2 * Note: this file was generated by the Gromacs sse2_double 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_sse2_double.h"
34 #include "kernelutil_x86_sse2_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomP1P1_VF_sse2_double
38 * Electrostatics interaction: CubicSplineTable
39 * VdW interaction: None
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_VF_sse2_double
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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
68 int vdwjidx0A,vdwjidx0B;
69 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
70 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
71 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
74 __m128i ifour = _mm_set1_epi32(4);
75 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
77 __m128d dummy_mask,cutoff_mask;
78 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
79 __m128d one = _mm_set1_pd(1.0);
80 __m128d two = _mm_set1_pd(2.0);
86 jindex = nlist->jindex;
88 shiftidx = nlist->shift;
90 shiftvec = fr->shift_vec[0];
91 fshift = fr->fshift[0];
92 facel = _mm_set1_pd(fr->epsfac);
93 charge = mdatoms->chargeA;
95 vftab = kernel_data->table_elec->data;
96 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
98 /* Avoid stupid compiler warnings */
106 /* Start outer loop over neighborlists */
107 for(iidx=0; iidx<nri; iidx++)
109 /* Load shift vector for this list */
110 i_shift_offset = DIM*shiftidx[iidx];
112 /* Load limits for loop over neighbors */
113 j_index_start = jindex[iidx];
114 j_index_end = jindex[iidx+1];
116 /* Get outer coordinate index */
118 i_coord_offset = DIM*inr;
120 /* Load i particle coords and add shift vector */
121 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
123 fix0 = _mm_setzero_pd();
124 fiy0 = _mm_setzero_pd();
125 fiz0 = _mm_setzero_pd();
127 /* Load parameters for i particles */
128 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
130 /* Reset potential sums */
131 velecsum = _mm_setzero_pd();
133 /* Start inner kernel loop */
134 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
137 /* Get j neighbor index, and coordinate index */
140 j_coord_offsetA = DIM*jnrA;
141 j_coord_offsetB = DIM*jnrB;
143 /* load j atom coordinates */
144 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
147 /* Calculate displacement vector */
148 dx00 = _mm_sub_pd(ix0,jx0);
149 dy00 = _mm_sub_pd(iy0,jy0);
150 dz00 = _mm_sub_pd(iz0,jz0);
152 /* Calculate squared distance and things based on it */
153 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
155 rinv00 = gmx_mm_invsqrt_pd(rsq00);
157 /* Load parameters for j particles */
158 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
160 /**************************
161 * CALCULATE INTERACTIONS *
162 **************************/
164 r00 = _mm_mul_pd(rsq00,rinv00);
166 /* Compute parameters for interactions between i and j atoms */
167 qq00 = _mm_mul_pd(iq0,jq0);
169 /* Calculate table index by multiplying r with table scale and truncate to integer */
170 rt = _mm_mul_pd(r00,vftabscale);
171 vfitab = _mm_cvttpd_epi32(rt);
172 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
173 vfitab = _mm_slli_epi32(vfitab,2);
175 /* CUBIC SPLINE TABLE ELECTROSTATICS */
176 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
177 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
178 GMX_MM_TRANSPOSE2_PD(Y,F);
179 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
180 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
181 GMX_MM_TRANSPOSE2_PD(G,H);
182 Heps = _mm_mul_pd(vfeps,H);
183 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
184 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
185 velec = _mm_mul_pd(qq00,VV);
186 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
187 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
189 /* Update potential sum for this i atom from the interaction with this j atom. */
190 velecsum = _mm_add_pd(velecsum,velec);
194 /* Calculate temporary vectorial force */
195 tx = _mm_mul_pd(fscal,dx00);
196 ty = _mm_mul_pd(fscal,dy00);
197 tz = _mm_mul_pd(fscal,dz00);
199 /* Update vectorial force */
200 fix0 = _mm_add_pd(fix0,tx);
201 fiy0 = _mm_add_pd(fiy0,ty);
202 fiz0 = _mm_add_pd(fiz0,tz);
204 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
206 /* Inner loop uses 43 flops */
213 j_coord_offsetA = DIM*jnrA;
215 /* load j atom coordinates */
216 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
219 /* Calculate displacement vector */
220 dx00 = _mm_sub_pd(ix0,jx0);
221 dy00 = _mm_sub_pd(iy0,jy0);
222 dz00 = _mm_sub_pd(iz0,jz0);
224 /* Calculate squared distance and things based on it */
225 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
227 rinv00 = gmx_mm_invsqrt_pd(rsq00);
229 /* Load parameters for j particles */
230 jq0 = _mm_load_sd(charge+jnrA+0);
232 /**************************
233 * CALCULATE INTERACTIONS *
234 **************************/
236 r00 = _mm_mul_pd(rsq00,rinv00);
238 /* Compute parameters for interactions between i and j atoms */
239 qq00 = _mm_mul_pd(iq0,jq0);
241 /* Calculate table index by multiplying r with table scale and truncate to integer */
242 rt = _mm_mul_pd(r00,vftabscale);
243 vfitab = _mm_cvttpd_epi32(rt);
244 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
245 vfitab = _mm_slli_epi32(vfitab,2);
247 /* CUBIC SPLINE TABLE ELECTROSTATICS */
248 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
249 F = _mm_setzero_pd();
250 GMX_MM_TRANSPOSE2_PD(Y,F);
251 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
252 H = _mm_setzero_pd();
253 GMX_MM_TRANSPOSE2_PD(G,H);
254 Heps = _mm_mul_pd(vfeps,H);
255 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
256 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
257 velec = _mm_mul_pd(qq00,VV);
258 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
259 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
261 /* Update potential sum for this i atom from the interaction with this j atom. */
262 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
263 velecsum = _mm_add_pd(velecsum,velec);
267 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
269 /* Calculate temporary vectorial force */
270 tx = _mm_mul_pd(fscal,dx00);
271 ty = _mm_mul_pd(fscal,dy00);
272 tz = _mm_mul_pd(fscal,dz00);
274 /* Update vectorial force */
275 fix0 = _mm_add_pd(fix0,tx);
276 fiy0 = _mm_add_pd(fiy0,ty);
277 fiz0 = _mm_add_pd(fiz0,tz);
279 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
281 /* Inner loop uses 43 flops */
284 /* End of innermost loop */
286 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
287 f+i_coord_offset,fshift+i_shift_offset);
290 /* Update potential energies */
291 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
293 /* Increment number of inner iterations */
294 inneriter += j_index_end - j_index_start;
296 /* Outer loop uses 8 flops */
299 /* Increment number of outer iterations */
302 /* Update outer/inner flops */
304 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*43);
307 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_double
308 * Electrostatics interaction: CubicSplineTable
309 * VdW interaction: None
310 * Geometry: Particle-Particle
311 * Calculate force/pot: Force
314 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_double
315 (t_nblist * gmx_restrict nlist,
316 rvec * gmx_restrict xx,
317 rvec * gmx_restrict ff,
318 t_forcerec * gmx_restrict fr,
319 t_mdatoms * gmx_restrict mdatoms,
320 nb_kernel_data_t * gmx_restrict kernel_data,
321 t_nrnb * gmx_restrict nrnb)
323 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
324 * just 0 for non-waters.
325 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
326 * jnr indices corresponding to data put in the four positions in the SIMD register.
328 int i_shift_offset,i_coord_offset,outeriter,inneriter;
329 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
331 int j_coord_offsetA,j_coord_offsetB;
332 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
334 real *shiftvec,*fshift,*x,*f;
335 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
337 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
338 int vdwjidx0A,vdwjidx0B;
339 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
340 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
341 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
344 __m128i ifour = _mm_set1_epi32(4);
345 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
347 __m128d dummy_mask,cutoff_mask;
348 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
349 __m128d one = _mm_set1_pd(1.0);
350 __m128d two = _mm_set1_pd(2.0);
356 jindex = nlist->jindex;
358 shiftidx = nlist->shift;
360 shiftvec = fr->shift_vec[0];
361 fshift = fr->fshift[0];
362 facel = _mm_set1_pd(fr->epsfac);
363 charge = mdatoms->chargeA;
365 vftab = kernel_data->table_elec->data;
366 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
368 /* Avoid stupid compiler warnings */
376 /* Start outer loop over neighborlists */
377 for(iidx=0; iidx<nri; iidx++)
379 /* Load shift vector for this list */
380 i_shift_offset = DIM*shiftidx[iidx];
382 /* Load limits for loop over neighbors */
383 j_index_start = jindex[iidx];
384 j_index_end = jindex[iidx+1];
386 /* Get outer coordinate index */
388 i_coord_offset = DIM*inr;
390 /* Load i particle coords and add shift vector */
391 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
393 fix0 = _mm_setzero_pd();
394 fiy0 = _mm_setzero_pd();
395 fiz0 = _mm_setzero_pd();
397 /* Load parameters for i particles */
398 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
400 /* Start inner kernel loop */
401 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
404 /* Get j neighbor index, and coordinate index */
407 j_coord_offsetA = DIM*jnrA;
408 j_coord_offsetB = DIM*jnrB;
410 /* load j atom coordinates */
411 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
414 /* Calculate displacement vector */
415 dx00 = _mm_sub_pd(ix0,jx0);
416 dy00 = _mm_sub_pd(iy0,jy0);
417 dz00 = _mm_sub_pd(iz0,jz0);
419 /* Calculate squared distance and things based on it */
420 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
422 rinv00 = gmx_mm_invsqrt_pd(rsq00);
424 /* Load parameters for j particles */
425 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
427 /**************************
428 * CALCULATE INTERACTIONS *
429 **************************/
431 r00 = _mm_mul_pd(rsq00,rinv00);
433 /* Compute parameters for interactions between i and j atoms */
434 qq00 = _mm_mul_pd(iq0,jq0);
436 /* Calculate table index by multiplying r with table scale and truncate to integer */
437 rt = _mm_mul_pd(r00,vftabscale);
438 vfitab = _mm_cvttpd_epi32(rt);
439 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
440 vfitab = _mm_slli_epi32(vfitab,2);
442 /* CUBIC SPLINE TABLE ELECTROSTATICS */
443 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
444 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
445 GMX_MM_TRANSPOSE2_PD(Y,F);
446 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
447 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
448 GMX_MM_TRANSPOSE2_PD(G,H);
449 Heps = _mm_mul_pd(vfeps,H);
450 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
451 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
452 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
456 /* Calculate temporary vectorial force */
457 tx = _mm_mul_pd(fscal,dx00);
458 ty = _mm_mul_pd(fscal,dy00);
459 tz = _mm_mul_pd(fscal,dz00);
461 /* Update vectorial force */
462 fix0 = _mm_add_pd(fix0,tx);
463 fiy0 = _mm_add_pd(fiy0,ty);
464 fiz0 = _mm_add_pd(fiz0,tz);
466 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
468 /* Inner loop uses 39 flops */
475 j_coord_offsetA = DIM*jnrA;
477 /* load j atom coordinates */
478 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
481 /* Calculate displacement vector */
482 dx00 = _mm_sub_pd(ix0,jx0);
483 dy00 = _mm_sub_pd(iy0,jy0);
484 dz00 = _mm_sub_pd(iz0,jz0);
486 /* Calculate squared distance and things based on it */
487 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
489 rinv00 = gmx_mm_invsqrt_pd(rsq00);
491 /* Load parameters for j particles */
492 jq0 = _mm_load_sd(charge+jnrA+0);
494 /**************************
495 * CALCULATE INTERACTIONS *
496 **************************/
498 r00 = _mm_mul_pd(rsq00,rinv00);
500 /* Compute parameters for interactions between i and j atoms */
501 qq00 = _mm_mul_pd(iq0,jq0);
503 /* Calculate table index by multiplying r with table scale and truncate to integer */
504 rt = _mm_mul_pd(r00,vftabscale);
505 vfitab = _mm_cvttpd_epi32(rt);
506 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
507 vfitab = _mm_slli_epi32(vfitab,2);
509 /* CUBIC SPLINE TABLE ELECTROSTATICS */
510 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
511 F = _mm_setzero_pd();
512 GMX_MM_TRANSPOSE2_PD(Y,F);
513 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
514 H = _mm_setzero_pd();
515 GMX_MM_TRANSPOSE2_PD(G,H);
516 Heps = _mm_mul_pd(vfeps,H);
517 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
518 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
519 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
523 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
525 /* Calculate temporary vectorial force */
526 tx = _mm_mul_pd(fscal,dx00);
527 ty = _mm_mul_pd(fscal,dy00);
528 tz = _mm_mul_pd(fscal,dz00);
530 /* Update vectorial force */
531 fix0 = _mm_add_pd(fix0,tx);
532 fiy0 = _mm_add_pd(fiy0,ty);
533 fiz0 = _mm_add_pd(fiz0,tz);
535 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
537 /* Inner loop uses 39 flops */
540 /* End of innermost loop */
542 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
543 f+i_coord_offset,fshift+i_shift_offset);
545 /* Increment number of inner iterations */
546 inneriter += j_index_end - j_index_start;
548 /* Outer loop uses 7 flops */
551 /* Increment number of outer iterations */
554 /* Update outer/inner flops */
556 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);