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
add ebx, edi movlpd xmm6, [esi + eax*8] ;# c6a movlpd xmm7, [esi + ebx*8] ;# c6b movhpd xmm6, [esi + eax*8 + 8] ;# c6a c12a movhpd xmm7, [esi + ebx*8 + 8] ;# c6b c12b movapd xmm4, xmm6 unpcklpd xmm4, xmm7 unpckhpd xmm6, xmm7 movd eax, mm0 movd ebx, mm1 movapd [esp + nb313nf_c6], xmm4 movapd [esp + nb313nf_c12], xmm6 mov esi, [ebp + nb313nf_pos] ;# base of pos[] lea eax, [eax + eax*2] ;# replace jnr with j3 lea ebx, [ebx + ebx*2] ;# move two coordinates to xmm0-xmm2 movlpd xmm0, [esi + eax*8] movlpd xmm1, [esi + eax*8 + 8] movlpd xmm2, [esi + eax*8 + 16] movhpd xmm0, [esi + ebx*8] movhpd xmm1, [esi + ebx*8 + 8] movhpd xmm2, [esi + ebx*8 + 16] ;# move ixO-izO to xmm4-xmm6 movapd xmm4, [esp + nb313nf_ixO] movapd xmm5, [esp + nb313nf_iyO] movapd xmm6, [esp + nb313nf_izO] ;# calc dr subpd xmm4, xmm0 subpd xmm5, xmm1 subpd xmm6, xmm2 ;# square it mulpd xmm4,xmm4 mulpd xmm5,xmm5 mulpd xmm6,xmm6 addpd xmm4, xmm5 addpd xmm4, xmm6 movapd xmm7, xmm4 ;# rsqO in xmm7 ;# move ixH1-izH1 to xmm4-xmm6 movapd xmm4, [esp + nb313nf_ixH1] movapd xmm5, [esp + nb313nf_iyH1] movapd xmm6, [esp + nb313nf_izH1] ;# calc dr subpd xmm4, xmm0 subpd xmm5, xmm1 subpd xmm6, xmm2 ;# square it mulpd xmm4,xmm4 mulpd xmm5,xmm5 mulpd xmm6,xmm6 addpd xmm6, xmm5 addpd xmm6, xmm4 ;# rsqH1 in xmm6 ;# move ixH2-izH2 to xmm3-xmm5 movapd xmm3, [esp + nb313nf_ixH2] movapd xmm4, [esp + nb313nf_iyH2] movapd xmm5, [esp + nb313nf_izH2] ;# calc dr subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 ;# 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 + nb313nf_iyM] movapd xmm4, [esp + nb313nf_izM] subpd xmm3, xmm1 subpd xmm4, xmm2 movapd xmm2, [esp + nb313nf_ixM] subpd xmm2, xmm0 ;# 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 + nb313nf_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 + nb313nf_rinvsqO], xmm0 ;# rsqH1 - seed in xmm2 cvtpd2ps xmm2, xmm6 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm0, [esp + nb313nf_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 + nb313nf_half] ;# iter1 ( new lu) movapd xmm2, xmm6 movapd xmm3, xmm0 mulpd xmm0, xmm0 ;# lu*lu mulpd xmm2, xmm0 ;# rsq*lu*lu movapd xmm0, [esp + nb313nf_three] subpd xmm0, xmm2 ;# 3-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm0, [esp + nb313nf_half] ;# rinv movapd [esp + nb313nf_rinvH1], xmm0 ;# rinvH1 mulpd xmm6, xmm0 movapd [esp + nb313nf_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 + nb313nf_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 + nb313nf_half] ;# iter1 ( new lu) movapd xmm2, xmm5 movapd xmm3, xmm0 mulpd xmm0, xmm0 ;# lu*lu mulpd xmm2, xmm0 ;# rsq*lu*lu movapd xmm0, [esp + nb313nf_three] subpd xmm0, xmm2 ;# 3-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm0, [esp + nb313nf_half] ;# rinv movapd [esp + nb313nf_rinvH2], xmm0 ;# rinv mulpd xmm5, xmm0 movapd [esp + nb313nf_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 + nb313nf_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 + nb313nf_half] ;# iter1 ( new lu) movapd xmm2, xmm4 movapd xmm3, xmm0 mulpd xmm0, xmm0 ;# lu*lu mulpd xmm2, xmm0 ;# rsq*lu*lu movapd xmm0, [esp + nb313nf_three] subpd xmm0, xmm2 ;# 3-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm0, [esp + nb313nf_half] ;# rinv movapd [esp + nb313nf_rinvM], xmm0 ;# rinv mulpd xmm4, xmm0 movapd [esp + nb313nf_rM], xmm4 ;# r ;# do O interactions movapd xmm0, [esp + nb313nf_rinvsqO] movapd xmm1, xmm0 mulpd xmm1, xmm1 ;# rinv4 mulpd xmm1, xmm0 ;#rinvsix movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# rinvtwelve mulpd xmm1, [esp + nb313nf_c6] mulpd xmm2, [esp + nb313nf_c12] movapd xmm3, xmm2 subpd xmm3, xmm1 ;# Vvdw=Vvdw12-Vvdw6 addpd xmm3, [esp + nb313nf_Vvdwtot] movapd [esp + nb313nf_Vvdwtot], xmm3 ;# Done with O interactions - now H1! movapd xmm7, [esp + nb313nf_rH1] mulpd xmm7, [esp + nb313nf_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 + nb313nf_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 movapd xmm3, [esp + nb313nf_qqH] mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm3 ;# vcoul=qq*VV ;# increment vcoul addpd xmm5, [esp + nb313nf_vctot] movapd [esp + nb313nf_vctot], xmm5 ;# H2 interactions movapd xmm7, [esp + nb313nf_rH2] mulpd xmm7, [esp + nb313nf_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 + nb313nf_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 movapd xmm3, [esp + nb313nf_qqH] mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm3 ;# vcoul=qq*VV ;# increment vcoul addpd xmm5, [esp + nb313nf_vctot] movapd [esp + nb313nf_vctot], xmm5 ;# M interactions movapd xmm7, [esp + nb313nf_rM] mulpd xmm7, [esp + nb313nf_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 + nb313nf_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 movapd xmm3, [esp + nb313nf_qqM] mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm3 ;# vcoul=qq*VV ;# increment vcoul addpd xmm5, [esp + nb313nf_vctot] movapd [esp + nb313nf_vctot], xmm5 ;# should we do one more iteration? sub dword ptr [esp + nb313nf_innerk], 2 jl .nb313nf_checksingle jmp .nb313nf_unroll_loop.nb313nf_checksingle: mov edx, [esp + nb313nf_innerk] and edx, 1 jnz .nb313nf_dosingle jmp .nb313nf_updateouterdata.nb313nf_dosingle: mov edx, [esp + nb313nf_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] mov esi, [ebp + nb313nf_charge] ;# base of charge[] xorpd xmm3, xmm3 movlpd xmm3, [esi + eax*8] movapd xmm4, xmm3 mulpd xmm3, [esp + nb313nf_iqM] mulpd xmm4, [esp + nb313nf_iqH] movd mm0, eax ;# use mmx registers as temp storage movapd [esp + nb313nf_qqM], xmm3 movapd [esp + nb313nf_qqH], xmm4 mov esi, [ebp + nb313nf_type] mov eax, [esi + eax*4] mov esi, [ebp + nb313nf_vdwparam] shl eax, 1 mov edi, [esp + nb313nf_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 + nb313nf_c6], xmm4 movapd [esp + nb313nf_c12], xmm6 mov esi, [ebp + nb313nf_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 + nb313nf_ixO] movapd xmm5, [esp + nb313nf_iyO] movapd xmm6, [esp + nb313nf_izO] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# 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 + nb313nf_ixH1] movapd xmm5, [esp + nb313nf_iyH1] movapd xmm6, [esp + nb313nf_izH1] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# 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 + nb313nf_ixH2] movapd xmm4, [esp + nb313nf_iyH2] movapd xmm5, [esp + nb313nf_izH2] ;# calc dr subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 ;# square it mulsd xmm3,xmm3 mulsd xmm4,xmm4 mulsd xmm5,xmm5 addsd xmm5, xmm4 addsd xmm5, xmm3 ;# move ixM-izM to xmm2-xmm4 movapd xmm3, [esp + nb313nf_iyM] movapd xmm4, [esp + nb313nf_izM] subpd xmm3, xmm1 subpd xmm4, xmm2 movapd xmm2, [esp + nb313nf_ixM] subpd xmm2, xmm0 ;# 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 cvtsd2ss xmm2, xmm7 movsd xmm3, xmm7 rcpps xmm2, xmm2 cvtss2sd xmm2, xmm2 movsd xmm1, [esp + nb313nf_two] movsd xmm0, xmm1 mulsd xmm7, xmm2 subsd xmm1, xmm7 mulsd xmm2, xmm1 ;# iter1 mulsd xmm3, xmm2 subsd xmm⌨️ 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?