nb_kernel202_ia32_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,166 行 · 第 1/5 页
S
2,166 行
rsqrtps xmm5, xmm4 movaps xmm2, xmm1 movaps xmm6, xmm5 mulps xmm1, xmm1 mulps xmm5, xmm5 movaps xmm3, [esp + nb202_three] movaps xmm7, xmm3 mulps xmm1, xmm0 mulps xmm5, xmm4 subps xmm3, xmm1 subps xmm7, xmm5 mulps xmm3, xmm2 mulps xmm7, xmm6 mulps xmm3, [esp + nb202_half] ;# rinv H1 - j water mulps xmm7, [esp + nb202_half] ;# rinv H2 - j water mulps xmm0, [esp + nb202_krf] ;# krsq mulps xmm4, [esp + nb202_krf] ;# krsq ;# assemble charges in xmm6 xorps xmm6, xmm6 movss xmm6, [esp + nb202_qqOH] movhps xmm6, [esp + nb202_qqHH] movaps xmm1, xmm0 movaps xmm5, xmm4 addps xmm0, xmm3 ;# krsq+ rinv addps xmm4, xmm7 ;# krsq+ rinv subps xmm0, [esp + nb202_crf] subps xmm4, [esp + nb202_crf] mulps xmm1, [esp + nb202_two] mulps xmm5, [esp + nb202_two] mulps xmm0, xmm6 ;# vcoul mulps xmm4, xmm6 ;# vcoul addps xmm4, xmm0 addps xmm4, [esp + nb202_vctot] movaps [esp + nb202_vctot], xmm4 movaps xmm0, xmm3 movaps xmm4, xmm7 mulps xmm3, xmm3 mulps xmm7, xmm7 subps xmm0, xmm1 subps xmm4, xmm5 mulps xmm0, xmm6 mulps xmm4, xmm6 mulps xmm0, xmm3 ;# fscal mulps xmm7, xmm4 ;# fscal movaps xmm1, xmm0 movaps xmm2, xmm0 mulps xmm0, [esp + nb202_dxH1O] mulps xmm1, [esp + nb202_dyH1O] mulps xmm2, [esp + nb202_dzH1O] ;# update forces H1 - j water movaps xmm3, [esp + nb202_fjxO] movaps xmm4, [esp + nb202_fjyO] movaps xmm5, [esp + nb202_fjzO] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 movaps [esp + nb202_fjxO], xmm3 movaps [esp + nb202_fjyO], xmm4 movaps [esp + nb202_fjzO], xmm5 addps xmm0, [esp + nb202_fixH1] addps xmm1, [esp + nb202_fiyH1] addps xmm2, [esp + nb202_fizH1] movaps [esp + nb202_fixH1], xmm0 movaps [esp + nb202_fiyH1], xmm1 movaps [esp + nb202_fizH1], xmm2 ;# do forces H2 - j water movaps xmm0, xmm7 movaps xmm1, xmm7 movaps xmm2, xmm7 mulps xmm0, [esp + nb202_dxH2O] mulps xmm1, [esp + nb202_dyH2O] mulps xmm2, [esp + nb202_dzH2O] movaps xmm3, [esp + nb202_fjxO] movaps xmm4, [esp + nb202_fjyO] movaps xmm5, [esp + nb202_fjzO] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 mov esi, [ebp + nb202_faction] movaps [esp + nb202_fjxO], xmm3 movaps [esp + nb202_fjyO], xmm4 movaps [esp + nb202_fjzO], xmm5 addps xmm0, [esp + nb202_fixH2] addps xmm1, [esp + nb202_fiyH2] addps xmm2, [esp + nb202_fizH2] movaps [esp + nb202_fixH2], xmm0 movaps [esp + nb202_fiyH2], xmm1 movaps [esp + nb202_fizH2], xmm2 ;# update j water forces from local variables movlps xmm0, [esi + eax*4] movlps xmm1, [esi + eax*4 + 12] movhps xmm1, [esi + eax*4 + 24] movaps xmm3, [esp + nb202_fjxO] movaps xmm4, [esp + nb202_fjyO] movaps xmm5, [esp + nb202_fjzO] movaps xmm6, xmm5 movaps xmm7, xmm5 shufps xmm6, xmm6, 2 ;# constant 00000010 shufps xmm7, xmm7, 3 ;# constant 00000011 addss xmm5, [esi + eax*4 + 8] addss xmm6, [esi + eax*4 + 20] addss xmm7, [esi + eax*4 + 32] movss [esi + eax*4 + 8], xmm5 movss [esi + eax*4 + 20], xmm6 movss [esi + eax*4 + 32], xmm7 movaps xmm5, xmm3 unpcklps xmm3, xmm4 unpckhps xmm5, xmm4 addps xmm0, xmm3 addps xmm1, xmm5 movlps [esi + eax*4], xmm0 movlps [esi + eax*4 + 12], xmm1 movhps [esi + eax*4 + 24], xmm1 dec dword ptr [esp + nb202_innerk] jz .nb202_updateouterdata jmp .nb202_single_loop.nb202_updateouterdata: mov ecx, [esp + nb202_ii3] mov edi, [ebp + nb202_faction] mov esi, [ebp + nb202_fshift] mov edx, [esp + nb202_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movaps xmm0, [esp + nb202_fixO] movaps xmm1, [esp + nb202_fiyO] movaps xmm2, [esp + nb202_fizO] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [edi + ecx*4] movss xmm4, [edi + ecx*4 + 4] movss xmm5, [edi + ecx*4 + 8] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [edi + ecx*4], xmm3 movss [edi + ecx*4 + 4], xmm4 movss [edi + ecx*4 + 8], xmm5 ;# accumulate force in xmm6/xmm7 for fshift movaps xmm6, xmm0 movss xmm7, xmm2 movlhps xmm6, xmm1 shufps xmm6, xmm6, 8 ;# constant 00001000 ;# accumulate H1i forces in xmm0, xmm1, xmm2 movaps xmm0, [esp + nb202_fixH1] movaps xmm1, [esp + nb202_fiyH1] movaps xmm2, [esp + nb202_fizH1] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [edi + ecx*4 + 12] movss xmm4, [edi + ecx*4 + 16] movss xmm5, [edi + ecx*4 + 20] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [edi + ecx*4 + 12], xmm3 movss [edi + ecx*4 + 16], xmm4 movss [edi + ecx*4 + 20], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addss xmm7, xmm2 movlhps xmm0, xmm1 shufps xmm0, xmm0, 8 ;# constant 00001000 addps xmm6, xmm0 ;# accumulate H2i forces in xmm0, xmm1, xmm2 movaps xmm0, [esp + nb202_fixH2] movaps xmm1, [esp + nb202_fiyH2] movaps xmm2, [esp + nb202_fizH2] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [edi + ecx*4 + 24] movss xmm4, [edi + ecx*4 + 28] movss xmm5, [edi + ecx*4 + 32] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [edi + ecx*4 + 24], xmm3 movss [edi + ecx*4 + 28], xmm4 movss [edi + ecx*4 + 32], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addss xmm7, xmm2 movlhps xmm0, xmm1 shufps xmm0, xmm0, 8 ;# constant 00001000 addps xmm6, xmm0 ;# increment fshift force movlps xmm3, [esi + edx*4] movss xmm4, [esi + edx*4 + 8] addps xmm3, xmm6 addss xmm4, xmm7 movlps [esi + edx*4], xmm3 movss [esi + edx*4 + 8], xmm4 ;# get n from stack mov esi, [esp + nb202_n] ;# get group index for i particle mov edx, [ebp + nb202_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [esp + nb202_vctot] ;# accumulate movhlps xmm6, xmm7 addps xmm7, xmm6 ;# pos 0-1 in xmm7 have the sum now movaps xmm6, xmm7 shufps xmm6, xmm6, 1 addss xmm7, xmm6 ;# add earlier value from mem mov eax, [ebp + nb202_Vc] addss xmm7, [eax + edx*4] ;# move back to mem movss [eax + edx*4], xmm7 ;# finish if last mov ecx, [esp + nb202_nn1] ;# esi already loaded with n inc esi sub ecx, esi jecxz .nb202_outerend ;# not last, iterate outer loop once more! mov [esp + nb202_n], esi jmp .nb202_outer.nb202_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb202_nri] ;# esi already loaded with n above sub ecx, esi jecxz .nb202_end ;# non-zero, do one more workunit jmp .nb202_threadloop.nb202_end: emms mov eax, [esp + nb202_nouter] mov ebx, [esp + nb202_ninner] mov ecx, [ebp + nb202_outeriter] mov edx, [ebp + nb202_inneriter] mov [ecx], eax mov [edx], ebx mov eax, [esp + nb202_salign] add esp, eax add esp, 1480 pop edi pop esi pop edx pop ecx pop ebx pop eax leave ret .globl nb_kernel202nf_ia32_sse.globl _nb_kernel202nf_ia32_ssenb_kernel202nf_ia32_sse: _nb_kernel202nf_ia32_sse: .equiv nb202nf_p_nri, 8.equiv nb202nf_iinr, 12.equiv nb202nf_jindex, 16.equiv nb202nf_jjnr, 20.equiv nb202nf_shift, 24.equiv nb202nf_shiftvec, 28.equiv nb202nf_fshift, 32.equiv nb202nf_gid, 36.equiv nb202nf_pos, 40.equiv nb202nf_faction, 44.equiv nb202nf_charge, 48.equiv nb202nf_p_facel, 52.equiv nb202nf_argkrf, 56.equiv nb202nf_argcrf, 60.equiv nb202nf_Vc, 64.equiv nb202nf_type, 68.equiv nb202nf_p_ntype, 72.equiv nb202nf_vdwparam, 76.equiv nb202nf_Vvdw, 80.equiv nb202nf_p_tabscale, 84.equiv nb202nf_VFtab, 88.equiv nb202nf_invsqrta, 92.equiv nb202nf_dvda, 96.equiv nb202nf_p_gbtabscale, 100.equiv nb202nf_GBtab, 104.equiv nb202nf_p_nthreads, 108.equiv nb202nf_count, 112.equiv nb202nf_mtx, 116.equiv nb202nf_outeriter, 120.equiv nb202nf_inneriter, 124.equiv nb202nf_work, 128 ;# stack offsets for local variables ;# bottom of stack is cache-aligned for sse use .equiv nb202nf_ixO, 0.equiv nb202nf_iyO, 16.equiv nb202nf_izO, 32.equiv nb202nf_ixH1, 48.equiv nb202nf_iyH1, 64.equiv nb202nf_izH1, 80.equiv nb202nf_ixH2, 96.equiv nb202nf_iyH2, 112.equiv nb202nf_izH2, 128.equiv nb202nf_jxO, 144.equiv nb202nf_jyO, 160.equiv nb202nf_jzO, 176.equiv nb202nf_jxH1, 192.equiv nb202nf_jyH1, 208.equiv nb202nf_jzH1, 224.equiv nb202nf_jxH2, 240.equiv nb202nf_jyH2, 256.equiv nb202nf_jzH2, 272.equiv nb202nf_qqOO, 288.equiv nb202nf_qqOH, 304.equiv nb202nf_qqHH, 320.equiv nb202nf_vctot, 336.equiv nb202nf_half, 352.equiv nb202nf_three, 368.equiv nb202nf_rsqOO, 384.equiv nb202nf_rsqOH1, 400.equiv nb202nf_rsqOH2, 416.equiv nb202nf_rsqH1O, 432.equiv nb202nf_rsqH1H1, 448.equiv nb202nf_rsqH1H2, 464.equiv nb202nf_rsqH2O, 480.equiv nb202nf_rsqH2H1, 496.equiv nb202nf_rsqH2H2, 512.equiv nb202nf_rinvOO, 528.equiv nb202nf_rinvOH1, 544.equiv nb202nf_rinvOH2, 560.equiv nb202nf_rinvH1O, 576.equiv nb202nf_rinvH1H1, 592.equiv nb202nf_rinvH1H2, 608.equiv nb202nf_rinvH2O, 624.equiv nb202nf_rinvH2H1, 640.equiv nb202nf_rinvH2H2, 656.equiv nb202nf_krf, 672.equiv nb202nf_crf, 688.equiv nb202nf_is3, 704.equiv nb202nf_ii3, 708.equiv nb202nf_innerjjnr, 712.equiv nb202nf_innerk, 716.equiv nb202nf_n, 720.equiv nb202nf_nn1, 724.equiv nb202nf_nri, 728.equiv nb202nf_nouter, 732.equiv nb202nf_ninner, 736.equiv nb202nf_salign, 740 push ebp mov ebp,esp push eax push ebx push ecx push edx push esi push edi sub esp, 744 ;# local stack space mov eax, esp and eax, 0xf sub esp, eax mov [esp + nb202nf_salign], eax emms ;# Move args passed by reference to stack mov ecx, [ebp + nb202nf_p_nri] mov ecx, [ecx] mov [esp + nb202nf_nri], ecx⌨️ 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?