nb_kernel302_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,217 行 · 第 1/5 页
S
2,217 行
subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb302_fixO] addpd xmm1, [esp + nb302_fiyO] addpd xmm2, [esp + nb302_fizO] movapd [esp + nb302_fjxH1], xmm3 movapd [esp + nb302_fjyH1], xmm4 movapd [esp + nb302_fjzH1], xmm5 movapd [esp + nb302_fixO], xmm0 movapd [esp + nb302_fiyO], xmm1 movapd [esp + nb302_fizO], xmm2 ;# O-H2 interaction movapd xmm0, [esp + nb302_rinvOH2] movapd xmm1, xmm0 mulpd xmm1, [esp + nb302_rsqOH2] ;# xmm1=r mulpd xmm1, [esp + nb302_tsc] cvttpd2pi mm6, xmm1 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm1, xmm6 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb302_VFtab] movd eax, mm6 psrlq mm6, 32 movd ebx, mm6 ;# indices in eax/ebx movlpd xmm4, [esi + eax*8] ;# Y1 movlpd xmm3, [esi + ebx*8] ;# Y2 movhpd xmm4, [esi + eax*8 + 8] ;# Y1 F1 movhpd xmm3, [esi + ebx*8 + 8] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + eax*8 + 16] ;# G1 movlpd xmm3, [esi + ebx*8 + 16] ;# G2 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 ;# coulomb 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 + nb302_two] ;# two*Heps2 movapd xmm3, [esp + nb302_qqOH] addpd xmm7, xmm6 addpd xmm7, xmm5 ;# xmm7=FF mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm3 ;# vcoul=qq*VV mulpd xmm3, xmm7 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm3 fijC addpd xmm5, [esp + nb302_vctot] movapd [esp + nb302_vctot], xmm5 xorpd xmm1, xmm1 mulpd xmm3, [esp + nb302_tsc] mulpd xmm3, xmm0 subpd xmm1, xmm3 movapd xmm0, xmm1 movapd xmm2, xmm1 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulpd xmm0, [esp + nb302_dxOH2] mulpd xmm1, [esp + nb302_dyOH2] mulpd xmm2, [esp + nb302_dzOH2] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb302_fixO] addpd xmm1, [esp + nb302_fiyO] addpd xmm2, [esp + nb302_fizO] movapd [esp + nb302_fjxH2], xmm3 movapd [esp + nb302_fjyH2], xmm4 movapd [esp + nb302_fjzH2], xmm5 movapd [esp + nb302_fixO], xmm0 movapd [esp + nb302_fiyO], xmm1 movapd [esp + nb302_fizO], xmm2 ;# H1-O interaction movapd xmm0, [esp + nb302_rinvH1O] movapd xmm1, xmm0 mulpd xmm1, [esp + nb302_rsqH1O] ;# xmm1=r mulpd xmm1, [esp + nb302_tsc] cvttpd2pi mm6, xmm1 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm1, xmm6 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb302_VFtab] movd eax, mm6 psrlq mm6, 32 movd ebx, mm6 ;# indices in eax/ebx movlpd xmm4, [esi + eax*8] ;# Y1 movlpd xmm3, [esi + ebx*8] ;# Y2 movhpd xmm4, [esi + eax*8 + 8] ;# Y1 F1 movhpd xmm3, [esi + ebx*8 + 8] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + eax*8 + 16] ;# G1 movlpd xmm3, [esi + ebx*8 + 16] ;# G2 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 ;# coulomb 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 + nb302_two] ;# two*Heps2 movapd xmm3, [esp + nb302_qqOH] addpd xmm7, xmm6 addpd xmm7, xmm5 ;# xmm7=FF mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm3 ;# vcoul=qq*VV mulpd xmm3, xmm7 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm3 fijC addpd xmm5, [esp + nb302_vctot] movapd [esp + nb302_vctot], xmm5 xorpd xmm1, xmm1 mulpd xmm3, [esp + nb302_tsc] mulpd xmm3, xmm0 subpd xmm1, xmm3 movapd xmm0, xmm1 movapd xmm2, xmm1 movapd xmm3, [esp + nb302_fjxO] movapd xmm4, [esp + nb302_fjyO] movapd xmm5, [esp + nb302_fjzO] mulpd xmm0, [esp + nb302_dxH1O] mulpd xmm1, [esp + nb302_dyH1O] mulpd xmm2, [esp + nb302_dzH1O] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb302_fixH1] addpd xmm1, [esp + nb302_fiyH1] addpd xmm2, [esp + nb302_fizH1] movapd [esp + nb302_fjxO], xmm3 movapd [esp + nb302_fjyO], xmm4 movapd [esp + nb302_fjzO], xmm5 movapd [esp + nb302_fixH1], xmm0 movapd [esp + nb302_fiyH1], xmm1 movapd [esp + nb302_fizH1], xmm2 ;# H1-H1 interaction movapd xmm0, [esp + nb302_rinvH1H1] movapd xmm1, xmm0 mulpd xmm1, [esp + nb302_rsqH1H1] ;# xmm1=r mulpd xmm1, [esp + nb302_tsc] cvttpd2pi mm6, xmm1 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm1, xmm6 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb302_VFtab] movd eax, mm6 psrlq mm6, 32 movd ebx, mm6 ;# indices in eax/ebx movlpd xmm4, [esi + eax*8] ;# Y1 movlpd xmm3, [esi + ebx*8] ;# Y2 movhpd xmm4, [esi + eax*8 + 8] ;# Y1 F1 movhpd xmm3, [esi + ebx*8 + 8] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + eax*8 + 16] ;# G1 movlpd xmm3, [esi + ebx*8 + 16] ;# G2 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 ;# coulomb 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 + nb302_two] ;# two*Heps2 movapd xmm3, [esp + nb302_qqHH] addpd xmm7, xmm6 addpd xmm7, xmm5 ;# xmm7=FF mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm3 ;# vcoul=qq*VV mulpd xmm3, xmm7 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm3 fijC addpd xmm5, [esp + nb302_vctot] movapd [esp + nb302_vctot], xmm5 xorpd xmm1, xmm1 mulpd xmm3, [esp + nb302_tsc] mulpd xmm3, xmm0 subpd xmm1, xmm3 movapd xmm0, xmm1 movapd xmm2, xmm1 movapd xmm3, [esp + nb302_fjxH1] movapd xmm4, [esp + nb302_fjyH1] movapd xmm5, [esp + nb302_fjzH1] mulpd xmm0, [esp + nb302_dxH1H1] mulpd xmm1, [esp + nb302_dyH1H1] mulpd xmm2, [esp + nb302_dzH1H1] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb302_fixH1] addpd xmm1, [esp + nb302_fiyH1] addpd xmm2, [esp + nb302_fizH1] movapd [esp + nb302_fjxH1], xmm3 movapd [esp + nb302_fjyH1], xmm4 movapd [esp + nb302_fjzH1], xmm5 movapd [esp + nb302_fixH1], xmm0 movapd [esp + nb302_fiyH1], xmm1 movapd [esp + nb302_fizH1], xmm2 ;# H1-H2 interaction movapd xmm0, [esp + nb302_rinvH1H2] movapd xmm1, xmm0 mulpd xmm1, [esp + nb302_rsqH1H2] ;# xmm1=r mulpd xmm1, [esp + nb302_tsc] cvttpd2pi mm6, xmm1 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm1, xmm6 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb302_VFtab] movd eax, mm6 psrlq mm6, 32 movd ebx, mm6 ;# indices in eax/ebx movlpd xmm4, [esi + eax*8] ;# Y1 movlpd xmm3, [esi + ebx*8] ;# Y2 movhpd xmm4, [esi + eax*8 + 8] ;# Y1 F1 movhpd xmm3, [esi + ebx*8 + 8] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + eax*8 + 16] ;# G1 movlpd xmm3, [esi + ebx*8 + 16] ;# G2 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 ;# coulomb 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 + nb302_two] ;# two*Heps2 movapd xmm3, [esp + nb302_qqHH] addpd xmm7, xmm6 addpd xmm7, xmm5 ;# xmm7=FF mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm3 ;# vcoul=qq*VV mulpd xmm3, xmm7 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm3 fijC addpd xmm5, [esp + nb302_vctot] movapd [esp + nb302_vctot], xmm5 xorpd xmm1, xmm1 mulpd xmm3, [esp + nb302_tsc] mulpd xmm3, xmm0 subpd xmm1, xmm3 movapd xmm0, xmm1 movapd xmm2, xmm1 movapd xmm3, [esp + nb302_fjxH2] movapd xmm4, [esp + nb302_fjyH2] movapd xmm5, [esp + nb302_fjzH2] mulpd xmm0, [esp + nb302_dxH1H2] mulpd xmm1, [esp + nb302_dyH1H2] mulpd xmm2, [esp + nb302_dzH1H2] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb302_fixH1] addpd xmm1, [esp + nb302_fiyH1] addpd xmm2, [esp + nb302_fizH1] movapd [esp + nb302_fjxH2], xmm3 movapd [esp + nb302_fjyH2], xmm4 movapd [esp + nb302_fjzH2], xmm5 movapd [esp + nb302_fixH1], xmm0 movapd [esp + nb302_fiyH1], xmm1 movapd [esp + nb302_fizH1], xmm2 ;# H2-O interaction movapd xmm0, [esp + nb302_rinvH2O] movapd xmm1, xmm0 mulpd xmm1, [esp + nb302_rsqH2O] ;# xmm1=r mulpd xmm1, [esp + nb302_tsc] cvttpd2pi mm6, xmm1 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm1, xmm6 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb302_VFtab] movd eax, mm6 psrlq mm6, 32 movd ebx, mm6 ;# indices in eax/ebx movlpd xmm4, [esi + eax*8] ;# Y1 movlpd xmm3, [esi + ebx*8] ;# Y2 movhpd xmm4, [esi + eax*8 + 8] ;# Y1 F1 movhpd xmm3, [esi + ebx*8 + 8] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + eax*8 + 16] ;# G1 movlpd xmm3, [esi + ebx*8 + 16] ;# G2 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 ;# coulomb 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 + nb302_two] ;# two*Heps2 movapd xmm3, [esp + nb302_qqOH] addpd xmm7, xmm6 addpd xmm7, xmm5 ;# xmm7=FF mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm3 ;# vcoul=qq*VV mulpd xmm3, xmm7 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm3 fijC addpd xmm5, [esp + nb302_vctot] movapd [esp + nb302_vctot], xmm5 xorpd xmm1, xmm1 mulpd xmm3, [esp + nb302_tsc] mulpd xmm3, xmm0 subpd xmm1, xmm3 movapd xmm0, xmm1 movapd xmm2, xmm1 movapd xmm3, [esp + nb302_fjxO] movapd xmm4, [esp + nb302_fjyO] movapd xmm5, [esp + nb302_fjzO] mulpd xmm0, [esp + nb302_dxH2O] mulpd xmm1, [esp + nb302_dyH2O] mulpd xmm2, [esp + nb302_dzH2O] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb302_fixH2] addpd xmm1, [esp + nb302_fiyH2] addpd xmm2, [esp + nb302_fizH2] movapd [esp + nb302_fjxO], xmm3 movapd [esp + nb302_fjyO], xmm4 movapd [esp + nb302_fjzO], xmm5 movapd [esp + nb302_fixH2], xmm0 movapd [esp + nb302_fiyH2], xmm1 movapd [esp + nb302_fizH2], xmm2 ;# H2-H1 interaction movapd xmm0, [esp + nb302_rinvH2H1] movapd xmm1, xmm0 mulpd xmm1, [esp + nb302_rsqH2H1] ;# xmm1=r mulpd xmm1, [esp + nb302_tsc] cvttpd2pi mm6, xmm1 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm1, xmm6 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb302_VFtab] movd eax, mm6 psrlq mm6, 32 movd ebx, mm6 ;# indices in eax/ebx movlpd xmm4, [esi + eax*8] ;# Y1 movlpd xmm3, [esi + ebx*8] ;# Y2 movhpd xmm4, [esi + eax*8 + 8] ;# Y1 F1 movhpd xmm3, [esi + ebx*8 + 8] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + eax*8 + 16] ;# G1 movlpd xmm3, [esi + ebx*8 + 16] ;# G2 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 ;# coulomb 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 + nb302_two] ;# two*Heps2 movapd xmm3, [esp + nb302_qqHH] addpd xmm7, xmm6 addpd xmm7, xmm5 ;# xmm7=FF mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm3 ;# vcoul=qq*VV
⌨️ 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?