nb_kernel030_ia32_3dnow.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,107 行 · 第 1/3 页
S
1,107 行
pfadd mm7, mm5 ;# mm7=FF pfmul mm5, mm1 ;# mm5=eps*Fp pfadd mm5, mm4 ;# mm5= VV movq mm6, [esp + nb030_c12] pfmul mm7, mm6 ;# fijR pfmul mm5, mm6 ;# Vvdw12 pfadd mm3, mm7 ;# total fscal fijD+ fijR ;# change sign of mm3 pxor mm1,mm1 pfsub mm1, mm3 pfmul mm1, [esp + nb030_tsc] pfmul mm0, mm1 ;# mm0 is total fscal now prefetchw [esp + nb030_dx1] ;# prefetch i forces to cache ;# spread fscalar to both positions movq mm1,mm0 punpckldq mm0,mm0 punpckhdq mm1,mm1 ;# calc vector force prefetchw [edi + eax*4] ;# prefetch the 1st faction to cache movq mm2, [esp + nb030_dx1] ;# fetch dr movd mm3, [esp + nb030_dz1] ;# update Vvdwtot pfadd mm5, [esp + nb030_Vvdwtot] ;# add the earlier value movq [esp + nb030_Vvdwtot], mm5 ;# store the sum prefetchw [edi + ebx*4] ;# prefetch the 2nd faction to cache pfmul mm2, mm0 ;# mult by fs pfmul mm3, mm0 movq mm4, [esp + nb030_dx2] ;# fetch dr movd mm5, [esp + nb030_dz2] pfmul mm4, mm1 ;# mult by fs pfmul mm5, mm1 ;# update i forces movq mm0, [esp + nb030_fix] movd mm1, [esp + nb030_fiz] pfadd mm0, mm2 pfadd mm1, mm3 pfadd mm0, mm4 pfadd mm1, mm5 movq [esp + nb030_fix], mm0 movd [esp + nb030_fiz], mm1 ;# update j forces movq mm0, [edi + eax*4] movd mm1, [edi + eax*4 + 8] movq mm6, [edi + ebx*4] movd mm7, [edi + ebx*4 + 8] pfsub mm0, mm2 pfsub mm1, mm3 pfsub mm6, mm4 pfsub mm7, mm5 movq [edi + eax*4], mm0 movd [edi + eax*4 +8], mm1 movq [edi + ebx*4], mm6 movd [edi + ebx*4 + 8], mm7 ;# should we do one more iteration? sub dword ptr [esp + nb030_innerk], 2 jl .nb030_finish_inner jmp .nb030_unroll_loop.nb030_finish_inner: and dword ptr [esp + nb030_innerk], 1 jnz .nb030_single_inner jmp .nb030_updateouterdata .nb030_single_inner: ;# a single j particle iteration here - compare with the unrolled code for comments mov eax, [esp + nb030_innerjjnr] mov eax, [eax] ;# eax=jnr offset mov esi, [ebp + nb030_vdwparam] mov ecx, [ebp + nb030_type] mov edx, [ecx + eax*4] ;# type [jnr1] shl edx, 1 add edx, [esp + nb030_ntia] ;# tja = ntia + 2*type movd mm5, [esi + edx*4] ;# mm5 = 1st c6 movq [esp + nb030_c6], mm5 movd mm5, [esi + edx*4 + 4] ;# mm5 = 1st c12 movq [esp + nb030_c12], mm5 mov esi, [ebp + nb030_pos] lea eax, [eax + eax*2] movq mm0, [esp + nb030_ix] movd mm1, [esp + nb030_iz] movq mm4, [esi + eax*4] movd mm5, [esi + eax*4 + 8] pfsubr mm4, mm0 pfsubr mm5, mm1 movq [esp + nb030_dx1], mm4 pfmul mm4,mm4 movd [esp + nb030_dz1], mm5 pfmul mm5,mm5 pfacc mm4, mm5 pfacc mm4, mm5 ;# mm0=rsq pfrsqrt mm0,mm4 movq mm2,mm0 pfmul mm0,mm0 pfrsqit1 mm0,mm4 pfrcpit2 mm0,mm2 ;# mm1=invsqrt pfmul mm4, mm0 movq mm1, mm4 ;# mm0 is invsqrt, and mm1 r ;# calculate potentials and scalar force pfmul mm1, [esp + nb030_tsc] ;# mm1=rt pf2iw mm4,mm1 movd [esp + nb030_n1], mm4 pi2fd mm4,mm4 pfsub mm1, mm4 ;# now mm1 is eps and mm4 is n0 movq mm2,mm1 pfmul mm2,mm2 ;# mm1 is eps, mm2 is eps2 mov edx, [ebp + nb030_VFtab] mov ecx, [esp + nb030_n1] shl ecx, 3 ;# dispersion table ;# load all the table values we need movd mm4, [edx + ecx*4] movd mm5, [edx + ecx*4 + 4] movd mm6, [edx + ecx*4 + 8] movd mm7, [edx + ecx*4 + 12] pfmul mm6, mm1 ;# mm6 = Geps pfmul mm7, mm2 ;# mm7 = Heps2 pfadd mm5, mm6 pfadd mm5, mm7 ;# mm5 = Fp pfmul mm7, [esp + nb030_two] ;# two*Heps2 pfadd mm7, mm6 pfadd mm7, mm5 ;# mm7=FF pfmul mm5, mm1 ;# mm5=eps*Fp pfadd mm5, mm4 ;# mm5= VV movq mm4, [esp + nb030_c6] pfmul mm7, mm4 ;# fijD pfmul mm5, mm4 ;# Vvdw6 movq mm3, mm7 ;# add to fscal ;# update Vvdwtot to release mm5! pfadd mm5, [esp + nb030_Vvdwtot] ;# add the earlier value movq [esp + nb030_Vvdwtot], mm5 ;# store the sum ;# repulsion table ;# load all the table values we need movd mm4, [edx + ecx*4 + 16] movd mm5, [edx + ecx*4 + 20] movd mm6, [edx + ecx*4 + 24] movd mm7, [edx + ecx*4 + 28] pfmul mm6, mm1 ;# mm6 = Geps pfmul mm7, mm2 ;# mm7 = Heps2 pfadd mm5, mm6 pfadd mm5, mm7 ;# mm5 = Fp pfmul mm7, [esp + nb030_two] ;# two*Heps2 pfadd mm7, mm6 pfadd mm7, mm5 ;# mm7=FF pfmul mm5, mm1 ;# mm5=eps*Fp pfadd mm5, mm4 ;# mm5= VV movq mm6, [esp + nb030_c12] pfmul mm7, mm6 ;# fijR pfmul mm5, mm6 ;# Vvdw12 pfadd mm3, mm7 ;# total fscal fijC+ fijD+ fijR ;# change sign of mm3 pxor mm1,mm1 pfsub mm1, mm3 pfmul mm0, [esp + nb030_tsc] pfmul mm0, mm1 ;# mm0 is total fscal now ;# update Vvdwtot pfadd mm5, [esp + nb030_Vvdwtot] ;# add the earlier value movq [esp + nb030_Vvdwtot], mm5 ;# store the sum ;# spread fscalar to both positions punpckldq mm0,mm0 ;# calc vectorial force prefetchw [edi + eax*4] ;# prefetch faction to cache movq mm2, [esp + nb030_dx1] movd mm3, [esp + nb030_dz1] pfmul mm2, mm0 pfmul mm3, mm0 ;# update i particle force movq mm0, [esp + nb030_fix] movd mm1, [esp + nb030_fiz] pfadd mm0, mm2 pfadd mm1, mm3 movq [esp + nb030_fix], mm0 movd [esp + nb030_fiz], mm1 ;# update j particle force movq mm0, [edi + eax*4] movd mm1, [edi + eax*4 + 8] pfsub mm0, mm2 pfsub mm1, mm3 movq [edi + eax*4], mm0 movd [edi + eax*4 +8], mm1 ;# done! .nb030_updateouterdata: mov ecx, [esp + nb030_ii3] movq mm6, [edi + ecx*4] ;# increment i force movd mm7, [edi + ecx*4 + 8] pfadd mm6, [esp + nb030_fix] pfadd mm7, [esp + nb030_fiz] movq [edi + ecx*4], mm6 movd [edi + ecx*4 +8], mm7 mov ebx, [ebp + nb030_fshift] ;# increment fshift force mov edx, [esp + nb030_is3] movq mm6, [ebx + edx*4] movd mm7, [ebx + edx*4 + 8] pfadd mm6, [esp + nb030_fix] pfadd mm7, [esp + nb030_fiz] movq [ebx + edx*4], mm6 movd [ebx + edx*4 + 8], mm7 ;# get n from stack mov esi, [esp + nb030_n] ;# get group index for i particle mov edx, [ebp + nb030_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] movq mm7, [esp + nb030_Vvdwtot] pfacc mm7,mm7 ;# get and sum the two parts of total potential mov eax, [ebp + nb030_Vvdw] movd mm6, [eax + edx*4] pfadd mm6, mm7 movd [eax + edx*4], mm6 ;# increment Vvdw[gid] ;# finish if last mov ecx, [esp + nb030_nn1] ;# esi already loaded with n inc esi sub ecx, esi jecxz .nb030_outerend ;# not last, iterate outer loop once more! mov [esp + nb030_n], esi jmp .nb030_outer.nb030_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb030_nri] ;# esi already loaded with n above sub ecx, esi jecxz .nb030_end ;# non-zero, do one more workunit jmp .nb030_threadloop.nb030_end: femms mov eax, [esp + nb030_nouter] mov ebx, [esp + nb030_ninner] mov ecx, [ebp + nb030_outeriter] mov edx, [ebp + nb030_inneriter] mov [ecx], eax mov [edx], ebx add esp, 140 pop edi pop esi pop edx pop ecx pop ebx pop eax leave ret .globl nb_kernel030nf_ia32_3dnow.globl _nb_kernel030nf_ia32_3dnownb_kernel030nf_ia32_3dnow: _nb_kernel030nf_ia32_3dnow: .equiv nb030nf_p_nri, 8.equiv nb030nf_iinr, 12.equiv nb030nf_jindex, 16.equiv nb030nf_jjnr, 20.equiv nb030nf_shift, 24.equiv nb030nf_shiftvec, 28.equiv nb030nf_fshift, 32.equiv nb030nf_gid, 36.equiv nb030nf_pos, 40 .equiv nb030nf_faction, 44.equiv nb030nf_charge, 48.equiv nb030nf_p_facel, 52.equiv nb030nf_p_krf, 56 .equiv nb030nf_p_crf, 60 .equiv nb030nf_Vc, 64 .equiv nb030nf_type, 68.equiv nb030nf_p_ntype, 72.equiv nb030nf_vdwparam, 76 .equiv nb030nf_Vvdw, 80 .equiv nb030nf_p_tabscale, 84 .equiv nb030nf_VFtab, 88.equiv nb030nf_invsqrta, 92 .equiv nb030nf_dvda, 96.equiv nb030nf_p_gbtabscale, 100.equiv nb030nf_GBtab, 104.equiv nb030nf_p_nthreads, 108.equiv nb030nf_count, 112.equiv nb030nf_mtx, 116.equiv nb030nf_outeriter, 120.equiv nb030nf_inneriter, 124.equiv nb030nf_work, 128 ;# stack offsets for local variables .equiv nb030nf_is3, 0.equiv nb030nf_ii3, 4.equiv nb030nf_ix, 8.equiv nb030nf_iy, 12.equiv nb030nf_iz, 16.equiv nb030nf_Vvdwtot, 20 .equiv nb030nf_c6, 28 .equiv nb030nf_c12, 36.equiv nb030nf_n1, 44 .equiv nb030nf_tsc, 52 .equiv nb030nf_ntia, 60.equiv nb030nf_innerjjnr, 64.equiv nb030nf_innerk, 68.equiv nb030nf_n, 72 ;# idx for outer loop.equiv nb030nf_nn1, 76 ;# number of outer iterations.equiv nb030nf_nri, 80.equiv nb030nf_ntype, 84.equiv nb030nf_nouter, 88.equiv nb030nf_ninner, 92 push ebp mov ebp,esp push eax push ebx push ecx push edx push esi push edi sub esp, 96 ;# local stack space femms ;# move data to local stack mov ecx, [ebp + nb030nf_p_nri] mov edx, [ebp + nb030nf_p_ntype] mov esi, [ebp + nb030nf_p_tabscale] mov ecx, [ecx] mov edx, [edx] mov [esp + nb030nf_nri], ecx mov [esp + nb030nf_ntype], edx movd mm3, [esi] punpckldq mm3,mm3 movq [esp + nb030nf_tsc], mm3 ;# zero iteration counters mov eax, 0 mov [esp + nb030nf_nouter], eax⌨️ 快捷键说明
复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?