nb_kernel134_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,244 行 · 第 1/5 页
S
2,244 行
movhpd xmm6, [esi + eax*8 + 24] ;# G1 H1 movhpd xmm3, [esi + ebx*8 + 24] ;# G2 H2 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 G2 unpckhpd xmm7, xmm3 ;# H1 H2 ;# dispersion table ready, in xmm4-xmm7 mulpd xmm6, xmm1 ;# xmm6=Geps mulpd xmm7, xmm2 ;# xmm7=Heps2 addpd xmm5, xmm6 addpd xmm5, xmm7 ;# xmm5=Fp mulpd xmm7, [esp + nb134_two] ;# two*Heps2 addpd xmm7, xmm6 addpd xmm7, xmm5 ;# xmm7=FF mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV movapd xmm4, [esp + nb134_c6] mulpd xmm7, xmm4 ;# fijD mulpd xmm5, xmm4 ;# Vvdw6 ;# put scalar force on stack Update Vvdwtot directly addpd xmm5, [esp + nb134_Vvdwtot] xorpd xmm3, xmm3 mulpd xmm7, [esp + nb134_tsc] subpd xmm3, xmm7 movapd [esp + nb134_fstmp], xmm3 movapd [esp + nb134_Vvdwtot], xmm5 ;# repulsion movlpd xmm4, [esi + eax*8 + 32] ;# Y1 movlpd xmm3, [esi + ebx*8 + 32] ;# Y2 movhpd xmm4, [esi + eax*8 + 40] ;# Y1 F1 movhpd xmm3, [esi + ebx*8 + 40] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + eax*8 + 48] ;# G1 movlpd xmm3, [esi + ebx*8 + 48] ;# G2 movhpd xmm6, [esi + eax*8 + 56] ;# G1 H1 movhpd xmm3, [esi + ebx*8 + 56] ;# G2 H2 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 G2 unpckhpd xmm7, xmm3 ;# H1 H2 ;# table ready, in xmm4-xmm7 mulpd xmm6, xmm1 ;# xmm6=Geps mulpd xmm7, xmm2 ;# xmm7=Heps2 addpd xmm5, xmm6 addpd xmm5, xmm7 ;# xmm5=Fp mulpd xmm7, [esp + nb134_two] ;# two*Heps2 addpd xmm7, xmm6 addpd xmm7, xmm5 ;# xmm7=FF mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV movapd xmm4, [esp + nb134_c12] mulpd xmm7, xmm4 mulpd xmm5, xmm4 addpd xmm5, [esp + nb134_Vvdwtot] movapd xmm3, [esp + nb134_fstmp] mulpd xmm7, [esp + nb134_tsc] subpd xmm3, xmm7 movapd [esp + nb134_Vvdwtot], xmm5 mulpd xmm0, xmm3 movapd xmm1, xmm0 movapd xmm2, xmm0 movd eax, mm0 movd ebx, mm1 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulpd xmm0, [esp + nb134_dxOO] mulpd xmm1, [esp + nb134_dyOO] mulpd xmm2, [esp + nb134_dzOO] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb134_fixO] addpd xmm1, [esp + nb134_fiyO] addpd xmm2, [esp + nb134_fizO] movapd [esp + nb134_fjxO], xmm3 movapd [esp + nb134_fjyO], xmm4 movapd [esp + nb134_fjzO], xmm5 movapd [esp + nb134_fixO], xmm0 movapd [esp + nb134_fiyO], xmm1 movapd [esp + nb134_fizO], xmm2 ;# H1-H1 interaction movapd xmm0, [esp + nb134_rinvH1H1] movapd xmm6, xmm0 ;# xmm7=rinv mulpd xmm0, xmm0 ;# xmm0=rinvsq mulpd xmm6, [esp + nb134_qqHH] ;# vcoul mulpd xmm0, xmm6 addpd xmm6, [esp + nb134_vctot] ;# local vctot summation variable movapd xmm1, xmm0 movapd xmm2, xmm0 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulpd xmm0, [esp + nb134_dxH1H1] mulpd xmm1, [esp + nb134_dyH1H1] mulpd xmm2, [esp + nb134_dzH1H1] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb134_fixH1] addpd xmm1, [esp + nb134_fiyH1] addpd xmm2, [esp + nb134_fizH1] movapd [esp + nb134_fjxH1], xmm3 movapd [esp + nb134_fjyH1], xmm4 movapd [esp + nb134_fjzH1], xmm5 movapd [esp + nb134_fixH1], xmm0 movapd [esp + nb134_fiyH1], xmm1 movapd [esp + nb134_fizH1], xmm2 ;# H1-H2 interaction movapd xmm0, [esp + nb134_rinvH1H2] movapd xmm7, xmm0 ;# xmm7=rinv mulpd xmm0, xmm0 ;# xmm0=rinvsq mulpd xmm7, [esp + nb134_qqHH] ;# vcoul mulpd xmm0, xmm7 addpd xmm6, xmm7 movapd xmm1, xmm0 movapd xmm2, xmm0 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulpd xmm0, [esp + nb134_dxH1H2] mulpd xmm1, [esp + nb134_dyH1H2] mulpd xmm2, [esp + nb134_dzH1H2] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb134_fixH1] addpd xmm1, [esp + nb134_fiyH1] addpd xmm2, [esp + nb134_fizH1] movapd [esp + nb134_fjxH2], xmm3 movapd [esp + nb134_fjyH2], xmm4 movapd [esp + nb134_fjzH2], xmm5 movapd [esp + nb134_fixH1], xmm0 movapd [esp + nb134_fiyH1], xmm1 movapd [esp + nb134_fizH1], xmm2 ;# H1-M interaction movapd xmm0, [esp + nb134_rinvH1M] movapd xmm7, xmm0 ;# xmm7=rinv mulpd xmm0, xmm0 ;# xmm0=rinvsq mulpd xmm7, [esp + nb134_qqMH] ;# vcoul mulpd xmm0, xmm7 addpd xmm6, xmm7 movapd xmm1, xmm0 movapd xmm2, xmm0 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulpd xmm0, [esp + nb134_dxH1M] mulpd xmm1, [esp + nb134_dyH1M] mulpd xmm2, [esp + nb134_dzH1M] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb134_fixH1] addpd xmm1, [esp + nb134_fiyH1] addpd xmm2, [esp + nb134_fizH1] movapd [esp + nb134_fjxM], xmm3 movapd [esp + nb134_fjyM], xmm4 movapd [esp + nb134_fjzM], xmm5 movapd [esp + nb134_fixH1], xmm0 movapd [esp + nb134_fiyH1], xmm1 movapd [esp + nb134_fizH1], xmm2 ;# H2-H1 interaction movapd xmm0, [esp + nb134_rinvH2H1] movapd xmm7, xmm0 ;# xmm7=rinv mulpd xmm0, xmm0 ;# xmm0=rinvsq mulpd xmm7, [esp + nb134_qqHH] ;# vcoul mulpd xmm0, xmm7 addpd xmm6, xmm7 movapd xmm1, xmm0 movapd xmm2, xmm0 movapd xmm3, [esp + nb134_fjxH1] movapd xmm4, [esp + nb134_fjyH1] movapd xmm5, [esp + nb134_fjzH1] mulpd xmm0, [esp + nb134_dxH2H1] mulpd xmm1, [esp + nb134_dyH2H1] mulpd xmm2, [esp + nb134_dzH2H1] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb134_fixH2] addpd xmm1, [esp + nb134_fiyH2] addpd xmm2, [esp + nb134_fizH2] movapd [esp + nb134_fjxH1], xmm3 movapd [esp + nb134_fjyH1], xmm4 movapd [esp + nb134_fjzH1], xmm5 movapd [esp + nb134_fixH2], xmm0 movapd [esp + nb134_fiyH2], xmm1 movapd [esp + nb134_fizH2], xmm2 ;# H2-H2 interaction movapd xmm0, [esp + nb134_rinvH2H2] movapd xmm7, xmm0 ;# xmm7=rinv mulpd xmm0, xmm0 ;# xmm0=rinvsq mulpd xmm7, [esp + nb134_qqHH] ;# vcoul mulpd xmm0, xmm7 addpd xmm6, xmm7 movapd xmm1, xmm0 movapd xmm2, xmm0 movapd xmm3, [esp + nb134_fjxH2] movapd xmm4, [esp + nb134_fjyH2] movapd xmm5, [esp + nb134_fjzH2] mulpd xmm0, [esp + nb134_dxH2H2] mulpd xmm1, [esp + nb134_dyH2H2] mulpd xmm2, [esp + nb134_dzH2H2] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb134_fixH2] addpd xmm1, [esp + nb134_fiyH2] addpd xmm2, [esp + nb134_fizH2] movapd [esp + nb134_fjxH2], xmm3 movapd [esp + nb134_fjyH2], xmm4 movapd [esp + nb134_fjzH2], xmm5 movapd [esp + nb134_fixH2], xmm0 movapd [esp + nb134_fiyH2], xmm1 movapd [esp + nb134_fizH2], xmm2 ;# H2-M interaction movapd xmm0, [esp + nb134_rinvH2M] movapd xmm7, xmm0 ;# xmm7=rinv mulpd xmm0, xmm0 ;# xmm0=rinvsq mulpd xmm7, [esp + nb134_qqMH] ;# vcoul mulpd xmm0, xmm7 addpd xmm6, xmm7 movapd xmm1, xmm0 movapd xmm2, xmm0 movapd xmm3, [esp + nb134_fjxM] movapd xmm4, [esp + nb134_fjyM] movapd xmm5, [esp + nb134_fjzM] mulpd xmm0, [esp + nb134_dxH2M] mulpd xmm1, [esp + nb134_dyH2M] mulpd xmm2, [esp + nb134_dzH2M] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb134_fixH2] addpd xmm1, [esp + nb134_fiyH2] addpd xmm2, [esp + nb134_fizH2] movapd [esp + nb134_fjxM], xmm3 movapd [esp + nb134_fjyM], xmm4 movapd [esp + nb134_fjzM], xmm5 movapd [esp + nb134_fixH2], xmm0 movapd [esp + nb134_fiyH2], xmm1 movapd [esp + nb134_fizH2], xmm2 ;# M-H1 interaction movapd xmm0, [esp + nb134_rinvMH1] movapd xmm7, xmm0 ;# xmm7=rinv mulpd xmm0, xmm0 ;# xmm0=rinvsq mulpd xmm7, [esp + nb134_qqMH] ;# vcoul mulpd xmm0, xmm7 addpd xmm6, xmm7 movapd xmm1, xmm0 movapd xmm2, xmm0 movapd xmm3, [esp + nb134_fjxH1] movapd xmm4, [esp + nb134_fjyH1] movapd xmm5, [esp + nb134_fjzH1] mulpd xmm0, [esp + nb134_dxMH1] mulpd xmm1, [esp + nb134_dyMH1] mulpd xmm2, [esp + nb134_dzMH1] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb134_fixM] addpd xmm1, [esp + nb134_fiyM] addpd xmm2, [esp + nb134_fizM] movapd [esp + nb134_fjxH1], xmm3 movapd [esp + nb134_fjyH1], xmm4 movapd [esp + nb134_fjzH1], xmm5 movapd [esp + nb134_fixM], xmm0 movapd [esp + nb134_fiyM], xmm1 movapd [esp + nb134_fizM], xmm2 ;# M-H2 interaction movapd xmm0, [esp + nb134_rinvMH2] movapd xmm7, xmm0 ;# xmm7=rinv mulpd xmm0, xmm0 ;# xmm0=rinvsq mulpd xmm7, [esp + nb134_qqMH] ;# vcoul mulpd xmm0, xmm7 addpd xmm6, xmm7 movapd xmm1, xmm0 movapd xmm2, xmm0 movapd xmm3, [esp + nb134_fjxH2] movapd xmm4, [esp + nb134_fjyH2] movapd xmm5, [esp + nb134_fjzH2] mulpd xmm0, [esp + nb134_dxMH2] mulpd xmm1, [esp + nb134_dyMH2] mulpd xmm2, [esp + nb134_dzMH2] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb134_fixM] addpd xmm1, [esp + nb134_fiyM] addpd xmm2, [esp + nb134_fizM] movapd [esp + nb134_fjxH2], xmm3 movapd [esp + nb134_fjyH2], xmm4 movapd [esp + nb134_fjzH2], xmm5 movapd [esp + nb134_fixM], xmm0 movapd [esp + nb134_fiyM], xmm1 movapd [esp + nb134_fizM], xmm2 ;# M-M interaction movapd xmm0, [esp + nb134_rinvMM] movapd xmm7, xmm0 ;# xmm7=rinv mulpd xmm0, xmm0 ;# xmm0=rinvsq mulpd xmm7, [esp + nb134_qqMM] ;# vcoul mulpd xmm0, xmm7 addpd xmm6, xmm7 movapd [esp + nb134_vctot], xmm6 movapd xmm1, xmm0 movapd xmm2, xmm0 movapd xmm3, [esp + nb134_fjxM] movapd xmm4, [esp + nb134_fjyM] movapd xmm5, [esp + nb134_fjzM] mulpd xmm0, [esp + nb134_dxMM] mulpd xmm1, [esp + nb134_dyMM] mulpd xmm2, [esp + nb134_dzMM] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb134_fixM] addpd xmm1, [esp + nb134_fiyM] addpd xmm2, [esp + nb134_fizM] movapd [esp + nb134_fjxM], xmm3 movapd [esp + nb134_fjyM], xmm4 movapd [esp + nb134_fjzM], xmm5 movapd [esp + nb134_fixM], xmm0 movapd [esp + nb134_fiyM], xmm1 movapd [esp + nb134_fizM], xmm2 mov edi, [ebp + nb134_faction] ;# Did all interactions - now update j forces ;# Step1 - transpose fjxO, fjyO and fjzO, fjxH1 movapd xmm0, [esp + nb134_fjxO] movapd xmm1, [esp + nb134_fjyO] movapd xmm2, [esp + nb134_fjzO] movapd xmm3, [esp + nb134_fjxH1] movapd xmm4, xmm0 movapd xmm5, xmm2 unpcklpd xmm0, xmm1 ;# fjOxA fjOyA unpckhpd xmm4, xmm1 ;# fjOxB fjOyB unpcklpd xmm2, xmm3 ;# fjOzA fjH1xA unpckhpd xmm5, xmm3 ;# fjOzB fjH1xB movlpd xmm1, [edi + eax*8] movlpd xmm3, [edi + ebx*8] movhpd xmm1, [edi + eax*8 + 8] movhpd xmm3, [edi + ebx*8 + 8] movlpd xmm6, [edi + eax*8 + 16] movlpd xmm7, [edi + ebx*8 + 16] movhpd xmm6, [edi + eax*8 + 24] movhpd xmm7, [edi + ebx*8 + 24] addpd xmm1, xmm0 addpd xmm3, xmm4 addpd xmm6, xmm2 addpd xmm7, xmm5 movlpd [edi + eax*8], xmm1 movlpd [edi + ebx*8], xmm3 movhpd [edi + eax*8 + 8], xmm1 movhpd [edi + ebx*8 + 8], xmm3 movlpd [edi + eax*8 + 16], xmm6 movlpd [edi + ebx*8 + 16], xmm7 movhpd [edi + eax*8 + 24], xmm6 movhpd [edi + ebx*8 + 24], xmm7 ;# Step2 - transpose fjyH1, fjzH1 and fjxH2, fjyH2 movapd xmm0, [esp + nb134_fjyH1] movapd xmm1, [esp + nb134_fjzH1] movapd xmm2, [esp + nb134_fjxH2] movapd xmm3, [esp + nb134_fjyH2] movapd xmm4, xmm0 movapd xmm5, xmm2 unpcklpd xmm0, xmm1 ;# fjOxA fjOyA unpckhpd xmm4, xmm1 ;# fjOxB fjOyB unpcklpd xmm2, xmm3 ;# fjOzA fjH1xA unpckhpd xmm5, xmm3 ;# fjOzB fjH1xB movlpd xmm1, [edi + eax*8 + 32] movlpd xmm3, [edi + ebx*8 + 32] movhpd xmm1, [edi + eax*8 + 40] movhpd xmm3, [edi + ebx*8 + 40] movlpd xmm6, [edi + eax*8 + 48] movlpd xmm7, [edi + ebx*8 + 48] movhpd xmm6, [edi + eax*8 + 56] movhpd xmm7, [edi + ebx*8 + 56] addpd xmm1, xmm0 addpd xmm3, xmm4 addpd xmm6, xmm2 addpd xmm7, xmm5 movlpd [edi + eax*8 + 32], xmm1 movlpd [edi + ebx*8 + 32], xmm3 movhpd [edi + eax*8 + 40], xmm1 movhpd [edi + ebx*8 + 40], xmm3 movlpd [edi + eax*8 + 48], xmm6 movlpd [edi + ebx*8 + 48], xmm7 movhpd [edi + eax*8 + 56], xmm6 movhpd [edi + ebx*8 + 56], xmm7 ;# Step3 - transpose fjzH2, fjxM and fjyM, fjzM movapd xmm0, [esp + nb134_fjzH2] movapd xmm1, [esp + nb134_fjxM] movapd xmm2, [esp + nb134_fjyM] movapd xmm3, [esp + nb134_fjzM] movapd xmm4, xmm0 movapd xmm5, xmm2
⌨️ 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?