2 * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse4_1_single.h"
34 #include "kernelutil_x86_sse4_1_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_sse4_1_single
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: None
40 * Geometry: Particle-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_sse4_1_single
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
60 int jnrA,jnrB,jnrC,jnrD;
61 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
62 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
63 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
65 real *shiftvec,*fshift,*x,*f;
66 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
68 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
70 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
71 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
72 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
73 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
74 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
76 __m128 dummy_mask,cutoff_mask;
77 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
78 __m128 one = _mm_set1_ps(1.0);
79 __m128 two = _mm_set1_ps(2.0);
85 jindex = nlist->jindex;
87 shiftidx = nlist->shift;
89 shiftvec = fr->shift_vec[0];
90 fshift = fr->fshift[0];
91 facel = _mm_set1_ps(fr->epsfac);
92 charge = mdatoms->chargeA;
94 /* Avoid stupid compiler warnings */
95 jnrA = jnrB = jnrC = jnrD = 0;
104 for(iidx=0;iidx<4*DIM;iidx++)
109 /* Start outer loop over neighborlists */
110 for(iidx=0; iidx<nri; iidx++)
112 /* Load shift vector for this list */
113 i_shift_offset = DIM*shiftidx[iidx];
115 /* Load limits for loop over neighbors */
116 j_index_start = jindex[iidx];
117 j_index_end = jindex[iidx+1];
119 /* Get outer coordinate index */
121 i_coord_offset = DIM*inr;
123 /* Load i particle coords and add shift vector */
124 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
126 fix0 = _mm_setzero_ps();
127 fiy0 = _mm_setzero_ps();
128 fiz0 = _mm_setzero_ps();
130 /* Load parameters for i particles */
131 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
133 /* Reset potential sums */
134 velecsum = _mm_setzero_ps();
136 /* Start inner kernel loop */
137 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
140 /* Get j neighbor index, and coordinate index */
145 j_coord_offsetA = DIM*jnrA;
146 j_coord_offsetB = DIM*jnrB;
147 j_coord_offsetC = DIM*jnrC;
148 j_coord_offsetD = DIM*jnrD;
150 /* load j atom coordinates */
151 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
152 x+j_coord_offsetC,x+j_coord_offsetD,
155 /* Calculate displacement vector */
156 dx00 = _mm_sub_ps(ix0,jx0);
157 dy00 = _mm_sub_ps(iy0,jy0);
158 dz00 = _mm_sub_ps(iz0,jz0);
160 /* Calculate squared distance and things based on it */
161 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
163 rinv00 = gmx_mm_invsqrt_ps(rsq00);
165 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
167 /* Load parameters for j particles */
168 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
169 charge+jnrC+0,charge+jnrD+0);
171 /**************************
172 * CALCULATE INTERACTIONS *
173 **************************/
175 /* Compute parameters for interactions between i and j atoms */
176 qq00 = _mm_mul_ps(iq0,jq0);
178 /* COULOMB ELECTROSTATICS */
179 velec = _mm_mul_ps(qq00,rinv00);
180 felec = _mm_mul_ps(velec,rinvsq00);
182 /* Update potential sum for this i atom from the interaction with this j atom. */
183 velecsum = _mm_add_ps(velecsum,velec);
187 /* Calculate temporary vectorial force */
188 tx = _mm_mul_ps(fscal,dx00);
189 ty = _mm_mul_ps(fscal,dy00);
190 tz = _mm_mul_ps(fscal,dz00);
192 /* Update vectorial force */
193 fix0 = _mm_add_ps(fix0,tx);
194 fiy0 = _mm_add_ps(fiy0,ty);
195 fiz0 = _mm_add_ps(fiz0,tz);
197 fjptrA = f+j_coord_offsetA;
198 fjptrB = f+j_coord_offsetB;
199 fjptrC = f+j_coord_offsetC;
200 fjptrD = f+j_coord_offsetD;
201 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
203 /* Inner loop uses 28 flops */
209 /* Get j neighbor index, and coordinate index */
210 jnrlistA = jjnr[jidx];
211 jnrlistB = jjnr[jidx+1];
212 jnrlistC = jjnr[jidx+2];
213 jnrlistD = jjnr[jidx+3];
214 /* Sign of each element will be negative for non-real atoms.
215 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
216 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
218 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
219 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
220 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
221 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
222 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
223 j_coord_offsetA = DIM*jnrA;
224 j_coord_offsetB = DIM*jnrB;
225 j_coord_offsetC = DIM*jnrC;
226 j_coord_offsetD = DIM*jnrD;
228 /* load j atom coordinates */
229 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
230 x+j_coord_offsetC,x+j_coord_offsetD,
233 /* Calculate displacement vector */
234 dx00 = _mm_sub_ps(ix0,jx0);
235 dy00 = _mm_sub_ps(iy0,jy0);
236 dz00 = _mm_sub_ps(iz0,jz0);
238 /* Calculate squared distance and things based on it */
239 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
241 rinv00 = gmx_mm_invsqrt_ps(rsq00);
243 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
245 /* Load parameters for j particles */
246 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
247 charge+jnrC+0,charge+jnrD+0);
249 /**************************
250 * CALCULATE INTERACTIONS *
251 **************************/
253 /* Compute parameters for interactions between i and j atoms */
254 qq00 = _mm_mul_ps(iq0,jq0);
256 /* COULOMB ELECTROSTATICS */
257 velec = _mm_mul_ps(qq00,rinv00);
258 felec = _mm_mul_ps(velec,rinvsq00);
260 /* Update potential sum for this i atom from the interaction with this j atom. */
261 velec = _mm_andnot_ps(dummy_mask,velec);
262 velecsum = _mm_add_ps(velecsum,velec);
266 fscal = _mm_andnot_ps(dummy_mask,fscal);
268 /* Calculate temporary vectorial force */
269 tx = _mm_mul_ps(fscal,dx00);
270 ty = _mm_mul_ps(fscal,dy00);
271 tz = _mm_mul_ps(fscal,dz00);
273 /* Update vectorial force */
274 fix0 = _mm_add_ps(fix0,tx);
275 fiy0 = _mm_add_ps(fiy0,ty);
276 fiz0 = _mm_add_ps(fiz0,tz);
278 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
279 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
280 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
281 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
282 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
284 /* Inner loop uses 28 flops */
287 /* End of innermost loop */
289 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
290 f+i_coord_offset,fshift+i_shift_offset);
293 /* Update potential energies */
294 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
296 /* Increment number of inner iterations */
297 inneriter += j_index_end - j_index_start;
299 /* Outer loop uses 8 flops */
302 /* Increment number of outer iterations */
305 /* Update outer/inner flops */
307 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*28);
310 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse4_1_single
311 * Electrostatics interaction: Coulomb
312 * VdW interaction: None
313 * Geometry: Particle-Particle
314 * Calculate force/pot: Force
317 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse4_1_single
318 (t_nblist * gmx_restrict nlist,
319 rvec * gmx_restrict xx,
320 rvec * gmx_restrict ff,
321 t_forcerec * gmx_restrict fr,
322 t_mdatoms * gmx_restrict mdatoms,
323 nb_kernel_data_t * gmx_restrict kernel_data,
324 t_nrnb * gmx_restrict nrnb)
326 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
327 * just 0 for non-waters.
328 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
329 * jnr indices corresponding to data put in the four positions in the SIMD register.
331 int i_shift_offset,i_coord_offset,outeriter,inneriter;
332 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
333 int jnrA,jnrB,jnrC,jnrD;
334 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
335 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
336 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
338 real *shiftvec,*fshift,*x,*f;
339 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
341 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
343 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
344 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
345 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
346 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
347 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
349 __m128 dummy_mask,cutoff_mask;
350 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
351 __m128 one = _mm_set1_ps(1.0);
352 __m128 two = _mm_set1_ps(2.0);
358 jindex = nlist->jindex;
360 shiftidx = nlist->shift;
362 shiftvec = fr->shift_vec[0];
363 fshift = fr->fshift[0];
364 facel = _mm_set1_ps(fr->epsfac);
365 charge = mdatoms->chargeA;
367 /* Avoid stupid compiler warnings */
368 jnrA = jnrB = jnrC = jnrD = 0;
377 for(iidx=0;iidx<4*DIM;iidx++)
382 /* Start outer loop over neighborlists */
383 for(iidx=0; iidx<nri; iidx++)
385 /* Load shift vector for this list */
386 i_shift_offset = DIM*shiftidx[iidx];
388 /* Load limits for loop over neighbors */
389 j_index_start = jindex[iidx];
390 j_index_end = jindex[iidx+1];
392 /* Get outer coordinate index */
394 i_coord_offset = DIM*inr;
396 /* Load i particle coords and add shift vector */
397 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
399 fix0 = _mm_setzero_ps();
400 fiy0 = _mm_setzero_ps();
401 fiz0 = _mm_setzero_ps();
403 /* Load parameters for i particles */
404 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
406 /* Start inner kernel loop */
407 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
410 /* Get j neighbor index, and coordinate index */
415 j_coord_offsetA = DIM*jnrA;
416 j_coord_offsetB = DIM*jnrB;
417 j_coord_offsetC = DIM*jnrC;
418 j_coord_offsetD = DIM*jnrD;
420 /* load j atom coordinates */
421 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
422 x+j_coord_offsetC,x+j_coord_offsetD,
425 /* Calculate displacement vector */
426 dx00 = _mm_sub_ps(ix0,jx0);
427 dy00 = _mm_sub_ps(iy0,jy0);
428 dz00 = _mm_sub_ps(iz0,jz0);
430 /* Calculate squared distance and things based on it */
431 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
433 rinv00 = gmx_mm_invsqrt_ps(rsq00);
435 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
437 /* Load parameters for j particles */
438 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
439 charge+jnrC+0,charge+jnrD+0);
441 /**************************
442 * CALCULATE INTERACTIONS *
443 **************************/
445 /* Compute parameters for interactions between i and j atoms */
446 qq00 = _mm_mul_ps(iq0,jq0);
448 /* COULOMB ELECTROSTATICS */
449 velec = _mm_mul_ps(qq00,rinv00);
450 felec = _mm_mul_ps(velec,rinvsq00);
454 /* Calculate temporary vectorial force */
455 tx = _mm_mul_ps(fscal,dx00);
456 ty = _mm_mul_ps(fscal,dy00);
457 tz = _mm_mul_ps(fscal,dz00);
459 /* Update vectorial force */
460 fix0 = _mm_add_ps(fix0,tx);
461 fiy0 = _mm_add_ps(fiy0,ty);
462 fiz0 = _mm_add_ps(fiz0,tz);
464 fjptrA = f+j_coord_offsetA;
465 fjptrB = f+j_coord_offsetB;
466 fjptrC = f+j_coord_offsetC;
467 fjptrD = f+j_coord_offsetD;
468 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
470 /* Inner loop uses 27 flops */
476 /* Get j neighbor index, and coordinate index */
477 jnrlistA = jjnr[jidx];
478 jnrlistB = jjnr[jidx+1];
479 jnrlistC = jjnr[jidx+2];
480 jnrlistD = jjnr[jidx+3];
481 /* Sign of each element will be negative for non-real atoms.
482 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
483 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
485 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
486 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
487 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
488 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
489 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
490 j_coord_offsetA = DIM*jnrA;
491 j_coord_offsetB = DIM*jnrB;
492 j_coord_offsetC = DIM*jnrC;
493 j_coord_offsetD = DIM*jnrD;
495 /* load j atom coordinates */
496 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
497 x+j_coord_offsetC,x+j_coord_offsetD,
500 /* Calculate displacement vector */
501 dx00 = _mm_sub_ps(ix0,jx0);
502 dy00 = _mm_sub_ps(iy0,jy0);
503 dz00 = _mm_sub_ps(iz0,jz0);
505 /* Calculate squared distance and things based on it */
506 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
508 rinv00 = gmx_mm_invsqrt_ps(rsq00);
510 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
512 /* Load parameters for j particles */
513 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
514 charge+jnrC+0,charge+jnrD+0);
516 /**************************
517 * CALCULATE INTERACTIONS *
518 **************************/
520 /* Compute parameters for interactions between i and j atoms */
521 qq00 = _mm_mul_ps(iq0,jq0);
523 /* COULOMB ELECTROSTATICS */
524 velec = _mm_mul_ps(qq00,rinv00);
525 felec = _mm_mul_ps(velec,rinvsq00);
529 fscal = _mm_andnot_ps(dummy_mask,fscal);
531 /* Calculate temporary vectorial force */
532 tx = _mm_mul_ps(fscal,dx00);
533 ty = _mm_mul_ps(fscal,dy00);
534 tz = _mm_mul_ps(fscal,dz00);
536 /* Update vectorial force */
537 fix0 = _mm_add_ps(fix0,tx);
538 fiy0 = _mm_add_ps(fiy0,ty);
539 fiz0 = _mm_add_ps(fiz0,tz);
541 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
542 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
543 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
544 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
545 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
547 /* Inner loop uses 27 flops */
550 /* End of innermost loop */
552 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
553 f+i_coord_offset,fshift+i_shift_offset);
555 /* Increment number of inner iterations */
556 inneriter += j_index_end - j_index_start;
558 /* Outer loop uses 7 flops */
561 /* Increment number of outer iterations */
564 /* Update outer/inner flops */
566 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);