/*
* This file is part of the GROMACS molecular simulation package.
*
- * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
+ * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
* Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
* and including many others, as listed in the AUTHORS file in the
* top-level source directory and at http://www.gromacs.org.
__m256 velec,felec,velecsum,facel,crf,krf,krf2;
real *charge;
/* #endif */
- /* #if 'GeneralizedBorn' in KERNEL_ELEC */
- __m256i gbitab;
- __m128i gbitab_lo,gbitab_hi;
- __m256 vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
- __m256 minushalf = _mm256_set1_ps(-0.5);
- real *invsqrta,*dvda,*gbtab;
- /* #endif */
/* #if KERNEL_VDW != 'None' */
int nvdwtype;
__m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
__m256 one_sixth = _mm256_set1_ps(1.0/6.0);
__m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
/* #endif */
- /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
+ /* #if 'Table' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
__m256i vfitab;
__m128i vfitab_lo,vfitab_hi;
__m128i ifour = _mm_set1_epi32(4);
/* #endif */
/* #endif */
- /* #if KERNEL_ELEC=='GeneralizedBorn' */
- invsqrta = fr->invsqrta;
- dvda = fr->dvda;
- gbtabscale = _mm256_set1_ps(fr->gbtab->scale);
- gbtab = fr->gbtab->data;
- gbinvepsdiff = _mm256_set1_ps((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
- /* #endif */
-
/* #if 'Water' in GEOMETRY_I */
/* Setup water-specific parameters */
inr = nlist->iinr[0];
/* #for I in PARTICLES_ELEC_I */
iq{I} = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+{I}]));
/* #define OUTERFLOPS OUTERFLOPS+1 */
- /* #if KERNEL_ELEC=='GeneralizedBorn' */
- isai{I} = _mm256_set1_ps(invsqrta[inr+{I}]);
- /* #endif */
/* #endfor */
/* #for I in PARTICLES_VDW_I */
vdwioffsetptr{I} = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
/* #if KERNEL_ELEC != 'None' */
velecsum = _mm256_setzero_ps();
/* #endif */
- /* #if 'GeneralizedBorn' in KERNEL_ELEC */
- vgbsum = _mm256_setzero_ps();
- /* #endif */
/* #if KERNEL_VDW != 'None' */
vvdwsum = _mm256_setzero_ps();
/* #endif */
/* #endif */
- /* #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
- dvdasum = _mm256_setzero_ps();
- /* #endif */
/* #for ROUND in ['Loop','Epilogue'] */
charge+jnrC+{J},charge+jnrD+{J},
charge+jnrE+{J},charge+jnrF+{J},
charge+jnrG+{J},charge+jnrH+{J});
- /* #if KERNEL_ELEC=='GeneralizedBorn' */
- isaj{J} = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+{J},invsqrta+jnrB+{J},
- invsqrta+jnrC+{J},invsqrta+jnrD+{J},
- invsqrta+jnrE+{J},invsqrta+jnrF+{J},
- invsqrta+jnrG+{J},invsqrta+jnrH+{J});
- /* #endif */
/* #endfor */
/* #for J in PARTICLES_VDW_J */
vdwjidx{J}A = 2*vdwtype[jnrA+{J}];
/* #define INNERFLOPS INNERFLOPS+3 */
/* #endif */
- /* #elif KERNEL_ELEC=='GeneralizedBorn' */
-
- /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
- isaprod = _mm256_mul_ps(isai{I},isaj{J});
- gbqqfactor = _mm256_xor_ps(signbit,_mm256_mul_ps(qq{I}{J},_mm256_mul_ps(isaprod,gbinvepsdiff)));
- gbscale = _mm256_mul_ps(isaprod,gbtabscale);
- /* #define INNERFLOPS INNERFLOPS+5 */
-
- /* Calculate generalized born table index - this is a separate table from the normal one,
- * but we use the same procedure by multiplying r with scale and truncating to integer.
- */
- rt = _mm256_mul_ps(r{I}{J},gbscale);
- gbitab = _mm256_cvttps_epi32(rt);
- gbeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
- /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
- gbitab_lo = _mm256_extractf128_si256(gbitab,0x0);
- gbitab_hi = _mm256_extractf128_si256(gbitab,0x1);
- gbitab_lo = _mm_slli_epi32(gbitab_lo,2);
- gbitab_hi = _mm_slli_epi32(gbitab_hi,2);
- Y = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
- _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
- F = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
- _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
- G = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
- _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
- H = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
- _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
- GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
- Heps = _mm256_mul_ps(gbeps,H);
- Fp = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
- VV = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
- vgb = _mm256_mul_ps(gbqqfactor,VV);
- /* #define INNERFLOPS INNERFLOPS+10 */
-
- /* #if 'Force' in KERNEL_VF */
- FF = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
- fgb = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
- dvdatmp = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r{I}{J})));
- /* #if ROUND == 'Epilogue' */
- dvdatmp = _mm256_andnot_ps(dummy_mask,dvdatmp);
- /* #endif */
- dvdasum = _mm256_add_ps(dvdasum,dvdatmp);
- /* #if ROUND == 'Loop' */
- fjptrA = dvda+jnrA;
- fjptrB = dvda+jnrB;
- fjptrC = dvda+jnrC;
- fjptrD = dvda+jnrD;
- fjptrE = dvda+jnrE;
- fjptrF = dvda+jnrF;
- fjptrG = dvda+jnrG;
- fjptrH = dvda+jnrH;
- /* #else */
- /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
- fjptrA = (jnrlistA>=0) ? dvda+jnrA : scratch;
- fjptrB = (jnrlistB>=0) ? dvda+jnrB : scratch;
- fjptrC = (jnrlistC>=0) ? dvda+jnrC : scratch;
- fjptrD = (jnrlistD>=0) ? dvda+jnrD : scratch;
- fjptrE = (jnrlistE>=0) ? dvda+jnrE : scratch;
- fjptrF = (jnrlistF>=0) ? dvda+jnrF : scratch;
- fjptrG = (jnrlistG>=0) ? dvda+jnrG : scratch;
- fjptrH = (jnrlistH>=0) ? dvda+jnrH : scratch;
- /* #endif */
- gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
- _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj{J},isaj{J})));
- /* #define INNERFLOPS INNERFLOPS+12 */
- /* #endif */
- velec = _mm256_mul_ps(qq{I}{J},rinv{I}{J});
- /* #define INNERFLOPS INNERFLOPS+1 */
- /* #if 'Force' in KERNEL_VF */
- felec = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv{I}{J}),fgb),rinv{I}{J});
- /* #define INNERFLOPS INNERFLOPS+3 */
- /* #endif */
-
/* #elif KERNEL_ELEC=='Ewald' */
/* EWALD ELECTROSTATICS */
/* #endif */
velecsum = _mm256_add_ps(velecsum,velec);
/* #define INNERFLOPS INNERFLOPS+1 */
- /* #if KERNEL_ELEC=='GeneralizedBorn' */
- /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
- vgb = _mm256_and_ps(vgb,cutoff_mask);
- /* #define INNERFLOPS INNERFLOPS+1 */
- /* #endif */
- /* #if ROUND == 'Epilogue' */
- vgb = _mm256_andnot_ps(dummy_mask,vgb);
- /* #endif */
- vgbsum = _mm256_add_ps(vgbsum,vgb);
- /* #define INNERFLOPS INNERFLOPS+1 */
- /* #endif */
/* #endif */
/* #if 'vdw' in INTERACTION_FLAGS[I][J] */
/* ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
/* #define OUTERFLOPS OUTERFLOPS+1 */
/* #endif */
- /* #if 'GeneralizedBorn' in KERNEL_ELEC */
- gmx_mm256_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
- /* #define OUTERFLOPS OUTERFLOPS+1 */
- /* #endif */
/* #if KERNEL_VDW != 'None' */
gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
/* #define OUTERFLOPS OUTERFLOPS+1 */
/* #endif */
/* #endif */
- /* #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
- dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai{I},isai{I}));
- gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
- /* #endif */
/* Increment number of inner iterations */
inneriter += j_index_end - j_index_start;
biod.pnpi.spb.ru / alexxy / gromacs.git / blobdiff