2 * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
34 #include "kernelutil_x86_sse2_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_sse2_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_sse2_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 j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
63 real shX,shY,shZ,rcutoff_scalar;
64 real *shiftvec,*fshift,*x,*f;
65 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
68 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
69 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
70 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
71 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
73 __m128 dummy_mask,cutoff_mask;
74 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
75 __m128 one = _mm_set1_ps(1.0);
76 __m128 two = _mm_set1_ps(2.0);
82 jindex = nlist->jindex;
84 shiftidx = nlist->shift;
86 shiftvec = fr->shift_vec[0];
87 fshift = fr->fshift[0];
88 facel = _mm_set1_ps(fr->epsfac);
89 charge = mdatoms->chargeA;
91 /* Avoid stupid compiler warnings */
92 jnrA = jnrB = jnrC = jnrD = 0;
101 /* Start outer loop over neighborlists */
102 for(iidx=0; iidx<nri; iidx++)
104 /* Load shift vector for this list */
105 i_shift_offset = DIM*shiftidx[iidx];
106 shX = shiftvec[i_shift_offset+XX];
107 shY = shiftvec[i_shift_offset+YY];
108 shZ = shiftvec[i_shift_offset+ZZ];
110 /* Load limits for loop over neighbors */
111 j_index_start = jindex[iidx];
112 j_index_end = jindex[iidx+1];
114 /* Get outer coordinate index */
116 i_coord_offset = DIM*inr;
118 /* Load i particle coords and add shift vector */
119 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
120 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
121 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
123 fix0 = _mm_setzero_ps();
124 fiy0 = _mm_setzero_ps();
125 fiz0 = _mm_setzero_ps();
127 /* Load parameters for i particles */
128 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
130 /* Reset potential sums */
131 velecsum = _mm_setzero_ps();
133 /* Start inner kernel loop */
134 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
137 /* Get j neighbor index, and coordinate index */
143 j_coord_offsetA = DIM*jnrA;
144 j_coord_offsetB = DIM*jnrB;
145 j_coord_offsetC = DIM*jnrC;
146 j_coord_offsetD = DIM*jnrD;
148 /* load j atom coordinates */
149 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
150 x+j_coord_offsetC,x+j_coord_offsetD,
153 /* Calculate displacement vector */
154 dx00 = _mm_sub_ps(ix0,jx0);
155 dy00 = _mm_sub_ps(iy0,jy0);
156 dz00 = _mm_sub_ps(iz0,jz0);
158 /* Calculate squared distance and things based on it */
159 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
161 rinv00 = gmx_mm_invsqrt_ps(rsq00);
163 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
165 /* Load parameters for j particles */
166 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
167 charge+jnrC+0,charge+jnrD+0);
169 /**************************
170 * CALCULATE INTERACTIONS *
171 **************************/
173 /* Compute parameters for interactions between i and j atoms */
174 qq00 = _mm_mul_ps(iq0,jq0);
176 /* COULOMB ELECTROSTATICS */
177 velec = _mm_mul_ps(qq00,rinv00);
178 felec = _mm_mul_ps(velec,rinvsq00);
180 /* Update potential sum for this i atom from the interaction with this j atom. */
181 velecsum = _mm_add_ps(velecsum,velec);
185 /* Calculate temporary vectorial force */
186 tx = _mm_mul_ps(fscal,dx00);
187 ty = _mm_mul_ps(fscal,dy00);
188 tz = _mm_mul_ps(fscal,dz00);
190 /* Update vectorial force */
191 fix0 = _mm_add_ps(fix0,tx);
192 fiy0 = _mm_add_ps(fiy0,ty);
193 fiz0 = _mm_add_ps(fiz0,tz);
195 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
196 f+j_coord_offsetC,f+j_coord_offsetD,
199 /* Inner loop uses 28 flops */
205 /* Get j neighbor index, and coordinate index */
211 /* Sign of each element will be negative for non-real atoms.
212 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
213 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
215 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
216 jnrA = (jnrA>=0) ? jnrA : 0;
217 jnrB = (jnrB>=0) ? jnrB : 0;
218 jnrC = (jnrC>=0) ? jnrC : 0;
219 jnrD = (jnrD>=0) ? jnrD : 0;
221 j_coord_offsetA = DIM*jnrA;
222 j_coord_offsetB = DIM*jnrB;
223 j_coord_offsetC = DIM*jnrC;
224 j_coord_offsetD = DIM*jnrD;
226 /* load j atom coordinates */
227 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
228 x+j_coord_offsetC,x+j_coord_offsetD,
231 /* Calculate displacement vector */
232 dx00 = _mm_sub_ps(ix0,jx0);
233 dy00 = _mm_sub_ps(iy0,jy0);
234 dz00 = _mm_sub_ps(iz0,jz0);
236 /* Calculate squared distance and things based on it */
237 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
239 rinv00 = gmx_mm_invsqrt_ps(rsq00);
241 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
243 /* Load parameters for j particles */
244 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
245 charge+jnrC+0,charge+jnrD+0);
247 /**************************
248 * CALCULATE INTERACTIONS *
249 **************************/
251 /* Compute parameters for interactions between i and j atoms */
252 qq00 = _mm_mul_ps(iq0,jq0);
254 /* COULOMB ELECTROSTATICS */
255 velec = _mm_mul_ps(qq00,rinv00);
256 felec = _mm_mul_ps(velec,rinvsq00);
258 /* Update potential sum for this i atom from the interaction with this j atom. */
259 velec = _mm_andnot_ps(dummy_mask,velec);
260 velecsum = _mm_add_ps(velecsum,velec);
264 fscal = _mm_andnot_ps(dummy_mask,fscal);
266 /* Calculate temporary vectorial force */
267 tx = _mm_mul_ps(fscal,dx00);
268 ty = _mm_mul_ps(fscal,dy00);
269 tz = _mm_mul_ps(fscal,dz00);
271 /* Update vectorial force */
272 fix0 = _mm_add_ps(fix0,tx);
273 fiy0 = _mm_add_ps(fiy0,ty);
274 fiz0 = _mm_add_ps(fiz0,tz);
276 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
277 f+j_coord_offsetC,f+j_coord_offsetD,
280 /* Inner loop uses 28 flops */
283 /* End of innermost loop */
285 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
286 f+i_coord_offset,fshift+i_shift_offset);
289 /* Update potential energies */
290 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
292 /* Increment number of inner iterations */
293 inneriter += j_index_end - j_index_start;
295 /* Outer loop uses 11 flops */
298 /* Increment number of outer iterations */
301 /* Update outer/inner flops */
303 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*11 + inneriter*28);
306 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse2_single
307 * Electrostatics interaction: Coulomb
308 * VdW interaction: None
309 * Geometry: Particle-Particle
310 * Calculate force/pot: Force
313 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse2_single
314 (t_nblist * gmx_restrict nlist,
315 rvec * gmx_restrict xx,
316 rvec * gmx_restrict ff,
317 t_forcerec * gmx_restrict fr,
318 t_mdatoms * gmx_restrict mdatoms,
319 nb_kernel_data_t * gmx_restrict kernel_data,
320 t_nrnb * gmx_restrict nrnb)
322 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
323 * just 0 for non-waters.
324 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
325 * jnr indices corresponding to data put in the four positions in the SIMD register.
327 int i_shift_offset,i_coord_offset,outeriter,inneriter;
328 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
329 int jnrA,jnrB,jnrC,jnrD;
330 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
331 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
332 real shX,shY,shZ,rcutoff_scalar;
333 real *shiftvec,*fshift,*x,*f;
334 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
336 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
337 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
338 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
339 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
340 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
342 __m128 dummy_mask,cutoff_mask;
343 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
344 __m128 one = _mm_set1_ps(1.0);
345 __m128 two = _mm_set1_ps(2.0);
351 jindex = nlist->jindex;
353 shiftidx = nlist->shift;
355 shiftvec = fr->shift_vec[0];
356 fshift = fr->fshift[0];
357 facel = _mm_set1_ps(fr->epsfac);
358 charge = mdatoms->chargeA;
360 /* Avoid stupid compiler warnings */
361 jnrA = jnrB = jnrC = jnrD = 0;
370 /* Start outer loop over neighborlists */
371 for(iidx=0; iidx<nri; iidx++)
373 /* Load shift vector for this list */
374 i_shift_offset = DIM*shiftidx[iidx];
375 shX = shiftvec[i_shift_offset+XX];
376 shY = shiftvec[i_shift_offset+YY];
377 shZ = shiftvec[i_shift_offset+ZZ];
379 /* Load limits for loop over neighbors */
380 j_index_start = jindex[iidx];
381 j_index_end = jindex[iidx+1];
383 /* Get outer coordinate index */
385 i_coord_offset = DIM*inr;
387 /* Load i particle coords and add shift vector */
388 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
389 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
390 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
392 fix0 = _mm_setzero_ps();
393 fiy0 = _mm_setzero_ps();
394 fiz0 = _mm_setzero_ps();
396 /* Load parameters for i particles */
397 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
399 /* Start inner kernel loop */
400 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
403 /* Get j neighbor index, and coordinate index */
409 j_coord_offsetA = DIM*jnrA;
410 j_coord_offsetB = DIM*jnrB;
411 j_coord_offsetC = DIM*jnrC;
412 j_coord_offsetD = DIM*jnrD;
414 /* load j atom coordinates */
415 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
416 x+j_coord_offsetC,x+j_coord_offsetD,
419 /* Calculate displacement vector */
420 dx00 = _mm_sub_ps(ix0,jx0);
421 dy00 = _mm_sub_ps(iy0,jy0);
422 dz00 = _mm_sub_ps(iz0,jz0);
424 /* Calculate squared distance and things based on it */
425 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
427 rinv00 = gmx_mm_invsqrt_ps(rsq00);
429 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
431 /* Load parameters for j particles */
432 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
433 charge+jnrC+0,charge+jnrD+0);
435 /**************************
436 * CALCULATE INTERACTIONS *
437 **************************/
439 /* Compute parameters for interactions between i and j atoms */
440 qq00 = _mm_mul_ps(iq0,jq0);
442 /* COULOMB ELECTROSTATICS */
443 velec = _mm_mul_ps(qq00,rinv00);
444 felec = _mm_mul_ps(velec,rinvsq00);
448 /* Calculate temporary vectorial force */
449 tx = _mm_mul_ps(fscal,dx00);
450 ty = _mm_mul_ps(fscal,dy00);
451 tz = _mm_mul_ps(fscal,dz00);
453 /* Update vectorial force */
454 fix0 = _mm_add_ps(fix0,tx);
455 fiy0 = _mm_add_ps(fiy0,ty);
456 fiz0 = _mm_add_ps(fiz0,tz);
458 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
459 f+j_coord_offsetC,f+j_coord_offsetD,
462 /* Inner loop uses 27 flops */
468 /* Get j neighbor index, and coordinate index */
474 /* Sign of each element will be negative for non-real atoms.
475 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
476 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
478 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
479 jnrA = (jnrA>=0) ? jnrA : 0;
480 jnrB = (jnrB>=0) ? jnrB : 0;
481 jnrC = (jnrC>=0) ? jnrC : 0;
482 jnrD = (jnrD>=0) ? jnrD : 0;
484 j_coord_offsetA = DIM*jnrA;
485 j_coord_offsetB = DIM*jnrB;
486 j_coord_offsetC = DIM*jnrC;
487 j_coord_offsetD = DIM*jnrD;
489 /* load j atom coordinates */
490 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
491 x+j_coord_offsetC,x+j_coord_offsetD,
494 /* Calculate displacement vector */
495 dx00 = _mm_sub_ps(ix0,jx0);
496 dy00 = _mm_sub_ps(iy0,jy0);
497 dz00 = _mm_sub_ps(iz0,jz0);
499 /* Calculate squared distance and things based on it */
500 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
502 rinv00 = gmx_mm_invsqrt_ps(rsq00);
504 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
506 /* Load parameters for j particles */
507 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
508 charge+jnrC+0,charge+jnrD+0);
510 /**************************
511 * CALCULATE INTERACTIONS *
512 **************************/
514 /* Compute parameters for interactions between i and j atoms */
515 qq00 = _mm_mul_ps(iq0,jq0);
517 /* COULOMB ELECTROSTATICS */
518 velec = _mm_mul_ps(qq00,rinv00);
519 felec = _mm_mul_ps(velec,rinvsq00);
523 fscal = _mm_andnot_ps(dummy_mask,fscal);
525 /* Calculate temporary vectorial force */
526 tx = _mm_mul_ps(fscal,dx00);
527 ty = _mm_mul_ps(fscal,dy00);
528 tz = _mm_mul_ps(fscal,dz00);
530 /* Update vectorial force */
531 fix0 = _mm_add_ps(fix0,tx);
532 fiy0 = _mm_add_ps(fiy0,ty);
533 fiz0 = _mm_add_ps(fiz0,tz);
535 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
536 f+j_coord_offsetC,f+j_coord_offsetD,
539 /* Inner loop uses 27 flops */
542 /* End of innermost loop */
544 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
545 f+i_coord_offset,fshift+i_shift_offset);
547 /* Increment number of inner iterations */
548 inneriter += j_index_end - j_index_start;
550 /* Outer loop uses 10 flops */
553 /* Increment number of outer iterations */
556 /* Update outer/inner flops */
558 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*10 + inneriter*27);