nb_kernel212_ia32_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,151 行 · 第 1/5 页
S
2,151 行
mulss xmm2, [esp + nb212_twelve] movaps [esp + nb212_Vvdwtot], xmm4 subss xmm2, xmm1 ;# fsD+ fsR addps xmm2, xmm3 ;# fsC+ fsD+ fsR addps xmm6, [esp + nb212_vctot] mulps xmm0, xmm2 ;# total fscal movaps [esp + nb212_vctot], xmm6 movaps xmm1, xmm0 movaps xmm2, xmm0 mulps xmm0, [esp + nb212_dxOO] mulps xmm1, [esp + nb212_dyOO] mulps xmm2, [esp + nb212_dzOO] ;# initial update for j forces xorps xmm3, xmm3 xorps xmm4, xmm4 xorps xmm5, xmm5 subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 movaps [esp + nb212_fjxO], xmm3 movaps [esp + nb212_fjyO], xmm4 movaps [esp + nb212_fjzO], xmm5 addps xmm0, [esp + nb212_fixO] addps xmm1, [esp + nb212_fiyO] addps xmm2, [esp + nb212_fizO] movaps [esp + nb212_fixO], xmm0 movaps [esp + nb212_fiyO], xmm1 movaps [esp + nb212_fizO], xmm2 ;# done with i O Now do i H1 & H2 simultaneously first get i particle coords: movaps xmm0, [esp + nb212_ixH1] movaps xmm1, [esp + nb212_iyH1] movaps xmm2, [esp + nb212_izH1] movaps xmm3, [esp + nb212_ixH2] movaps xmm4, [esp + nb212_iyH2] movaps xmm5, [esp + nb212_izH2] subps xmm0, [esp + nb212_jxO] subps xmm1, [esp + nb212_jyO] subps xmm2, [esp + nb212_jzO] subps xmm3, [esp + nb212_jxO] subps xmm4, [esp + nb212_jyO] subps xmm5, [esp + nb212_jzO] movaps [esp + nb212_dxH1O], xmm0 movaps [esp + nb212_dyH1O], xmm1 movaps [esp + nb212_dzH1O], xmm2 movaps [esp + nb212_dxH2O], xmm3 movaps [esp + nb212_dyH2O], xmm4 movaps [esp + nb212_dzH2O], xmm5 mulps xmm0, xmm0 mulps xmm1, xmm1 mulps xmm2, xmm2 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 addps xmm0, xmm1 addps xmm4, xmm3 addps xmm0, xmm2 ;# have rsqH1 in xmm0 addps xmm4, xmm5 ;# have rsqH2 in xmm4 ;# do invsqrt rsqrtps xmm1, xmm0 rsqrtps xmm5, xmm4 movaps xmm2, xmm1 movaps xmm6, xmm5 mulps xmm1, xmm1 mulps xmm5, xmm5 movaps xmm3, [esp + nb212_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 + nb212_half] ;# rinv H1 - j water mulps xmm7, [esp + nb212_half] ;# rinv H2 - j water mulps xmm0, [esp + nb212_krf] ;# krsq mulps xmm4, [esp + nb212_krf] ;# krsq ;# assemble charges in xmm6 xorps xmm6, xmm6 movss xmm6, [esp + nb212_qqOH] movhps xmm6, [esp + nb212_qqHH] movaps xmm1, xmm0 movaps xmm5, xmm4 addps xmm0, xmm3 ;# krsq+ rinv addps xmm4, xmm7 ;# krsq+ rinv subps xmm0, [esp + nb212_crf] subps xmm4, [esp + nb212_crf] mulps xmm1, [esp + nb212_two] mulps xmm5, [esp + nb212_two] mulps xmm0, xmm6 ;# vcoul mulps xmm4, xmm6 ;# vcoul addps xmm4, xmm0 addps xmm4, [esp + nb212_vctot] movaps [esp + nb212_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 + nb212_dxH1O] mulps xmm1, [esp + nb212_dyH1O] mulps xmm2, [esp + nb212_dzH1O] ;# update forces H1 - j water movaps xmm3, [esp + nb212_fjxO] movaps xmm4, [esp + nb212_fjyO] movaps xmm5, [esp + nb212_fjzO] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 movaps [esp + nb212_fjxO], xmm3 movaps [esp + nb212_fjyO], xmm4 movaps [esp + nb212_fjzO], xmm5 addps xmm0, [esp + nb212_fixH1] addps xmm1, [esp + nb212_fiyH1] addps xmm2, [esp + nb212_fizH1] movaps [esp + nb212_fixH1], xmm0 movaps [esp + nb212_fiyH1], xmm1 movaps [esp + nb212_fizH1], xmm2 ;# do forces H2 - j water movaps xmm0, xmm7 movaps xmm1, xmm7 movaps xmm2, xmm7 mulps xmm0, [esp + nb212_dxH2O] mulps xmm1, [esp + nb212_dyH2O] mulps xmm2, [esp + nb212_dzH2O] movaps xmm3, [esp + nb212_fjxO] movaps xmm4, [esp + nb212_fjyO] movaps xmm5, [esp + nb212_fjzO] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 mov esi, [ebp + nb212_faction] movaps [esp + nb212_fjxO], xmm3 movaps [esp + nb212_fjyO], xmm4 movaps [esp + nb212_fjzO], xmm5 addps xmm0, [esp + nb212_fixH2] addps xmm1, [esp + nb212_fiyH2] addps xmm2, [esp + nb212_fizH2] movaps [esp + nb212_fixH2], xmm0 movaps [esp + nb212_fiyH2], xmm1 movaps [esp + nb212_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 + nb212_fjxO] movaps xmm4, [esp + nb212_fjyO] movaps xmm5, [esp + nb212_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 + nb212_innerk] jz .nb212_updateouterdata jmp .nb212_single_loop.nb212_updateouterdata: mov ecx, [esp + nb212_ii3] mov edi, [ebp + nb212_faction] mov esi, [ebp + nb212_fshift] mov edx, [esp + nb212_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movaps xmm0, [esp + nb212_fixO] movaps xmm1, [esp + nb212_fiyO] movaps xmm2, [esp + nb212_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 + nb212_fixH1] movaps xmm1, [esp + nb212_fiyH1] movaps xmm2, [esp + nb212_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 + nb212_fixH2] movaps xmm1, [esp + nb212_fiyH2] movaps xmm2, [esp + nb212_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 + nb212_n] ;# get group index for i particle mov edx, [ebp + nb212_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [esp + nb212_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 + nb212_Vc] addss xmm7, [eax + edx*4] ;# move back to mem movss [eax + edx*4], xmm7 ;# accumulate total lj energy and update it movaps xmm7, [esp + nb212_Vvdwtot] ;# 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 + nb212_Vvdw] addss xmm7, [eax + edx*4] ;# move back to mem movss [eax + edx*4], xmm7 ;# finish if last mov ecx, [esp + nb212_nn1] ;# esi already loaded with n inc esi sub ecx, esi jecxz .nb212_outerend ;# not last, iterate outer loop once more! mov [esp + nb212_n], esi jmp .nb212_outer.nb212_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb212_nri] ;# esi already loaded with n above sub ecx, esi jecxz .nb212_end ;# non-zero, do one more workunit jmp .nb212_threadloop.nb212_end: emms mov eax, [esp + nb212_nouter] mov ebx, [esp + nb212_ninner] mov ecx, [ebp + nb212_outeriter] mov edx, [ebp + nb212_inneriter] mov [ecx], eax mov [edx], ebx mov eax, [esp + nb212_salign] add esp, eax add esp, 1560 pop edi pop esi pop edx pop ecx pop ebx pop eax leave ret.globl nb_kernel212nf_ia32_sse.globl _nb_kernel212nf_ia32_ssenb_kernel212nf_ia32_sse: _nb_kernel212nf_ia32_sse: .equiv nb212nf_p_nri, 8.equiv nb212nf_iinr, 12.equiv nb212nf_jindex, 16.equiv nb212nf_jjnr, 20.equiv nb212nf_shift, 24.equiv nb212nf_shiftvec, 28.equiv nb212nf_fshift, 32.equiv nb212nf_gid, 36.equiv nb212nf_pos, 40.equiv nb212nf_faction, 44.equiv nb212nf_charge, 48.equiv nb212nf_p_facel, 52.equiv nb212nf_argkrf, 56.equiv nb212nf_argcrf, 60.equiv nb212nf_Vc, 64.equiv nb212nf_type, 68.equiv nb212nf_p_ntype, 72.equiv nb212nf_vdwparam, 76.equiv nb212nf_Vvdw, 80.equiv nb212nf_p_tabscale, 84.equiv nb212nf_VFtab, 88.equiv nb212nf_invsqrta, 92.equiv nb212nf_dvda, 96.equiv nb212nf_p_gbtabscale, 100.equiv nb212nf_GBtab, 104.equiv nb212nf_p_nthreads, 108.equiv nb212nf_count, 112.equiv nb212nf_mtx, 116.equiv nb212nf_outeriter, 120⌨️ 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?