 vdwparam+vdwioffset0+vdwjidx0D,
916 /* LENNARD-JONES DISPERSION/REPULSION */
918 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
919 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
923 /* Update vectorial force */
924 fix0 = _mm_macc_ps(dx00,fscal,fix0);
925 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
926 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
928 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
929 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
930 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
932 /**************************
933 * CALCULATE INTERACTIONS *
934 **************************/
936 r10 = _mm_mul_ps(rsq10,rinv10);
938 /* Compute parameters for interactions between i and j atoms */
939 qq10 = _mm_mul_ps(iq1,jq0);
941 /* Calculate table index by multiplying r with table scale and truncate to integer */
942 rt = _mm_mul_ps(r10,vftabscale);
943 vfitab = _mm_cvttps_epi32(rt);
945 vfeps = _mm_frcz_ps(rt);
947 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
949 twovfeps = _mm_add_ps(vfeps,vfeps);
950 vfitab = _mm_slli_epi32(vfitab,2);
952 /* CUBIC SPLINE TABLE ELECTROSTATICS */
953 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
954 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
955 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
956 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
957 _MM_TRANSPOSE4_PS(Y,F,G,H);
958 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
959 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
960 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
964 /* Update vectorial force */
965 fix1 = _mm_macc_ps(dx10,fscal,fix1);
966 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
967 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
969 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
970 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
971 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
973 /**************************
974 * CALCULATE INTERACTIONS *
975 **************************/
977 r20 = _mm_mul_ps(rsq20,rinv20);
979 /* Compute parameters for interactions between i and j atoms */
980 qq20 = _mm_mul_ps(iq2,jq0);
982 /* Calculate table index by multiplying r with table scale and truncate to integer */
983 rt = _mm_mul_ps(r20,vftabscale);
984 vfitab = _mm_cvttps_epi32(rt);
986 vfeps = _mm_frcz_ps(rt);
988 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
990 twovfeps = _mm_add_ps(vfeps,vfeps);
991 vfitab = _mm_slli_epi32(vfitab,2);
993 /* CUBIC SPLINE TABLE ELECTROSTATICS */
994 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
995 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
996 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
997 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
998 _MM_TRANSPOSE4_PS(Y,F,G,H);
999 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1000 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1001 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1005 /* Update vectorial force */
1006 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1007 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1008 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1010 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1011 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1012 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1014 /**************************
1015 * CALCULATE INTERACTIONS *
1016 **************************/
1018 r30 = _mm_mul_ps(rsq30,rinv30);
1020 /* Compute parameters for interactions between i and j atoms */
1021 qq30 = _mm_mul_ps(iq3,jq0);
1023 /* Calculate table index by multiplying r with table scale and truncate to integer */
1024 rt = _mm_mul_ps(r30,vftabscale);
1025 vfitab = _mm_cvttps_epi32(rt);
1027 vfeps = _mm_frcz_ps(rt);
1029 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1031 twovfeps = _mm_add_ps(vfeps,vfeps);
1032 vfitab = _mm_slli_epi32(vfitab,2);
1034 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1035 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1036 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1037 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1038 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1039 _MM_TRANSPOSE4_PS(Y,F,G,H);
1040 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1041 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1042 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1046 /* Update vectorial force */
1047 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1048 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1049 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1051 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1052 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1053 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1055 fjptrA = f+j_coord_offsetA;
1056 fjptrB = f+j_coord_offsetB;
1057 fjptrC = f+j_coord_offsetC;
1058 fjptrD = f+j_coord_offsetD;
1060 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1062 /* Inner loop uses 156 flops */
1065 if(jidx<j_index_end)
1068 /* Get j neighbor index, and coordinate index */
1069 jnrlistA = jjnr[jidx];
1070 jnrlistB = jjnr[jidx+1];
1071 jnrlistC = jjnr[jidx+2];
1072 jnrlistD = jjnr[jidx+3];
1073 /* Sign of each element will be negative for non-real atoms.
1074 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1075 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1077 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1078 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1079 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1080 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1081 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1082 j_coord_offsetA = DIM*jnrA;
1083 j_coord_offsetB = DIM*jnrB;
1084 j_coord_offsetC = DIM*jnrC;
1085 j_coord_offsetD = DIM*jnrD;
1087 /* load j atom coordinates */
1088 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1089 x+j_coord_offsetC,x+j_coord_offsetD,
1092 /* Calculate displacement vector */
1093 dx00 = _mm_sub_ps(ix0,jx0);
1094 dy00 = _mm_sub_ps(iy0,jy0);
1095 dz00 = _mm_sub_ps(iz0,jz0);
1096 dx10 = _mm_sub_ps(ix1,jx0);
1097 dy10 = _mm_sub_ps(iy1,jy0);
1098 dz10 = _mm_sub_ps(iz1,jz0);
1099 dx20 = _mm_sub_ps(ix2,jx0);
1100 dy20 = _mm_sub_ps(iy2,jy0);
1101 dz20 = _mm_sub_ps(iz2,jz0);
1102 dx30 = _mm_sub_ps(ix3,jx0);
1103 dy30 = _mm_sub_ps(iy3,jy0);
1104 dz30 = _mm_sub_ps(iz3,jz0);
1106 /* Calculate squared distance and things based on it */
1107 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1108 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1109 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1110 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1112 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1113 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1114 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1116 rinvsq00 = gmx_mm_inv_ps(rsq00);
1118 /* Load parameters for j particles */
1119 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1120 charge+jnrC+0,charge+jnrD+0);
1121 vdwjidx0A = 2*vdwtype[jnrA+0];
1122 vdwjidx0B = 2*vdwtype[jnrB+0];
1123 vdwjidx0C = 2*vdwtype[jnrC+0];
1124 vdwjidx0D = 2*vdwtype[jnrD+0];
1126 fjx0 = _mm_setzero_ps();
1127 fjy0 = _mm_setzero_ps();
1128 fjz0 = _mm_setzero_ps();
1130 /**************************
1131 * CALCULATE INTERACTIONS *
1132 **************************/
1134 /* Compute parameters for interactions between i and j atoms */
1135 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1136 vdwparam+vdwioffset0+vdwjidx0B,
1137 vdwparam+vdwioffset0+vdwjidx0C,
1138 vdwparam+vdwioffset0+vdwjidx0D,
1141 /* LENNARD-JONES DISPERSION/REPULSION */
1143 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1144 fvdw = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1148 fscal = _mm_andnot_ps(dummy_mask,fscal);
1150 /* Update vectorial force */
1151 fix0 = _mm_macc_ps(dx00,fscal,fix0);
1152 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
1153 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
1155 fjx0 = _mm_macc_ps(dx00,fscal,fjx0);
1156 fjy0 = _mm_macc_ps(dy00,fscal,fjy0);
1157 fjz0 = _mm_macc_ps(dz00,fscal,fjz0);
1159 /**************************
1160 * CALCULATE INTERACTIONS *
1161 **************************/
1163 r10 = _mm_mul_ps(rsq10,rinv10);
1164 r10 = _mm_andnot_ps(dummy_mask,r10);
1166 /* Compute parameters for interactions between i and j atoms */
1167 qq10 = _mm_mul_ps(iq1,jq0);
1169 /* Calculate table index by multiplying r with table scale and truncate to integer */
1170 rt = _mm_mul_ps(r10,vftabscale);
1171 vfitab = _mm_cvttps_epi32(rt);
1173 vfeps = _mm_frcz_ps(rt);
1175 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1177 twovfeps = _mm_add_ps(vfeps,vfeps);
1178 vfitab = _mm_slli_epi32(vfitab,2);
1180 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1181 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1182 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1183 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1184 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1185 _MM_TRANSPOSE4_PS(Y,F,G,H);
1186 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1187 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1188 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1192 fscal = _mm_andnot_ps(dummy_mask,fscal);
1194 /* Update vectorial force */
1195 fix1 = _mm_macc_ps(dx10,fscal,fix1);
1196 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1197 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1199 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1200 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1201 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1203 /**************************
1204 * CALCULATE INTERACTIONS *
1205 **************************/
1207 r20 = _mm_mul_ps(rsq20,rinv20);
1208 r20 = _mm_andnot_ps(dummy_mask,r20);
1210 /* Compute parameters for interactions between i and j atoms */
1211 qq20 = _mm_mul_ps(iq2,jq0);
1213 /* Calculate table index by multiplying r with table scale and truncate to integer */
1214 rt = _mm_mul_ps(r20,vftabscale);
1215 vfitab = _mm_cvttps_epi32(rt);
1217 vfeps = _mm_frcz_ps(rt);
1219 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1221 twovfeps = _mm_add_ps(vfeps,vfeps);
1222 vfitab = _mm_slli_epi32(vfitab,2);
1224 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1225 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1226 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1227 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1228 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1229 _MM_TRANSPOSE4_PS(Y,F,G,H);
1230 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1231 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1232 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1236 fscal = _mm_andnot_ps(dummy_mask,fscal);
1238 /* Update vectorial force */
1239 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1240 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1241 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1243 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1244 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1245 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1247 /**************************
1248 * CALCULATE INTERACTIONS *
1249 **************************/
1251 r30 = _mm_mul_ps(rsq30,rinv30);
1252 r30 = _mm_andnot_ps(dummy_mask,r30);
1254 /* Compute parameters for interactions between i and j atoms */
1255 qq30 = _mm_mul_ps(iq3,jq0);
1257 /* Calculate table index by multiplying r with table scale and truncate to integer */
1258 rt = _mm_mul_ps(r30,vftabscale);
1259 vfitab = _mm_cvttps_epi32(rt);
1261 vfeps = _mm_frcz_ps(rt);
1263 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1265 twovfeps = _mm_add_ps(vfeps,vfeps);
1266 vfitab = _mm_slli_epi32(vfitab,2);
1268 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1269 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1270 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1271 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1272 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1273 _MM_TRANSPOSE4_PS(Y,F,G,H);
1274 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1275 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1276 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1280 fscal = _mm_andnot_ps(dummy_mask,fscal);
1282 /* Update vectorial force */
1283 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1284 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1285 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1287 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1288 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1289 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1291 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1292 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1293 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1294 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1296 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1298 /* Inner loop uses 159 flops */
1301 /* End of innermost loop */
1303 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1304 f+i_coord_offset,fshift+i_shift_offset);
1306 /* Increment number of inner iterations */
1307 inneriter += j_index_end - j_index_start;
1309 /* Outer loop uses 24 flops */
1312 /* Increment number of outer iterations */
1315 /* Update outer/inner flops */
1317 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*159);