nb_kernel133_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,398 行 · 第 1/5 页
S
2,398 行
addpd xmm5, xmm4 addpd xmm5, xmm3 ;# move ixM-izM to xmm2-xmm4 movapd xmm3, [esp + nb133_iyM] movapd xmm4, [esp + nb133_izM] subpd xmm3, xmm1 subpd xmm4, xmm2 movapd xmm2, [esp + nb133_ixM] subpd xmm2, xmm0 ;# store dr movapd [esp + nb133_dxM], xmm2 movapd [esp + nb133_dyM], xmm3 movapd [esp + nb133_dzM], xmm4 ;# square it mulpd xmm2,xmm2 mulpd xmm3,xmm3 mulpd xmm4,xmm4 addpd xmm4, xmm3 addpd xmm4, xmm2 ;# rsqM in xmm4, rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 movapd [esp + nb133_rsqO], xmm7 ;# start with rsqH1 - put seed in xmm2 cvtpd2ps xmm2, xmm6 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm1, [esp + nb133_three] mulpd xmm2, xmm6 ;# rsq*lu*lu subpd xmm1, xmm2 ;# 30-rsq*lu*lu mulpd xmm1, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm1, [esp + nb133_half] ;# iter1 ( new lu) movapd xmm3, xmm1 mulpd xmm1, xmm1 ;# lu*lu mulpd xmm6, xmm1 ;# rsq*lu*lu movapd xmm1, [esp + nb133_three] subpd xmm1, xmm6 ;# 3-rsq*lu*lu mulpd xmm1, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm1, [esp + nb133_half] ;# rinv movapd [esp + nb133_rinvH1], xmm1 ;# rsqH2 - seed in xmm2 cvtpd2ps xmm2, xmm5 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm1, [esp + nb133_three] mulpd xmm2, xmm5 ;# rsq*lu*lu subpd xmm1, xmm2 ;# 30-rsq*lu*lu mulpd xmm1, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm1, [esp + nb133_half] ;# iter1 ( new lu) movapd xmm3, xmm1 mulpd xmm1, xmm1 ;# lu*lu mulpd xmm5, xmm1 ;# rsq*lu*lu movapd xmm1, [esp + nb133_three] subpd xmm1, xmm5 ;# 3-rsq*lu*lu mulpd xmm1, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm1, [esp + nb133_half] ;# rinv movapd [esp + nb133_rinvH2], xmm1 ;# rsqM - seed in xmm2 cvtpd2ps xmm2, xmm4 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm1, [esp + nb133_three] mulpd xmm2, xmm4 ;# rsq*lu*lu subpd xmm1, xmm2 ;# 30-rsq*lu*lu mulpd xmm1, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm1, [esp + nb133_half] ;# iter1 ( new lu) movapd xmm3, xmm1 mulpd xmm1, xmm1 ;# lu*lu mulpd xmm4, xmm1 ;# rsq*lu*lu movapd xmm1, [esp + nb133_three] subpd xmm1, xmm4 ;# 3-rsq*lu*lu mulpd xmm1, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm1, [esp + nb133_half] ;# rinv movapd [esp + nb133_rinvM], xmm1 ;# rsqO - put seed in xmm2 cvtpd2ps xmm2, xmm7 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm4, [esp + nb133_three] mulpd xmm2, xmm7 ;# rsq*lu*lu subpd xmm4, xmm2 ;# 30-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm4, [esp + nb133_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulpd xmm4, xmm4 ;# lu*lu mulpd xmm7, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb133_three] subpd xmm4, xmm7 ;# 3-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm4, [esp + nb133_half] ;# rinv movapd xmm7, xmm4 ;# rinvO in xmm7 movapd xmm4, [esp + nb133_rsqO] movapd xmm0, xmm7 ;# LJ table interaction. mulpd xmm4, xmm7 ;# xmm4=r mulpd xmm4, [esp + nb133_tsc] cvttpd2pi mm6, xmm4 ;# mm6 = lu idx cvtpi2pd xmm5, mm6 subpd xmm4, xmm5 movapd xmm1, xmm4 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 3 ;# idx *= 8 movd mm0, eax movd mm1, ebx mov esi, [ebp + nb133_VFtab] movd eax, mm6 psrlq mm6, 32 movd ebx, mm6 ;# dispersion 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 ;# 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 + nb133_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 + nb133_c6] mulpd xmm7, xmm4 ;# fijD mulpd xmm5, xmm4 ;# Vvdw6 ;# put scalar force on stack Update Vvdwtot directly addpd xmm5, [esp + nb133_Vvdwtot] xorpd xmm3, xmm3 mulpd xmm7, [esp + nb133_tsc] subpd xmm3, xmm7 movapd [esp + nb133_fstmp], xmm3 movapd [esp + nb133_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 + nb133_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 + nb133_c12] mulpd xmm7, xmm4 mulpd xmm5, xmm4 addpd xmm5, [esp + nb133_Vvdwtot] movapd xmm3, [esp + nb133_fstmp] mulpd xmm7, [esp + nb133_tsc] subpd xmm3, xmm7 movapd [esp + nb133_Vvdwtot], xmm5 mulpd xmm3, xmm0 movapd xmm0, [esp + nb133_dxO] movapd xmm1, [esp + nb133_dyO] movapd xmm2, [esp + nb133_dzO] movd eax, mm0 movd ebx, mm1 mov edi, [ebp + nb133_faction] mulpd xmm0, xmm3 mulpd xmm1, xmm3 mulpd xmm2, xmm3 ;# update O forces movapd xmm3, [esp + nb133_fixO] movapd xmm4, [esp + nb133_fiyO] movapd xmm7, [esp + nb133_fizO] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb133_fixO], xmm3 movapd [esp + nb133_fiyO], xmm4 movapd [esp + nb133_fizO], xmm7 ;# update j forces with water O movapd [esp + nb133_fjx], xmm0 movapd [esp + nb133_fjy], xmm1 movapd [esp + nb133_fjz], xmm2 ;# H1 interactions movapd xmm6, [esp + nb133_rinvH1] movapd xmm4, xmm6 mulpd xmm4, xmm4 ;# xmm6=rinv, xmm4=rinvsq mulpd xmm6, [esp + nb133_qqH] ;# vcoul mulpd xmm4, xmm6 ;# fscal addpd xmm6, [esp + nb133_vctot] movapd [esp + nb133_vctot], xmm6 movapd xmm0, [esp + nb133_dxH1] movapd xmm1, [esp + nb133_dyH1] movapd xmm2, [esp + nb133_dzH1] mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 ;# update H1 forces movapd xmm3, [esp + nb133_fixH1] movapd xmm4, [esp + nb133_fiyH1] movapd xmm7, [esp + nb133_fizH1] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb133_fixH1], xmm3 movapd [esp + nb133_fiyH1], xmm4 movapd [esp + nb133_fizH1], xmm7 ;# update j forces with water H1 addpd xmm0, [esp + nb133_fjx] addpd xmm1, [esp + nb133_fjy] addpd xmm2, [esp + nb133_fjz] movapd [esp + nb133_fjx], xmm0 movapd [esp + nb133_fjy], xmm1 movapd [esp + nb133_fjz], xmm2 ;# H2 interactions movapd xmm6, [esp + nb133_rinvH2] movapd xmm4, xmm6 mulpd xmm4, xmm4 ;# xmm6=rinv, xmm4=rinvsq mulpd xmm6, [esp + nb133_qqH] ;# vcoul mulpd xmm4, xmm6 ;# fscal addpd xmm6, [esp + nb133_vctot] movapd [esp + nb133_vctot], xmm6 movapd xmm0, [esp + nb133_dxH2] movapd xmm1, [esp + nb133_dyH2] movapd xmm2, [esp + nb133_dzH2] mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 ;# update H2 forces movapd xmm3, [esp + nb133_fixH2] movapd xmm4, [esp + nb133_fiyH2] movapd xmm7, [esp + nb133_fizH2] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb133_fixH2], xmm3 movapd [esp + nb133_fiyH2], xmm4 movapd [esp + nb133_fizH2], xmm7 ;# update j forces with water H2 addpd xmm0, [esp + nb133_fjx] addpd xmm1, [esp + nb133_fjy] addpd xmm2, [esp + nb133_fjz] movapd [esp + nb133_fjx], xmm0 movapd [esp + nb133_fjy], xmm1 movapd [esp + nb133_fjz], xmm2 ;# M interactions movapd xmm6, [esp + nb133_rinvM] movapd xmm4, xmm6 mulpd xmm4, xmm4 ;# xmm6=rinv, xmm4=rinvsq mulpd xmm6, [esp + nb133_qqM] ;# vcoul mulpd xmm4, xmm6 ;# fscal addpd xmm6, [esp + nb133_vctot] movapd [esp + nb133_vctot], xmm6 movapd xmm0, [esp + nb133_dxM] movapd xmm1, [esp + nb133_dyM] movapd xmm2, [esp + nb133_dzM] mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 ;# update H2 forces movapd xmm3, [esp + nb133_fixM] movapd xmm4, [esp + nb133_fiyM] movapd xmm7, [esp + nb133_fizM] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb133_fixM], xmm3 movapd [esp + nb133_fiyM], xmm4 movapd [esp + nb133_fizM], xmm7 mov edi, [ebp + nb133_faction] ;# update j forces addpd xmm0, [esp + nb133_fjx] addpd xmm1, [esp + nb133_fjy] addpd xmm2, [esp + nb133_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 + nb133_innerk], 2 jl .nb133_checksingle jmp .nb133_unroll_loop.nb133_checksingle: mov edx, [esp + nb133_innerk] and edx, 1 jnz .nb133_dosingle jmp .nb133_updateouterdata.nb133_dosingle: mov edx, [esp + nb133_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] add dword ptr [esp + nb133_innerjjnr], 4 mov esi, [ebp + nb133_charge] ;# base of charge[] xorpd xmm3, xmm3 movlpd xmm3, [esi + eax*8] movapd xmm4, xmm3 mulsd xmm3, [esp + nb133_iqM] mulsd xmm4, [esp + nb133_iqH] movd mm0, eax ;# use mmx registers as temp storage movapd [esp + nb133_qqM], xmm3 movapd [esp + nb133_qqH], xmm4 mov esi, [ebp + nb133_type] mov eax, [esi + eax*4] mov esi, [ebp + nb133_vdwparam] shl eax, 1 mov edi, [esp + nb133_ntia] add eax, edi movlpd xmm6, [esi + eax*8] ;# c6a movhpd xmm6, [esi + eax*8 + 8] ;# c6a c12a xorpd xmm7, xmm7 movapd xmm4, xmm6 unpcklpd xmm4, xmm7 unpckhpd xmm6, xmm7 movd eax, mm0 movd ebx, mm1 movapd [esp + nb133_c6], xmm4 movapd [esp + nb133_c12], xmm6 mov esi, [ebp + nb133_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 + nb133_ixO] movapd xmm5, [esp + nb133_iyO] movapd xmm6, [esp + nb133_izO] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# store dr movapd [esp + nb133_dxO], xmm4 movapd [esp + nb133_dyO], xmm5 movapd [esp + nb133_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 movapd [esp + nb133_rsqO], xmm7 ;# move ixH1-izH1 to xmm4-xmm6 movapd xmm4, [esp + nb133_ixH1] movapd xmm5, [esp + nb133_iyH1] movapd xmm6, [esp + nb133_izH1] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# store dr movapd [esp + nb133_dxH1], xmm4 movapd [esp + nb133_dyH1], xmm5 movapd [esp + nb133_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 + nb133_ixH2] movapd xmm4, [esp + nb133_iyH2] movapd xmm5, [esp + nb133_izH2] ;# calc dr subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 ;# store dr movapd [esp + nb133_dxH2], xmm3 movapd [esp + nb133_dyH2], xmm4 movapd [esp + nb133_dzH2], xmm5 ;# square it mulsd xmm3,xmm3 mulsd xmm4,xmm4⌨️ 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?