nb_kernel311_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,357 行 · 第 1/5 页
S
2,357 行
movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm4, [esp + nb311_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 + nb311_half] ;# iter1 ( new lu) movapd xmm2, xmm7 movapd xmm3, xmm4 mulpd xmm4, xmm4 ;# lu*lu mulpd xmm2, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb311_three] subpd xmm4, xmm2 ;# 3-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm4, [esp + nb311_half] ;# rinv movapd [esp + nb311_rinvO], xmm4 ;# rinvO in xmm4 mulpd xmm7, xmm4 movapd [esp + nb311_rO], xmm7 ;# r in xmm7 ;# rsqH1 - seed in xmm2 cvtpd2ps xmm2, xmm6 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm4, [esp + nb311_three] mulpd xmm2, xmm6 ;# rsq*lu*lu subpd xmm4, xmm2 ;# 30-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm4, [esp + nb311_half] ;# iter1 ( new lu) movapd xmm2, xmm6 movapd xmm3, xmm4 mulpd xmm4, xmm4 ;# lu*lu mulpd xmm2, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb311_three] subpd xmm4, xmm2 ;# 3-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm4, [esp + nb311_half] ;# rinv movapd [esp + nb311_rinvH1], xmm4 ;# rinvH1 mulpd xmm6, xmm4 movapd [esp + nb311_rH1], xmm6 ;# rH1 ;# rsqH2 - seed in xmm2 cvtpd2ps xmm2, xmm5 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm4, [esp + nb311_three] mulpd xmm2, xmm5 ;# rsq*lu*lu subpd xmm4, xmm2 ;# 30-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm4, [esp + nb311_half] ;# iter1 ( new lu) movapd xmm2, xmm5 movapd xmm3, xmm4 mulpd xmm4, xmm4 ;# lu*lu mulpd xmm2, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb311_three] subpd xmm4, xmm2 ;# 3-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm4, [esp + nb311_half] ;# rinv movapd [esp + nb311_rinvH2], xmm4 ;# rinv mulpd xmm5, xmm4 movapd [esp + nb311_rH2], xmm5 ;# r ;# do O interactions ;# rO is still in xmm7 movapd xmm0, [esp + nb311_rinvO] mulpd xmm7, [esp + nb311_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 movd mm0, eax movd mm1, ebx mov esi, [ebp + nb311_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 + nb311_two] ;# two*Heps2 movapd xmm0, [esp + nb311_qqO] addpd xmm7, xmm6 addpd xmm7, xmm5 ;# xmm7=FF mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm0 ;# vcoul=qq*VV mulpd xmm0, xmm7 ;# fijC=FF*qq ;# do nontable L-J movapd xmm2, [esp + nb311_rinvO] mulpd xmm2, xmm2 ;# at this point mm5 contains vcoul and xmm0 fijC ;# increment vcoul - then we can get rid of mm5 addpd xmm5, [esp + nb311_vctot] movapd [esp + nb311_vctot], xmm5 movapd xmm1, xmm2 mulpd xmm1, xmm1 mulpd xmm1, xmm2 ;# xmm1=rinvsix movapd xmm4, xmm1 mulpd xmm4, xmm4 ;# xmm4=rinvtwelve mulpd xmm1, [esp + nb311_c6] mulpd xmm4, [esp + nb311_c12] movapd xmm3, xmm4 subpd xmm3, xmm1 ;# xmm3=Vvdw12-Vvdw6 mulpd xmm1, [esp + nb311_six] mulpd xmm4, [esp + nb311_twelve] subpd xmm4, xmm1 addpd xmm3, [esp + nb311_Vvdwtot] mulpd xmm4, [esp + nb311_rinvO] mulpd xmm0, [esp + nb311_tsc] subpd xmm4, xmm0 movapd [esp + nb311_Vvdwtot], xmm3 mulpd xmm4, [esp + nb311_rinvO] movapd xmm0, [esp + nb311_dxO] movapd xmm1, [esp + nb311_dyO] movapd xmm2, [esp + nb311_dzO] mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 ;# tx in xmm0-xmm2 ;# update O forces movapd xmm3, [esp + nb311_fixO] movapd xmm4, [esp + nb311_fiyO] movapd xmm7, [esp + nb311_fizO] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb311_fixO], xmm3 movapd [esp + nb311_fiyO], xmm4 movapd [esp + nb311_fizO], xmm7 ;# update j forces with water O movapd [esp + nb311_fjx], xmm0 movapd [esp + nb311_fjy], xmm1 movapd [esp + nb311_fjz], xmm2 ;# Done with O interactions - now H1! movapd xmm7, [esp + nb311_rH1] mulpd xmm7, [esp + nb311_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 + nb311_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 + nb311_two] ;# two*Heps2 movapd xmm3, [esp + nb311_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 + nb311_vctot] mulpd xmm3, [esp + nb311_rinvH1] movapd [esp + nb311_vctot], xmm5 mulpd xmm3, [esp + nb311_tsc] subpd xmm4, xmm3 movapd xmm0, [esp + nb311_dxH1] movapd xmm1, [esp + nb311_dyH1] movapd xmm2, [esp + nb311_dzH1] mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 ;# update H1 forces movapd xmm3, [esp + nb311_fixH1] movapd xmm4, [esp + nb311_fiyH1] movapd xmm7, [esp + nb311_fizH1] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb311_fixH1], xmm3 movapd [esp + nb311_fiyH1], xmm4 movapd [esp + nb311_fizH1], xmm7 ;# update j forces with water H1 addpd xmm0, [esp + nb311_fjx] addpd xmm1, [esp + nb311_fjy] addpd xmm2, [esp + nb311_fjz] movapd [esp + nb311_fjx], xmm0 movapd [esp + nb311_fjy], xmm1 movapd [esp + nb311_fjz], xmm2 ;# Done with H1, finally we do H2 interactions movapd xmm7, [esp + nb311_rH2] mulpd xmm7, [esp + nb311_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 + nb311_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 + nb311_two] ;# two*Heps2 movapd xmm3, [esp + nb311_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 + nb311_vctot] mulpd xmm3, [esp + nb311_rinvH2] movapd [esp + nb311_vctot], xmm5 mulpd xmm3, [esp + nb311_tsc] subpd xmm4, xmm3 movapd xmm0, [esp + nb311_dxH2] movapd xmm1, [esp + nb311_dyH2] movapd xmm2, [esp + nb311_dzH2] mulpd xmm0, xmm4 mulpd xmm1, xmm4 mulpd xmm2, xmm4 movd eax, mm0 movd ebx, mm1 ;# update H2 forces movapd xmm3, [esp + nb311_fixH2] movapd xmm4, [esp + nb311_fiyH2] movapd xmm7, [esp + nb311_fizH2] addpd xmm3, xmm0 addpd xmm4, xmm1 addpd xmm7, xmm2 movapd [esp + nb311_fixH2], xmm3 movapd [esp + nb311_fiyH2], xmm4 movapd [esp + nb311_fizH2], xmm7 mov edi, [ebp + nb311_faction] ;# update j forces addpd xmm0, [esp + nb311_fjx] addpd xmm1, [esp + nb311_fjy] addpd xmm2, [esp + nb311_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 + nb311_innerk], 2 jl .nb311_checksingle jmp .nb311_unroll_loop.nb311_checksingle: mov edx, [esp + nb311_innerk] and edx, 1 jnz .nb311_dosingle jmp .nb311_updateouterdata.nb311_dosingle: mov edx, [esp + nb311_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] mov esi, [ebp + nb311_charge] ;# base of charge[] xorpd xmm3, xmm3 movlpd xmm3, [esi + eax*8] movapd xmm4, xmm3 mulpd xmm3, [esp + nb311_iqO] mulpd xmm4, [esp + nb311_iqH] movd mm0, eax ;# use mmx registers as temp storage movapd [esp + nb311_qqO], xmm3 movapd [esp + nb311_qqH], xmm4 mov esi, [ebp + nb311_type] mov eax, [esi + eax*4] mov esi, [ebp + nb311_vdwparam] shl eax, 1 mov edi, [esp + nb311_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 + nb311_c6], xmm4 movapd [esp + nb311_c12], xmm6 mov esi, [ebp + nb311_pos] ;# base of pos[] lea eax, [eax + eax*2] ;# replace jnr with j3 ;# move coords 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 + nb311_ixO] movapd xmm5, [esp + nb311_iyO] movapd xmm6, [esp + nb311_izO] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# store dr movapd [esp + nb311_dxO], xmm4 movapd [esp + nb311_dyO], xmm5 movapd [esp + nb311_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 ;# move ixH1-izH1 to xmm4-xmm6 movapd xmm4, [esp + nb311_ixH1] movapd xmm5, [esp + nb311_iyH1] movapd xmm6, [esp + nb311_izH1] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# store dr movapd [esp + nb311_dxH1], xmm4 movapd [esp + nb311_dyH1], xmm5 movapd [esp + nb311_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 + nb311_ixH2] movapd xmm4, [esp + nb311_iyH2] movapd xmm5, [esp + nb311_izH2] ;# calc dr subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 ;# store dr movapd [esp + nb311_dxH2], xmm3 movapd [esp + nb311_dyH2], xmm4 movapd [esp + nb311_dzH2], xmm5 ;# square it mulsd xmm3,xmm3 mulsd xmm4,xmm4 mulsd xmm5,xmm5 addsd xmm5, xmm4 addsd xmm5, xmm3 ;# rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ;# start with rsqO - put seed in xmm2 cvtsd2ss xmm2, xmm7 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [esp + nb311_three]⌨️ 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?