nb_kernel101_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,668 行 · 第 1/3 页
S
1,668 行
movapd xmm1, [esp + nb101_dyH2] movapd xmm2, [esp + nb101_dzH2] movapd [esp + nb101_vctot], xmm5 mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 ;# update H2 forces movapd xmm3, [esp + nb101_fixH2] movapd xmm4, [esp + nb101_fiyH2] movapd xmm7, [esp + nb101_fizH2] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb101_fixH2], xmm3 movapd [esp + nb101_fiyH2], xmm4 movapd [esp + nb101_fizH2], xmm7 mov edi, [ebp + nb101_faction] ;# update j forces addpd xmm0, [esp + nb101_fjx] addpd xmm1, [esp + nb101_fjy] addpd xmm2, [esp + nb101_fjz] movlpd xmm3, [edi + eax*8] movlpd xmm4, [edi + eax*8 + 8] movlpd xmm5, [edi + eax*8 + 16] movhpd xmm3, [edi + ebx*8] movhpd xmm4, [edi + ebx*8 + 8] movhpd xmm5, [edi + ebx*8 + 16] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 movlpd [edi + eax*8], xmm3 movlpd [edi + eax*8 + 8], xmm4 movlpd [edi + eax*8 + 16], xmm5 movhpd [edi + ebx*8], xmm3 movhpd [edi + ebx*8 + 8], xmm4 movhpd [edi + ebx*8 + 16], xmm5 ;# should we do one more iteration? sub dword ptr [esp + nb101_innerk], 2 jl .nb101_checksingle jmp .nb101_unroll_loop.nb101_checksingle: mov edx, [esp + nb101_innerk] and edx, 1 jnz .nb101_dosingle jmp .nb101_updateouterdata.nb101_dosingle: mov edx, [esp + nb101_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] mov esi, [ebp + nb101_charge] ;# base of charge[] xorpd xmm6, xmm6 movlpd xmm6, [esi + eax*8] ;# jq A movapd xmm3, [esp + nb101_iqO] movapd xmm4, [esp + nb101_iqH] mulsd xmm3, xmm6 ;# qqO mulsd xmm4, xmm6 ;# qqH movapd [esp + nb101_qqO], xmm3 movapd [esp + nb101_qqH], xmm4 mov esi, [ebp + nb101_pos] ;# base of pos[] lea eax, [eax + eax*2] ;# replace jnr with j3 ;# move coordinates to xmm0-xmm2 movlpd xmm0, [esi + eax*8] movlpd xmm1, [esi + eax*8 + 8] movlpd xmm2, [esi + eax*8 + 16] ;# move ixO-izO to xmm4-xmm6 movapd xmm4, [esp + nb101_ixO] movapd xmm5, [esp + nb101_iyO] movapd xmm6, [esp + nb101_izO] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# store dr movapd [esp + nb101_dxO], xmm4 movapd [esp + nb101_dyO], xmm5 movapd [esp + nb101_dzO], xmm6 ;# square it mulsd xmm4,xmm4 mulsd xmm5,xmm5 mulsd xmm6,xmm6 addsd xmm4, xmm5 addsd xmm4, xmm6 movapd xmm7, xmm4 ;# rsqO in xmm7 ;# move ixH1-izH1 to xmm4-xmm6 movapd xmm4, [esp + nb101_ixH1] movapd xmm5, [esp + nb101_iyH1] movapd xmm6, [esp + nb101_izH1] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# store dr movapd [esp + nb101_dxH1], xmm4 movapd [esp + nb101_dyH1], xmm5 movapd [esp + nb101_dzH1], xmm6 ;# square it mulsd xmm4,xmm4 mulsd xmm5,xmm5 mulsd xmm6,xmm6 addsd xmm6, xmm5 addsd xmm6, xmm4 ;# rsqH1 in xmm6 ;# move ixH2-izH2 to xmm3-xmm5 movapd xmm3, [esp + nb101_ixH2] movapd xmm4, [esp + nb101_iyH2] movapd xmm5, [esp + nb101_izH2] ;# calc dr subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 ;# store dr movapd [esp + nb101_dxH2], xmm3 movapd [esp + nb101_dyH2], xmm4 movapd [esp + nb101_dzH2], xmm5 ;# square it mulsd xmm3,xmm3 mulsd xmm4,xmm4 mulsd xmm5,xmm5 addsd xmm5, xmm4 addsd xmm5, xmm3 ;# rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ;# start with rsqO - put seed in xmm2 cvtsd2ss xmm2, xmm7 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [esp + nb101_three] mulsd xmm2, xmm7 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [esp + nb101_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm7, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb101_three] subsd xmm4, xmm7 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb101_half] ;# rinv movapd xmm7, xmm4 ;# rinvO in xmm7 ;# rsqH1 - seed in xmm2 cvtsd2ss xmm2, xmm6 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [esp + nb101_three] mulsd xmm2, xmm6 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [esp + nb101_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm6, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb101_three] subsd xmm4, xmm6 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb101_half] ;# rinv movapd xmm6, xmm4 ;# rinvH1 in xmm6 ;# rsqH2 - seed in xmm2 cvtsd2ss xmm2, xmm5 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [esp + nb101_three] mulsd xmm2, xmm5 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [esp + nb101_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm5, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb101_three] subsd xmm4, xmm5 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb101_half] ;# rinv movapd xmm5, xmm4 ;# rinvH2 in xmm5 ;# do O interactions movapd xmm4, xmm7 mulsd xmm4, xmm4 ;# xmm7=rinv, xmm4=rinvsq mulsd xmm7, [esp + nb101_qqO] ;# xmm7=vcoul mulsd xmm4, xmm7 ;# total fsO in xmm4 addsd xmm7, [esp + nb101_vctot] movlpd [esp + nb101_vctot], xmm7 movapd xmm0, [esp + nb101_dxO] movapd xmm1, [esp + nb101_dyO] movapd xmm2, [esp + nb101_dzO] mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 ;# update O forces movapd xmm3, [esp + nb101_fixO] movapd xmm4, [esp + nb101_fiyO] movapd xmm7, [esp + nb101_fizO] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb101_fixO], xmm3 movlpd [esp + nb101_fiyO], xmm4 movlpd [esp + nb101_fizO], xmm7 ;# update j forces with water O movlpd [esp + nb101_fjx], xmm0 movlpd [esp + nb101_fjy], xmm1 movlpd [esp + nb101_fjz], xmm2 ;# H1 interactions movapd xmm4, xmm6 mulsd xmm4, xmm4 ;# xmm6=rinv, xmm4=rinvsq mulsd xmm6, [esp + nb101_qqH] ;# xmm6=vcoul mulsd xmm4, xmm6 ;# total fsH1 in xmm4 addsd xmm6, [esp + nb101_vctot] movapd xmm0, [esp + nb101_dxH1] movapd xmm1, [esp + nb101_dyH1] movapd xmm2, [esp + nb101_dzH1] movlpd [esp + nb101_vctot], xmm6 mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 ;# update H1 forces movapd xmm3, [esp + nb101_fixH1] movapd xmm4, [esp + nb101_fiyH1] movapd xmm7, [esp + nb101_fizH1] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb101_fixH1], xmm3 movlpd [esp + nb101_fiyH1], xmm4 movlpd [esp + nb101_fizH1], xmm7 ;# update j forces with water H1 addsd xmm0, [esp + nb101_fjx] addsd xmm1, [esp + nb101_fjy] addsd xmm2, [esp + nb101_fjz] movsd [esp + nb101_fjx], xmm0 movsd [esp + nb101_fjy], xmm1 movsd [esp + nb101_fjz], xmm2 ;# H2 interactions movapd xmm4, xmm5 mulsd xmm4, xmm4 ;# xmm5=rinv, xmm4=rinvsq mulsd xmm5, [esp + nb101_qqH] ;# xmm5=vcoul mulsd xmm4, xmm5 ;# total fsH1 in xmm4 addsd xmm5, [esp + nb101_vctot] movapd xmm0, [esp + nb101_dxH2] movapd xmm1, [esp + nb101_dyH2] movapd xmm2, [esp + nb101_dzH2] movlpd [esp + nb101_vctot], xmm5 mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 ;# update H2 forces movapd xmm3, [esp + nb101_fixH2] movapd xmm4, [esp + nb101_fiyH2] movapd xmm7, [esp + nb101_fizH2] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb101_fixH2], xmm3 movlpd [esp + nb101_fiyH2], xmm4 movlpd [esp + nb101_fizH2], xmm7 mov edi, [ebp + nb101_faction] ;# update j forces addsd xmm0, [esp + nb101_fjx] addsd xmm1, [esp + nb101_fjy] addsd xmm2, [esp + nb101_fjz] movlpd xmm3, [edi + eax*8] movlpd xmm4, [edi + eax*8 + 8] movlpd xmm5, [edi + eax*8 + 16] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 movlpd [edi + eax*8], xmm3 movlpd [edi + eax*8 + 8], xmm4 movlpd [edi + eax*8 + 16], xmm5.nb101_updateouterdata: mov ecx, [esp + nb101_ii3] mov edi, [ebp + nb101_faction] mov esi, [ebp + nb101_fshift] mov edx, [esp + nb101_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movapd xmm0, [esp + nb101_fixO] movapd xmm1, [esp + nb101_fiyO] movapd xmm2, [esp + nb101_fizO] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addsd xmm0, xmm3 addsd xmm1, xmm4 addsd xmm2, xmm5 ;# sum is in low xmm0-xmm2 ;# increment i force movsd xmm3, [edi + ecx*8] movsd xmm4, [edi + ecx*8 + 8] movsd xmm5, [edi + ecx*8 + 16] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm5, xmm2 movsd [edi + ecx*8], xmm3 movsd [edi + ecx*8 + 8], xmm4 movsd [edi + ecx*8 + 16], xmm5 ;# accumulate force in xmm6/xmm7 for fshift movapd xmm6, xmm0 movsd xmm7, xmm2 unpcklpd xmm6, xmm1 ;# accumulate H1i forces in xmm0, xmm1, xmm2 movapd xmm0, [esp + nb101_fixH1] movapd xmm1, [esp + nb101_fiyH1] movapd xmm2, [esp + nb101_fizH1] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addsd xmm0, xmm3 addsd xmm1, xmm4 addsd xmm2, xmm5 ;# sum is in low xmm0-xmm2 ;# increment i force movsd xmm3, [edi + ecx*8 + 24] movsd xmm4, [edi + ecx*8 + 32] movsd xmm5, [edi + ecx*8 + 40] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm5, xmm2 movsd [edi + ecx*8 + 24], xmm3 movsd [edi + ecx*8 + 32], xmm4 movsd [edi + ecx*8 + 40], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addsd xmm7, xmm2 unpcklpd xmm0, xmm1 addpd xmm6, xmm0 ;# accumulate H2i forces in xmm0, xmm1, xmm2 movapd xmm0, [esp + nb101_fixH2] movapd xmm1, [esp + nb101_fiyH2] movapd xmm2, [esp + nb101_fizH2] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addsd xmm0, xmm3 addsd xmm1, xmm4 addsd xmm2, xmm5 ;# sum is in low xmm0-xmm2 movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 ;# increment i force movsd xmm3, [edi + ecx*8 + 48] movsd xmm4, [edi + ecx*8 + 56] movsd xmm5, [edi + ecx*8 + 64] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm5, xmm2 movsd [edi + ecx*8 + 48], xmm3 movsd [edi + ecx*8 + 56], xmm4 movsd [edi + ecx*8 + 64], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addsd xmm7, xmm2 unpcklpd xmm0, xmm1 addpd xmm6, xmm0 ;# increment fshift force movlpd xmm3, [esi + edx*8] movhpd xmm3, [esi + edx*8 + 8] movsd xmm4, [esi + edx*8 + 16] addpd xmm3, xmm6 addsd xmm4, xmm7 movlpd [esi + edx*8], xmm3 movhpd [esi + edx*8 + 8], xmm3 movsd [esi + edx*8 + 16], xmm4 ;# get n from stack mov esi, [esp + nb101_n] ;# get group index for i particle mov edx, [ebp + nb101_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movapd xmm7, [esp + nb101_vctot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# pos 0-1 in xmm7 have the sum now ;# add earlier value from mem mov eax, [ebp + nb101_Vc] addsd xmm7, [eax + edx*8] ;# move back to mem movsd [eax + edx*8], xmm7 ;# finish if last mov ecx, [esp + nb101_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb101_outerend ;# not last, iterate outer loop once more! mov [esp + nb101_n], esi jmp .nb101_outer.nb101_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb101_nri] ;# esi already loaded with n above sub ecx, esi jz .nb101_end ;# non-zero, do one more workunit jmp .nb101_threadloop.nb101_end: emms mov eax, [esp + nb101_nouter] mov ebx, [esp + nb101_ninner] mov ecx, [ebp + nb101_outeriter] mov edx, [ebp + nb101_inneriter] mov [ecx], eax mov [edx], ebx mov eax, [esp + nb101_salign] add esp, eax add esp, 632 pop edi pop esi pop edx pop ecx pop ebx pop eax leave ret.globl nb_kernel101nf_ia32_sse2.globl _nb_kernel101nf_ia32_sse2nb_kernel101nf_ia32_sse2: _nb_kernel101nf_ia32_sse2: .equiv nb101nf_p_nri, 8.equiv nb101nf_iinr, 12.equiv nb101nf_jindex, 16.equiv nb101nf_jjnr, 20.equiv nb101nf_shift, 24.equiv nb101nf_shiftvec, 28.equiv nb101nf_fshift, 32.equiv nb101nf_gid, 36.equiv nb101nf_pos, 40.equiv nb101nf_faction, 44.equiv nb101nf_charge, 48.equiv nb101nf_p_facel, 52.equiv nb101nf_argkrf, 56.equiv nb101nf_argcrf, 60.equiv nb101nf_Vc, 64.equiv nb101nf_type, 68.equiv nb101nf_p_ntype, 72.equiv nb101nf_vdwparam, 76.equiv nb101nf_Vvdw, 80.equiv nb101nf_p_tabscale, 84.equiv nb101nf_VFtab, 88.equiv nb101nf_invsqrta, 92.equiv nb101nf_dvda, 96.equiv nb101nf_p_gbtabscale, 100.equiv nb101nf_GBtab, 104.equiv nb101nf_p_nthreads, 108.equiv nb101nf_count, 112.equiv nb101nf_mtx, 116.equiv nb101nf_outeriter, 120.equiv nb101nf_inneriter, 124.equiv nb101nf_work, 128 ;# stack offsets for local variables ;# bottom of stack is cache-aligned for sse use .equiv nb101nf_ixO, 0.equiv nb101nf_iyO, 16.equiv nb101nf_izO, 32.equiv nb101nf_ixH1, 48.equiv nb101nf_iyH1, 64.equiv nb101nf_izH1, 80.equiv nb101nf_ixH2, 96.equiv nb101nf_iyH2, 112.equiv nb101nf_izH2, 128.equiv nb101nf_iqO, 144.equiv nb101nf_iqH, 160.equiv nb101nf_qqO, 176.equiv nb101nf_qqH, 192.equiv nb101nf_vctot, 208.equiv nb101nf_half, 224.equiv nb101nf_three, 240.equiv nb101nf_is3, 256.equiv nb101nf_ii3, 260.equiv nb101nf_innerjjnr, 264.equiv nb101nf_innerk, 268.equiv nb101nf_n, 272.equiv nb101nf_nn1, 276.equiv nb101nf_nri, 280.equiv nb101nf_nouter, 284.equiv nb101nf_ninner, 288.equiv nb101nf_salign, 292 push ebp mov ebp,esp push eax push ebx push ecx push edx push esi push edi sub esp, 296 ;# local stack space mov eax, esp and eax, 0xf sub esp, 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?