nb_kernel313_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,380 行 · 第 1/5 页
S
2,380 行
movapd xmm4, [esp + nb313_iyH2] movapd xmm5, [esp + nb313_izH2] ;# calc dr subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 ;# store dr movapd [esp + nb313_dxH2], xmm3 movapd [esp + nb313_dyH2], xmm4 movapd [esp + nb313_dzH2], xmm5 ;# square it mulpd xmm3,xmm3 mulpd xmm4,xmm4 mulpd xmm5,xmm5 addpd xmm5, xmm4 addpd xmm5, xmm3 ;# move ixM-izM to xmm2-xmm4 movapd xmm3, [esp + nb313_iyM] movapd xmm4, [esp + nb313_izM] subpd xmm3, xmm1 subpd xmm4, xmm2 movapd xmm2, [esp + nb313_ixM] subpd xmm2, xmm0 ;# store dr movapd [esp + nb313_dxM], xmm2 movapd [esp + nb313_dyM], xmm3 movapd [esp + nb313_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 ;# 1/x for O - rsqO is in xmm7 cvtpd2ps xmm2, xmm7 movapd xmm3, xmm7 rcpps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm1, [esp + nb313_two] movapd xmm0, xmm1 mulpd xmm7, xmm2 subpd xmm1, xmm7 mulpd xmm2, xmm1 ;# iter1 mulpd xmm3, xmm2 subpd xmm0, xmm3 mulpd xmm0, xmm2 ;# xmm0=rinvsq movapd [esp + nb313_rinvsqO], xmm0 ;# rsqH1 - seed in xmm2 cvtpd2ps xmm2, xmm6 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm0, [esp + nb313_three] mulpd xmm2, xmm6 ;# rsq*lu*lu subpd xmm0, xmm2 ;# 30-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm0, [esp + nb313_half] ;# iter1 ( new lu) movapd xmm2, xmm6 movapd xmm3, xmm0 mulpd xmm0, xmm0 ;# lu*lu mulpd xmm2, xmm0 ;# rsq*lu*lu movapd xmm0, [esp + nb313_three] subpd xmm0, xmm2 ;# 3-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm0, [esp + nb313_half] ;# rinv movapd [esp + nb313_rinvH1], xmm0 ;# rinvH1 mulpd xmm6, xmm0 movapd [esp + nb313_rH1], xmm6 ;# rH1 ;# rsqH2 - seed in xmm2 cvtpd2ps xmm2, xmm5 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm0, [esp + nb313_three] mulpd xmm2, xmm5 ;# rsq*lu*lu subpd xmm0, xmm2 ;# 30-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm0, [esp + nb313_half] ;# iter1 ( new lu) movapd xmm2, xmm5 movapd xmm3, xmm0 mulpd xmm0, xmm0 ;# lu*lu mulpd xmm2, xmm0 ;# rsq*lu*lu movapd xmm0, [esp + nb313_three] subpd xmm0, xmm2 ;# 3-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm0, [esp + nb313_half] ;# rinv movapd [esp + nb313_rinvH2], xmm0 ;# rinv mulpd xmm5, xmm0 movapd [esp + nb313_rH2], xmm5 ;# r ;# rsqM - seed in xmm2 cvtpd2ps xmm2, xmm4 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm0, [esp + nb313_three] mulpd xmm2, xmm4 ;# rsq*lu*lu subpd xmm0, xmm2 ;# 30-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm0, [esp + nb313_half] ;# iter1 ( new lu) movapd xmm2, xmm4 movapd xmm3, xmm0 mulpd xmm0, xmm0 ;# lu*lu mulpd xmm2, xmm0 ;# rsq*lu*lu movapd xmm0, [esp + nb313_three] subpd xmm0, xmm2 ;# 3-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm0, [esp + nb313_half] ;# rinv movapd [esp + nb313_rinvM], xmm0 ;# rinv mulpd xmm4, xmm0 movapd [esp + nb313_rM], xmm4 ;# r ;# do O interactions movapd xmm0, [esp + nb313_rinvsqO] movapd xmm1, xmm0 mulpd xmm1, xmm1 ;# rinv4 mulpd xmm1, xmm0 ;#rinvsix movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# rinvtwelve mulpd xmm1, [esp + nb313_c6] mulpd xmm2, [esp + nb313_c12] movapd xmm3, xmm2 subpd xmm3, xmm1 ;# Vvdw=Vvdw12-Vvdw6 addpd xmm3, [esp + nb313_Vvdwtot] mulpd xmm1, [esp + nb313_six] mulpd xmm2, [esp + nb313_twelve] subpd xmm2, xmm1 mulpd xmm2, xmm0 movapd xmm4, xmm2 ;# total fsO movapd [esp + nb313_Vvdwtot], xmm3 movapd xmm0, [esp + nb313_dxO] movapd xmm1, [esp + nb313_dyO] movapd xmm2, [esp + nb313_dzO] mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 ;# tx in xmm0-xmm2 ;# update O forces movapd xmm3, [esp + nb313_fixO] movapd xmm4, [esp + nb313_fiyO] movapd xmm7, [esp + nb313_fizO] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb313_fixO], xmm3 movapd [esp + nb313_fiyO], xmm4 movapd [esp + nb313_fizO], xmm7 ;# update j forces with water O movapd [esp + nb313_fjx], xmm0 movapd [esp + nb313_fjy], xmm1 movapd [esp + nb313_fjz], xmm2 ;# Done with O interactions - now H1! movapd xmm7, [esp + nb313_rH1] mulpd xmm7, [esp + nb313_tsc] cvttpd2pi mm6, xmm7 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 movd mm0, eax movd mm1, ebx pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb313_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 + nb313_two] ;# two*Heps2 movapd xmm3, [esp + nb313_qqH] 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 ;# increment vcoul xorpd xmm4, xmm4 addpd xmm5, [esp + nb313_vctot] mulpd xmm3, [esp + nb313_rinvH1] movapd [esp + nb313_vctot], xmm5 mulpd xmm3, [esp + nb313_tsc] subpd xmm4, xmm3 movapd xmm0, [esp + nb313_dxH1] movapd xmm1, [esp + nb313_dyH1] movapd xmm2, [esp + nb313_dzH1] mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 ;# update H1 forces movapd xmm3, [esp + nb313_fixH1] movapd xmm4, [esp + nb313_fiyH1] movapd xmm7, [esp + nb313_fizH1] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb313_fixH1], xmm3 movapd [esp + nb313_fiyH1], xmm4 movapd [esp + nb313_fizH1], xmm7 ;# update j forces with water H1 addpd xmm0, [esp + nb313_fjx] addpd xmm1, [esp + nb313_fjy] addpd xmm2, [esp + nb313_fjz] movapd [esp + nb313_fjx], xmm0 movapd [esp + nb313_fjy], xmm1 movapd [esp + nb313_fjz], xmm2 ;# H2 interactions movapd xmm7, [esp + nb313_rH2] mulpd xmm7, [esp + nb313_tsc] cvttpd2pi mm6, xmm7 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb313_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 + nb313_two] ;# two*Heps2 movapd xmm3, [esp + nb313_qqH] 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 ;# increment vcoul xorpd xmm4, xmm4 addpd xmm5, [esp + nb313_vctot] mulpd xmm3, [esp + nb313_rinvH2] movapd [esp + nb313_vctot], xmm5 mulpd xmm3, [esp + nb313_tsc] subpd xmm4, xmm3 movapd xmm0, [esp + nb313_dxH2] movapd xmm1, [esp + nb313_dyH2] movapd xmm2, [esp + nb313_dzH2] mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 movd eax, mm0 movd ebx, mm1 ;# update H2 forces movapd xmm3, [esp + nb313_fixH2] movapd xmm4, [esp + nb313_fiyH2] movapd xmm7, [esp + nb313_fizH2] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb313_fixH2], xmm3 movapd [esp + nb313_fiyH2], xmm4 movapd [esp + nb313_fizH2], xmm7 ;# update j forces with water H1 addpd xmm0, [esp + nb313_fjx] addpd xmm1, [esp + nb313_fjy] addpd xmm2, [esp + nb313_fjz] movapd [esp + nb313_fjx], xmm0 movapd [esp + nb313_fjy], xmm1 movapd [esp + nb313_fjz], xmm2 ;# M interactions movapd xmm7, [esp + nb313_rM] mulpd xmm7, [esp + nb313_tsc] cvttpd2pi mm6, xmm7 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 movd mm0, eax movd mm1, ebx pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb313_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 + nb313_two] ;# two*Heps2 movapd xmm3, [esp + nb313_qqM] 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 ;# increment vcoul xorpd xmm4, xmm4 addpd xmm5, [esp + nb313_vctot] mulpd xmm3, [esp + nb313_rinvM] movapd [esp + nb313_vctot], xmm5 mulpd xmm3, [esp + nb313_tsc] subpd xmm4, xmm3 movapd xmm0, [esp + nb313_dxM] movapd xmm1, [esp + nb313_dyM] movapd xmm2, [esp + nb313_dzM] mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 movd eax, mm0 movd ebx, mm1 ;# update H2 forces movapd xmm3, [esp + nb313_fixM] movapd xmm4, [esp + nb313_fiyM] movapd xmm7, [esp + nb313_fizM] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb313_fixM], xmm3 movapd [esp + nb313_fiyM], xmm4 movapd [esp + nb313_fizM], xmm7 mov edi, [ebp + nb313_faction] ;# update j forces addpd xmm0, [esp + nb313_fjx] addpd xmm1, [esp + nb313_fjy] addpd xmm2, [esp + nb313_fjz] ;# the fj's - start by accumulating forces from memory 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 + nb313_innerk], 2 jl .nb313_checksingle jmp .nb313_unroll_loop.nb313_checksingle: mov edx, [esp + nb313_innerk] and edx, 1 jnz .nb313_dosingle jmp .nb313_updateouterdata.nb313_dosingle: mov edx, [esp + nb313_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] mov esi, [ebp + nb313_charge] ;# base of charge[] xorpd xmm3, xmm3 movlpd xmm3, [esi + eax*8] movapd xmm4, xmm3 mulpd xmm3, [esp + nb313_iqM] mulpd xmm4, [esp + nb313_iqH] movd mm0, eax ;# use mmx registers as temp storage movapd [esp + nb313_qqM], xmm3 movapd [esp + nb313_qqH], xmm4 mov esi, [ebp + nb313_type] mov eax, [esi + eax*4] mov esi, [ebp + nb313_vdwparam] shl eax, 1 mov edi, [esp + nb313_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 movapd [esp + nb313_c6], 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?