nb_kernel231_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,366 行 · 第 1/5 页
S
2,366 行
mulsd xmm4, xmm4 ;# lu*lu mulsd xmm7, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb231_three] subsd xmm4, xmm7 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb231_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 + nb231_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 + nb231_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm6, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb231_three] subsd xmm4, xmm6 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb231_half] ;# rinv movapd xmm6, xmm4 ;# rinvH1 in xmm6 movsd [esp + nb231_rinvH1], xmm6 ;# rsqH2 - seed in xmm2 cvtsd2ss xmm2, xmm5 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [esp + nb231_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 + nb231_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm5, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb231_three] subsd xmm4, xmm5 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb231_half] ;# rinv movapd xmm5, xmm4 ;# rinvH2 in xmm5 movsd [esp + nb231_rinvH2], xmm5 ;# do O interactions movsd xmm0, xmm7 movsd xmm6, xmm7 movsd xmm1, [esp + nb231_krsqO] addsd xmm0, xmm1 mulsd xmm1, [esp + nb231_two] subsd xmm0, [esp + nb231_crf] ;# xmm0=rinv+ krsq-crf subsd xmm6, xmm1 mulsd xmm0, [esp + nb231_qqO] mulsd xmm6, [esp + nb231_qqO] mulsd xmm6, xmm7 movsd [esp + nb231_fstmp], xmm6 ;# save to temp. storage addsd xmm0, [esp + nb231_vctot] movsd [esp + nb231_vctot], xmm0 movsd xmm0, xmm7 movsd xmm4, [esp + nb231_rsqO] ;# LJ table interaction. xmm0=rinv, xmm4=rsq mulsd xmm4, xmm0 ;# xmm4=r mulsd xmm4, [esp + nb231_tsc] cvttsd2si ebx, xmm4 ;# mm6 = lu idx cvtsi2sd xmm5, ebx subpd xmm4, xmm5 movapd xmm1, xmm4 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 shl ebx, 3 mov esi, [ebp + nb231_VFtab] ;# dispersion movlpd xmm4, [esi + ebx*8] ;# Y1 movhpd xmm4, [esi + ebx*8 + 8] ;# Y1 F1 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + ebx*8 + 16] ;# G1 movhpd xmm6, [esi + ebx*8 + 24] ;# G1 H1 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 G2 unpckhpd xmm7, xmm3 ;# H1 H2 ;# dispersion table ready, in xmm4-xmm7 mulsd xmm6, xmm1 ;# xmm6=Geps mulsd xmm7, xmm2 ;# xmm7=Heps2 addsd xmm5, xmm6 addsd xmm5, xmm7 ;# xmm5=Fp mulsd xmm7, [esp + nb231_two] ;# two*Heps2 addsd xmm7, xmm6 addsd xmm7, xmm5 ;# xmm7=FF mulsd xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV movsd xmm4, [esp + nb231_c6] mulsd xmm7, xmm4 ;# fijD mulsd xmm5, xmm4 ;# Vvdw6 ;# put scalar force on stack Update Vvdwtot directly addsd xmm5, [esp + nb231_Vvdwtot] movsd xmm3, [esp + nb231_fstmp] mulsd xmm7, [esp + nb231_tsc] subsd xmm3, xmm7 movsd [esp + nb231_fstmp], xmm3 movsd [esp + nb231_Vvdwtot], xmm5 ;# repulsion movlpd xmm4, [esi + ebx*8 + 32] ;# Y1 movhpd xmm4, [esi + ebx*8 + 40] ;# Y1 F1 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + ebx*8 + 48] ;# G1 movhpd xmm6, [esi + ebx*8 + 56] ;# G1 H1 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 G2 unpckhpd xmm7, xmm3 ;# H1 H2 ;# table ready, in xmm4-xmm7 mulsd xmm6, xmm1 ;# xmm6=Geps mulsd xmm7, xmm2 ;# xmm7=Heps2 addsd xmm5, xmm6 addsd xmm5, xmm7 ;# xmm5=Fp mulsd xmm7, [esp + nb231_two] ;# two*Heps2 addsd xmm7, xmm6 addsd xmm7, xmm5 ;# xmm7=FF mulsd xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV movsd xmm4, [esp + nb231_c12] mulsd xmm7, xmm4 mulsd xmm5, xmm4 addsd xmm5, [esp + nb231_Vvdwtot] movsd xmm3, [esp + nb231_fstmp] mulsd xmm7, [esp + nb231_tsc] subsd xmm3, xmm7 movsd [esp + nb231_Vvdwtot], xmm5 mulsd xmm3, xmm0 movsd xmm0, [esp + nb231_dxO] movsd xmm1, [esp + nb231_dyO] movsd xmm2, [esp + nb231_dzO] mov edi, [ebp + nb231_faction] mulsd xmm0, xmm3 mulsd xmm1, xmm3 mulsd xmm2, xmm3 ;# update O forces movapd xmm3, [esp + nb231_fixO] movapd xmm4, [esp + nb231_fiyO] movapd xmm7, [esp + nb231_fizO] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb231_fixO], xmm3 movlpd [esp + nb231_fiyO], xmm4 movlpd [esp + nb231_fizO], xmm7 ;# update j forces with water O movlpd [esp + nb231_fjx], xmm0 movlpd [esp + nb231_fjy], xmm1 movlpd [esp + nb231_fjz], xmm2 ;# H1 interactions movsd xmm6, [esp + nb231_rinvH1] movsd xmm4, xmm6 mulsd xmm4, xmm4 ;# xmm6=rinv, xmm4=rinvsq movsd xmm7, xmm6 movsd xmm0, [esp + nb231_krsqH1] addsd xmm6, xmm0 ;# xmm6=rinv+ krsq mulsd xmm0, [esp + nb231_two] subsd xmm6, [esp + nb231_crf] subsd xmm7, xmm0 ;# xmm7=rinv-2*krsq mulsd xmm6, [esp + nb231_qqH] ;# vcoul mulsd xmm7, [esp + nb231_qqH] mulsd xmm4, xmm7 ;# total fsH1 in xmm4 addsd xmm6, [esp + nb231_vctot] movapd xmm0, [esp + nb231_dxH1] movapd xmm1, [esp + nb231_dyH1] movapd xmm2, [esp + nb231_dzH1] movlpd [esp + nb231_vctot], xmm6 mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 ;# update H1 forces movapd xmm3, [esp + nb231_fixH1] movapd xmm4, [esp + nb231_fiyH1] movapd xmm7, [esp + nb231_fizH1] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb231_fixH1], xmm3 movlpd [esp + nb231_fiyH1], xmm4 movlpd [esp + nb231_fizH1], xmm7 ;# update j forces with water H1 addsd xmm0, [esp + nb231_fjx] addsd xmm1, [esp + nb231_fjy] addsd xmm2, [esp + nb231_fjz] movlpd [esp + nb231_fjx], xmm0 movlpd [esp + nb231_fjy], xmm1 movlpd [esp + nb231_fjz], xmm2 ;# H2 interactions movapd xmm5, [esp + nb231_rinvH2] movapd xmm4, xmm5 mulpd xmm4, xmm4 ;# xmm5=rinv, xmm4=rinvsq movapd xmm7, xmm5 movsd xmm0, [esp + nb231_krsqH2] addsd xmm5, xmm0 ;# xmm5=rinv+ krsq mulsd xmm0, [esp + nb231_two] subsd xmm5, [esp + nb231_crf] subsd xmm7, xmm0 ;# xmm7=rinv-2*krsq mulsd xmm5, [esp + nb231_qqH] ;# vcoul mulsd xmm7, [esp + nb231_qqH] mulsd xmm4, xmm7 ;# total fsH2 in xmm4 addsd xmm5, [esp + nb231_vctot] movapd xmm0, [esp + nb231_dxH2] movapd xmm1, [esp + nb231_dyH2] movapd xmm2, [esp + nb231_dzH2] movlpd [esp + nb231_vctot], xmm5 mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 ;# update H2 forces movapd xmm3, [esp + nb231_fixH2] movapd xmm4, [esp + nb231_fiyH2] movapd xmm7, [esp + nb231_fizH2] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb231_fixH2], xmm3 movlpd [esp + nb231_fiyH2], xmm4 movlpd [esp + nb231_fizH2], xmm7 mov edi, [ebp + nb231_faction] ;# update j forces addsd xmm0, [esp + nb231_fjx] addsd xmm1, [esp + nb231_fjy] addsd xmm2, [esp + nb231_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 .nb231_updateouterdata: mov ecx, [esp + nb231_ii3] mov edi, [ebp + nb231_faction] mov esi, [ebp + nb231_fshift] mov edx, [esp + nb231_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movapd xmm0, [esp + nb231_fixO] movapd xmm1, [esp + nb231_fiyO] movapd xmm2, [esp + nb231_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 movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, 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 + nb231_fixH1] movapd xmm1, [esp + nb231_fiyH1] movapd xmm2, [esp + nb231_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 + nb231_fixH2] movapd xmm1, [esp + nb231_fiyH2] movapd xmm2, [esp + nb231_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 + nb231_n] ;# get group index for i particle mov edx, [ebp + nb231_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movapd xmm7, [esp + nb231_vctot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# low xmm7 has the sum now ;# add earlier value from mem mov eax, [ebp + nb231_Vc] addsd xmm7, [eax + edx*8] ;# move back to mem movsd [eax + edx*8], xmm7 ;# accumulate total lj energy and update it movapd xmm7, [esp + nb231_Vvdwtot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# low xmm7 has the sum now ;# add earlier value from mem mov eax, [ebp + nb231_Vvdw] addsd xmm7, [eax + edx*8] ;# move back to mem movsd [eax + edx*8], xmm7 ;# finish if last mov ecx, [esp + nb231_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb231_outerend ;# not last, iterate outer loop once more! mov [esp + nb231_n], esi jmp .nb231_outer.nb231_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb231_nri] ;# esi already loaded with n above sub ecx, esi jz .nb231_end ;# non-zero, do one more workunit jmp .nb231_threadloop.nb231_end: emms mov eax, [esp + nb231_nouter] mov ebx, [esp + nb231_ninner] mov ecx, [ebp + nb231_outeriter] mov edx, [ebp + nb231_inneriter] mov [ecx], eax mov [edx], ebx mov eax, [esp + nb231_salign] add esp, eax add esp, 860 pop edi pop esi pop edx pop ecx pop ebx pop eax leave ret.globl nb_kernel231nf_ia32_sse2.globl _nb_kernel231nf_ia32_sse2nb_kernel231nf_ia32_sse2: _nb_kernel231nf_ia32_sse2: .equiv nb231nf_p_nri, 8.equiv nb231nf_iinr, 12.equiv nb231nf_jindex, 16.equiv nb231nf_jjnr, 20.equiv nb231nf_shift, 24.equiv nb231nf_shiftvec, 28.equiv nb231nf_fshift, 32.equiv nb231nf_gid, 36.equiv nb231nf_pos, 40.equiv nb231nf_faction, 44⌨️ 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?