nb_kernel234_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,167 行 · 第 1/5 页
S
2,167 行
mulsd xmm3, xmm2 ;# luA*(3-rsqA*luA*luA) mulsd xmm7, xmm6 ;# luB*(3-rsqB*luB*luB) mulsd xmm3, [esp + nb234_half] ;# iter1 of mulsd xmm7, [esp + nb234_half] ;# iter1 of movapd xmm2, xmm3 ;# copy of luA movapd xmm6, xmm7 ;# copy of luB mulsd xmm3, xmm3 ;# luA*luA mulsd xmm7, xmm7 ;# luB*luB movapd xmm1, [esp + nb234_three] mulsd xmm3, xmm0 ;# rsqA*luA*luA mulsd xmm7, xmm4 ;# rsqB*luB*luB movapd xmm5, xmm1 subsd xmm1, xmm3 ;# 3-rsqA*luA*luA subsd xmm5, xmm7 ;# 3-rsqB*luB*luB mulsd xmm1, xmm2 ;# luA*(3-rsqA*luA*luA) mulsd xmm5, xmm6 ;# luB*(3-rsqB*luB*luB) mulsd xmm1, [esp + nb234_half] ;# rinv mulsd xmm5, [esp + nb234_half] ;# rinv movapd [esp + nb234_rinvOO], xmm1 movapd [esp + nb234_rinvH1H1], xmm5 movapd xmm0, [esp + nb234_rsqH1H2] movapd xmm4, [esp + nb234_rsqH1M] cvtsd2ss xmm1, xmm0 cvtsd2ss xmm5, xmm4 rsqrtss xmm1, xmm1 rsqrtss xmm5, xmm5 cvtss2sd xmm1, xmm1 cvtss2sd xmm5, xmm5 movapd xmm2, xmm1 ;# copy of luA movapd xmm6, xmm5 ;# copy of luB mulsd xmm1, xmm1 ;# luA*luA mulsd xmm5, xmm5 ;# luB*luB movapd xmm3, [esp + nb234_three] mulsd xmm1, xmm0 ;# rsqA*luA*luA mulsd xmm5, xmm4 ;# rsqB*luB*luB movapd xmm7, xmm3 subsd xmm3, xmm1 ;# 3-rsqA*luA*luA subsd xmm7, xmm5 ;# 3-rsqB*luB*luB mulsd xmm3, xmm2 ;# luA*(3-rsqA*luA*luA) mulsd xmm7, xmm6 ;# luB*(3-rsqB*luB*luB) mulsd xmm3, [esp + nb234_half] ;# iter1 mulsd xmm7, [esp + nb234_half] ;# iter1 movapd xmm2, xmm3 ;# copy of luA movapd xmm6, xmm7 ;# copy of luB mulsd xmm3, xmm3 ;# luA*luA mulsd xmm7, xmm7 ;# luB*luB movapd xmm1, [esp + nb234_three] mulsd xmm3, xmm0 ;# rsqA*luA*luA mulsd xmm7, xmm4 ;# rsqB*luB*luB movapd xmm5, xmm1 subsd xmm1, xmm3 ;# 3-rsqA*luA*luA subsd xmm5, xmm7 ;# 3-rsqB*luB*luB mulsd xmm1, xmm2 ;# luA*(3-rsqA*luA*luA) mulsd xmm5, xmm6 ;# luB*(3-rsqB*luB*luB) mulsd xmm1, [esp + nb234_half] ;# rinv mulsd xmm5, [esp + nb234_half] ;# rinv movapd [esp + nb234_rinvH1H2], xmm1 movapd [esp + nb234_rinvH1M], xmm5 movapd xmm0, [esp + nb234_rsqH2H1] movapd xmm4, [esp + nb234_rsqH2H2] cvtsd2ss xmm1, xmm0 cvtsd2ss xmm5, xmm4 rsqrtss xmm1, xmm1 rsqrtss xmm5, xmm5 cvtss2sd xmm1, xmm1 cvtss2sd xmm5, xmm5 movapd xmm2, xmm1 ;# copy of luA movapd xmm6, xmm5 ;# copy of luB mulsd xmm1, xmm1 ;# luA*luA mulsd xmm5, xmm5 ;# luB*luB movapd xmm3, [esp + nb234_three] mulsd xmm1, xmm0 ;# rsqA*luA*luA mulsd xmm5, xmm4 ;# rsqB*luB*luB movapd xmm7, xmm3 subsd xmm3, xmm1 ;# 3-rsqA*luA*luA subsd xmm7, xmm5 ;# 3-rsqB*luB*luB mulsd xmm3, xmm2 ;# luA*(3-rsqA*luA*luA) mulsd xmm7, xmm6 ;# luB*(3-rsqB*luB*luB) mulsd xmm3, [esp + nb234_half] ;# iter1a mulsd xmm7, [esp + nb234_half] ;# iter1b movapd xmm2, xmm3 ;# copy of luA movapd xmm6, xmm7 ;# copy of luB mulsd xmm3, xmm3 ;# luA*luA mulsd xmm7, xmm7 ;# luB*luB movapd xmm1, [esp + nb234_three] mulsd xmm3, xmm0 ;# rsqA*luA*luA mulsd xmm7, xmm4 ;# rsqB*luB*luB movapd xmm5, xmm1 subsd xmm1, xmm3 ;# 3-rsqA*luA*luA subsd xmm5, xmm7 ;# 3-rsqB*luB*luB mulsd xmm1, xmm2 ;# luA*(3-rsqA*luA*luA) mulsd xmm5, xmm6 ;# luB*(3-rsqB*luB*luB) mulsd xmm1, [esp + nb234_half] ;# rinv mulsd xmm5, [esp + nb234_half] ;# rinv movapd [esp + nb234_rinvH2H1], xmm1 movapd [esp + nb234_rinvH2H2], xmm5 movapd xmm0, [esp + nb234_rsqMH1] movapd xmm4, [esp + nb234_rsqH2M] cvtsd2ss xmm1, xmm0 cvtsd2ss xmm5, xmm4 rsqrtss xmm1, xmm1 rsqrtss xmm5, xmm5 cvtss2sd xmm1, xmm1 cvtss2sd xmm5, xmm5 movapd xmm2, xmm1 ;# copy of luA movapd xmm6, xmm5 ;# copy of luB mulsd xmm1, xmm1 ;# luA*luA mulsd xmm5, xmm5 ;# luB*luB movapd xmm3, [esp + nb234_three] mulsd xmm1, xmm0 ;# rsqA*luA*luA mulsd xmm5, xmm4 ;# rsqB*luB*luB movapd xmm7, xmm3 subsd xmm3, xmm1 ;# 3-rsqA*luA*luA subsd xmm7, xmm5 ;# 3-rsqB*luB*luB mulsd xmm3, xmm2 ;# luA*(3-rsqA*luA*luA) mulsd xmm7, xmm6 ;# luB*(3-rsqB*luB*luB) mulsd xmm3, [esp + nb234_half] ;# iter1a mulsd xmm7, [esp + nb234_half] ;# iter1b movapd xmm2, xmm3 ;# copy of luA movapd xmm6, xmm7 ;# copy of luB mulsd xmm3, xmm3 ;# luA*luA mulsd xmm7, xmm7 ;# luB*luB movapd xmm1, [esp + nb234_three] mulsd xmm3, xmm0 ;# rsqA*luA*luA mulsd xmm7, xmm4 ;# rsqB*luB*luB movapd xmm5, xmm1 subsd xmm1, xmm3 ;# 3-rsqA*luA*luA subsd xmm5, xmm7 ;# 3-rsqB*luB*luB mulsd xmm1, xmm2 ;# luA*(3-rsqA*luA*luA) mulsd xmm5, xmm6 ;# luB*(3-rsqB*luB*luB) mulsd xmm1, [esp + nb234_half] ;# rinv mulsd xmm5, [esp + nb234_half] ;# rinv movapd [esp + nb234_rinvMH1], xmm1 movapd [esp + nb234_rinvH2M], xmm5 ;# start with OO interaction movsd xmm0, [esp + nb234_rinvOO] movsd xmm4, [esp + nb234_rsqOO] mulsd xmm4, xmm0 ;# xmm4=r mulsd xmm4, [esp + nb234_tsc] cvttsd2si ebx, xmm4 ;# mm6 = lu idx cvtsi2sd xmm5, ebx subsd xmm4, xmm5 movsd xmm1, xmm4 ;# xmm1=eps movsd xmm2, xmm1 mulsd xmm2, xmm2 ;# xmm2=eps2 shl ebx, 3 mov esi, [ebp + nb234_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 + nb234_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 + nb234_c6] mulsd xmm7, xmm4 ;# fijD mulsd xmm5, xmm4 ;# Vvdw6 ;# put scalar force on stack Update Vvdwtot directly addsd xmm5, [esp + nb234_Vvdwtot] xorpd xmm3, xmm3 mulsd xmm7, [esp + nb234_tsc] subsd xmm3, xmm7 movsd [esp + nb234_fstmp], xmm3 movsd [esp + nb234_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 + nb234_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 + nb234_c12] mulsd xmm7, xmm4 mulsd xmm5, xmm4 addsd xmm5, [esp + nb234_Vvdwtot] movsd xmm3, [esp + nb234_fstmp] mulsd xmm7, [esp + nb234_tsc] subsd xmm3, xmm7 movsd [esp + nb234_Vvdwtot], xmm5 mulsd xmm0, xmm3 movsd xmm1, xmm0 movsd xmm2, xmm0 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulsd xmm0, [esp + nb234_dxOO] mulsd xmm1, [esp + nb234_dyOO] mulsd xmm2, [esp + nb234_dzOO] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 addsd xmm0, [esp + nb234_fixO] addsd xmm1, [esp + nb234_fiyO] addsd xmm2, [esp + nb234_fizO] movsd [esp + nb234_fjxO], xmm3 movsd [esp + nb234_fjyO], xmm4 movsd [esp + nb234_fjzO], xmm5 movsd [esp + nb234_fixO], xmm0 movsd [esp + nb234_fiyO], xmm1 movsd [esp + nb234_fizO], xmm2 ;# H1-H1 interaction movsd xmm0, [esp + nb234_rinvH1H1] movsd xmm7, xmm0 ;# xmm7=rinv movsd xmm5, [esp + nb234_krf] mulsd xmm0, xmm0 ;# xmm0=rinvsq mulsd xmm5, [esp + nb234_rsqH1H1] ;# xmm5=krsq movsd xmm6, xmm5 addsd xmm6, xmm7 ;# xmm6=rinv+ krsq subsd xmm6, [esp + nb234_crf] mulsd xmm6, [esp + nb234_qqHH] ;# xmm6=voul=qq*(rinv+ krsq-crf) mulsd xmm5, [esp + nb234_two] subsd xmm7, xmm5 ;# xmm7=rinv-2*krsq mulsd xmm7, [esp + nb234_qqHH] ;# xmm7 = coul part of fscal addsd xmm6, [esp + nb234_vctot] ;# local vctot summation variable mulsd xmm0, xmm7 movsd [esp + nb234_vctot], xmm6 movapd xmm1, xmm0 movapd xmm2, xmm0 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulsd xmm0, [esp + nb234_dxH1H1] mulsd xmm1, [esp + nb234_dyH1H1] mulsd xmm2, [esp + nb234_dzH1H1] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 addsd xmm0, [esp + nb234_fixH1] addsd xmm1, [esp + nb234_fiyH1] addsd xmm2, [esp + nb234_fizH1] movsd [esp + nb234_fjxH1], xmm3 movsd [esp + nb234_fjyH1], xmm4 movsd [esp + nb234_fjzH1], xmm5 movsd [esp + nb234_fixH1], xmm0 movsd [esp + nb234_fiyH1], xmm1 movsd [esp + nb234_fizH1], xmm2 ;# H1-H2 interaction movsd xmm0, [esp + nb234_rinvH1H2] movsd xmm7, xmm0 ;# xmm7=rinv movsd xmm5, [esp + nb234_krf] mulsd xmm0, xmm0 ;# xmm0=rinvsq mulsd xmm5, [esp + nb234_rsqH1H2] ;# xmm5=krsq movsd xmm6, xmm5 addsd xmm6, xmm7 ;# xmm6=rinv+ krsq subsd xmm6, [esp + nb234_crf] mulsd xmm6, [esp + nb234_qqHH] ;# xmm6=voul=qq*(rinv+ krsq-crf) mulsd xmm5, [esp + nb234_two] subsd xmm7, xmm5 ;# xmm7=rinv-2*krsq mulsd xmm7, [esp + nb234_qqHH] ;# xmm7 = coul part of fscal addsd xmm6, [esp + nb234_vctot] ;# local vctot summation variable mulsd xmm0, xmm7 movsd [esp + nb234_vctot], xmm6 movapd xmm1, xmm0 movapd xmm2, xmm0 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulsd xmm0, [esp + nb234_dxH1H2] mulsd xmm1, [esp + nb234_dyH1H2] mulsd xmm2, [esp + nb234_dzH1H2] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 addsd xmm0, [esp + nb234_fixH1] addsd xmm1, [esp + nb234_fiyH1] addsd xmm2, [esp + nb234_fizH1] movsd [esp + nb234_fjxH2], xmm3 movsd [esp + nb234_fjyH2], xmm4 movsd [esp + nb234_fjzH2], xmm5 movsd [esp + nb234_fixH1], xmm0 movsd [esp + nb234_fiyH1], xmm1 movsd [esp + nb234_fizH1], xmm2 ;# H1-M interaction movsd xmm0, [esp + nb234_rinvH1M] movsd xmm7, xmm0 ;# xmm7=rinv movsd xmm5, [esp + nb234_krf] mulsd xmm0, xmm0 ;# xmm0=rinvsq mulsd xmm5, [esp + nb234_rsqH1M] ;# xmm5=krsq movsd xmm6, xmm5 addsd xmm6, xmm7 ;# xmm6=rinv+ krsq subsd xmm6, [esp + nb234_crf] mulsd xmm6, [esp + nb234_qqMH] ;# xmm6=voul=qq*(rinv+ krsq-crf) mulsd xmm5, [esp + nb234_two] subsd xmm7, xmm5 ;# xmm7=rinv-2*krsq mulsd xmm7, [esp + nb234_qqMH] ;# xmm7 = coul part of fscal addsd xmm6, [esp + nb234_vctot] ;# local vctot summation variable mulsd xmm0, xmm7 movsd [esp + nb234_vctot], xmm6 movapd xmm1, xmm0 movapd xmm2, xmm0 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulsd xmm0, [esp + nb234_dxH1M] mulsd xmm1, [esp + nb234_dyH1M] mulsd xmm2, [esp + nb234_dzH1M] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 addsd xmm0, [esp + nb234_fixH1] addsd xmm1, [esp + nb234_fiyH1] addsd xmm2, [esp + nb234_fizH1] movsd [esp + nb234_fjxM], xmm3 movsd [esp + nb234_fjyM], xmm4 movsd [esp + nb234_fjzM], xmm5 movsd [esp + nb234_fixH1], xmm0 movsd [esp + nb234_fiyH1], xmm1 movsd [esp + nb234_fizH1], xmm2 ;# H2-H1 interaction movsd xmm0, [esp + nb234_rinvH2H1] movsd xmm7, xmm0 ;# xmm7=rinv movsd xmm5, [esp + nb234_krf] mulsd xmm0, xmm0 ;# xmm0=rinvsq mulsd xmm5, [esp + nb234_rsqH2H1] ;# xmm5=krsq movsd xmm6, xmm5 addsd xmm6, xmm7 ;# xmm6=rinv+ krsq subsd xmm6, [esp + nb234_crf] mulsd xmm6, [esp + nb234_qqHH] ;# xmm6=voul=qq*(rinv+ krsq-crf) mulsd xmm5, [esp + nb234_two] subsd xmm7, xmm5 ;# xmm7=rinv-2*krsq mulsd xmm7, [esp + nb234_qqHH] ;# xmm7 = coul part of fscal addsd xmm6, [esp + nb234_vctot] ;# local vctot summation variable mulsd xmm0, xmm7 movsd [esp + nb234_vctot], xmm6 movapd xmm1, xmm0 movapd xmm2, xmm0 movapd xmm3, [esp + nb234_fjxH1] movapd xmm4, [esp + nb234_fjyH1] movapd xmm5, [esp + nb234_fjzH1] mulsd xmm0, [esp + nb234_dxH2H1] mulsd xmm1, [esp + nb234_dyH2H1] mulsd xmm2, [esp + nb234_dzH2H1] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 addsd xmm0, [esp + nb234_fixH2] addsd xmm1, [esp + nb234_fiyH2] addsd xmm2, [esp + nb234_fizH2] movsd [esp + nb234_fjxH1], xmm3 movsd [esp + nb234_fjyH1], xmm4 movsd [esp + nb234_fjzH1], xmm5 movsd [esp + nb234_fixH2], xmm0 movsd [esp + nb234_fiyH2], xmm1 movsd [esp + nb234_fizH2], xmm2 ;# H2-H2 interaction movsd xmm0, [esp + nb234_rinvH2H2] movsd xmm7, xmm0 ;# xmm7=rinv movsd xmm5, [esp + nb234_krf] mulsd xmm0, xmm0 ;# xmm0=rinvsq mulsd xmm5, [esp + nb234_rsqH2H2] ;# xmm5=krsq movsd xmm6, xmm5 addsd xmm6, xmm7 ;# xmm6=rinv+ krsq subsd xmm6⌨️ 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?