nb_kernel133_x86_64_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,375 行 · 第 1/5 页
S
2,375 行
;# convert back to float cvtpi2pd xmm4, mm6 ;# multiply by 8 pslld mm6, 3 ;# calculate eps subpd xmm3, xmm4 ;# xmm3=eps ;# move to integer registers movd r10d, mm6 psrlq mm6, 32 movd r11d, mm6 ;# xmm3=eps ;# xmm15=rinv mov rsi, [rbp + nb133_VFtab] ;# indices in r10, r11. Load dispersion and repulsion tables in parallel. movapd xmm4, [rsi + r10*8] ;# Y1d F1d movapd xmm12, [rsi + r11*8] ;# Y2d F2d movapd xmm8, [rsi + r10*8 + 32] ;# Y1r F1r movapd xmm13, [rsi + r11*8 + 32] ;# Y2r F2r movapd xmm5, xmm4 movapd xmm9, xmm8 unpcklpd xmm4, xmm12 ;# Y1d Y2d unpckhpd xmm5, xmm12 ;# F1d F2d unpcklpd xmm8, xmm13 ;# Y1r Y2r unpckhpd xmm9, xmm13 ;# F1r F2r movapd xmm6, [rsi + r10*8 + 16] ;# G1d H1d movapd xmm12, [rsi + r11*8 + 16] ;# G2d H2d movapd xmm10, [rsi + r10*8 + 48] ;# G1r H1r movapd xmm13, [rsi + r11*8 + 48] ;# G2r H2r movapd xmm7, xmm6 movapd xmm11, xmm10 unpcklpd xmm6, xmm12 ;# G1d G2d unpckhpd xmm7, xmm12 ;# H1d H2d unpcklpd xmm10, xmm13 ;# G1r G2r unpckhpd xmm11, xmm13 ;# H1r H2r ;# dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulpd xmm7, xmm3 ;# Heps mulpd xmm11, xmm3 mulpd xmm6, xmm3 ;# Geps mulpd xmm10, xmm3 mulpd xmm7, xmm3 ;# Heps2 mulpd xmm11, xmm3 addpd xmm5, xmm6 ;# F+Geps addpd xmm9, xmm10 addpd xmm5, xmm7 ;# F+Geps+Heps2 = Fp addpd xmm9, xmm11 addpd xmm7, xmm7 ;# 2*Heps2 addpd xmm11, xmm11 addpd xmm7, xmm6 ;# 2*Heps2+Geps addpd xmm11, xmm10 addpd xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps addpd xmm11, xmm9 mulpd xmm5, xmm3 ;# eps*Fp mulpd xmm9, xmm3 movapd xmm12, [rsp + nb133_c6] movapd xmm13, [rsp + nb133_c12] addpd xmm5, xmm4 ;# VV addpd xmm9, xmm8 mulpd xmm5, xmm12 ;# VV*c6 = vnb6 mulpd xmm9, xmm13 ;# VV*c12 = vnb12 addpd xmm5, xmm9 addpd xmm5, [rsp + nb133_Vvdwtot] movapd [rsp + nb133_Vvdwtot], xmm5 mulpd xmm7, xmm12 ;# FF*c6 = fnb6 mulpd xmm11, xmm13 ;# FF*c12 = fnb12 addpd xmm7, xmm11 mulpd xmm7, [rsp + nb133_tsc] mulpd xmm7, xmm15 ;# -fscal xorpd xmm9, xmm9 subpd xmm9, xmm7 ;# fscal movapd xmm10, xmm9 movapd xmm11, xmm9 mulpd xmm9, [rsp + nb133_dxO] ;# fx/fy/fz mulpd xmm10, [rsp + nb133_dyO] mulpd xmm11, [rsp + nb133_dzO] ;# save j force temporarily movapd [rsp + nb133_fjx], xmm9 movapd [rsp + nb133_fjy], xmm10 movapd [rsp + nb133_fjz], xmm11 ;# increment i O force addpd xmm9, [rsp + nb133_fixO] addpd xmm10, [rsp + nb133_fiyO] addpd xmm11, [rsp + nb133_fizO] movapd [rsp + nb133_fixO], xmm9 movapd [rsp + nb133_fiyO], xmm10 movapd [rsp + nb133_fizO], xmm11 ;# finished O LJ interaction. ;# do H1, H2, and M interactions in parallel. ;# xmm0-xmm2 still contain j coordinates. movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 movapd xmm6, xmm0 movapd xmm7, xmm1 movapd xmm8, xmm2 subpd xmm0, [rsp + nb133_ixH1] subpd xmm1, [rsp + nb133_iyH1] subpd xmm2, [rsp + nb133_izH1] subpd xmm3, [rsp + nb133_ixH2] subpd xmm4, [rsp + nb133_iyH2] subpd xmm5, [rsp + nb133_izH2] subpd xmm6, [rsp + nb133_ixM] subpd xmm7, [rsp + nb133_iyM] subpd xmm8, [rsp + nb133_izM] movapd [rsp + nb133_dxH1], xmm0 movapd [rsp + nb133_dyH1], xmm1 movapd [rsp + nb133_dzH1], xmm2 mulpd xmm0, xmm0 mulpd xmm1, xmm1 mulpd xmm2, xmm2 movapd [rsp + nb133_dxH2], xmm3 movapd [rsp + nb133_dyH2], xmm4 movapd [rsp + nb133_dzH2], xmm5 mulpd xmm3, xmm3 mulpd xmm4, xmm4 mulpd xmm5, xmm5 movapd [rsp + nb133_dxM], xmm6 movapd [rsp + nb133_dyM], xmm7 movapd [rsp + nb133_dzM], xmm8 mulpd xmm6, xmm6 mulpd xmm7, xmm7 mulpd xmm8, xmm8 addpd xmm0, xmm1 addpd xmm0, xmm2 addpd xmm3, xmm4 addpd xmm3, xmm5 addpd xmm6, xmm7 addpd xmm6, xmm8 ;# start doing invsqrt for j atoms cvtpd2ps xmm1, xmm0 cvtpd2ps xmm4, xmm3 cvtpd2ps xmm7, xmm6 rsqrtps xmm1, xmm1 rsqrtps xmm4, xmm4 rsqrtps xmm7, xmm7 cvtps2pd xmm1, xmm1 cvtps2pd xmm4, xmm4 cvtps2pd xmm7, xmm7 movapd xmm2, xmm1 movapd xmm5, xmm4 movapd xmm8, xmm7 mulpd xmm1, xmm1 ;# lu*lu mulpd xmm4, xmm4 ;# lu*lu mulpd xmm7, xmm7 ;# lu*lu movapd xmm9, [rsp + nb133_three] movapd xmm10, xmm9 movapd xmm11, xmm9 mulpd xmm1, xmm0 ;# rsq*lu*lu mulpd xmm4, xmm3 ;# rsq*lu*lu mulpd xmm7, xmm6 ;# rsq*lu*lu subpd xmm9, xmm1 subpd xmm10, xmm4 subpd xmm11, xmm7 ;# 3-rsq*lu*lu mulpd xmm9, xmm2 mulpd xmm10, xmm5 mulpd xmm11, xmm8 ;# lu*(3-rsq*lu*lu) movapd xmm15, [rsp + nb133_half] mulpd xmm9, xmm15 ;# first iteration for rinvH1 mulpd xmm10, xmm15 ;# first iteration for rinvH2 mulpd xmm11, xmm15 ;# first iteration for rinvM ;# second iteration step movapd xmm2, xmm9 movapd xmm5, xmm10 movapd xmm8, xmm11 mulpd xmm2, xmm2 ;# lu*lu mulpd xmm5, xmm5 ;# lu*lu mulpd xmm8, xmm8 ;# lu*lu movapd xmm1, [rsp + nb133_three] movapd xmm4, xmm1 movapd xmm7, xmm1 mulpd xmm2, xmm0 ;# rsq*lu*lu mulpd xmm5, xmm3 ;# rsq*lu*lu mulpd xmm8, xmm6 ;# rsq*lu*lu subpd xmm1, xmm2 subpd xmm4, xmm5 subpd xmm7, xmm8 ;# 3-rsq*lu*lu mulpd xmm9, xmm1 mulpd xmm10, xmm4 mulpd xmm11, xmm7 ;# lu*(3-rsq*lu*lu) movapd xmm15, [rsp + nb133_half] mulpd xmm9, xmm15 ;# rinvH1 mulpd xmm10, xmm15 ;# rinvH2 mulpd xmm11, xmm15 ;# rinvM ;# interactions movapd xmm0, xmm9 movapd xmm1, xmm10 movapd xmm2, xmm11 mulpd xmm9, xmm9 mulpd xmm10, xmm10 mulpd xmm11, xmm11 mulpd xmm0, [rsp + nb133_qqH] mulpd xmm1, [rsp + nb133_qqH] mulpd xmm2, [rsp + nb133_qqM] mulpd xmm9, xmm0 mulpd xmm10, xmm1 mulpd xmm11, xmm2 addpd xmm0, [rsp + nb133_vctot] addpd xmm1, xmm2 addpd xmm0, xmm1 movapd [rsp + nb133_vctot], xmm0 ;# move j forces to xmm0-xmm2 mov rdi, [rbp + nb133_faction] movlpd xmm0, [rdi + rax*8] movlpd xmm1, [rdi + rax*8 + 8] movlpd xmm2, [rdi + rax*8 + 16] movhpd xmm0, [rdi + rbx*8] movhpd xmm1, [rdi + rbx*8 + 8] movhpd xmm2, [rdi + rbx*8 + 16] movapd xmm7, xmm9 movapd xmm8, xmm9 movapd xmm13, xmm11 movapd xmm14, xmm11 movapd xmm15, xmm11 movapd xmm11, xmm10 movapd xmm12, xmm10 ;# add forces from O interaction addpd xmm0, [rsp + nb133_fjx] addpd xmm1, [rsp + nb133_fjy] addpd xmm2, [rsp + nb133_fjz] mulpd xmm7, [rsp + nb133_dxH1] mulpd xmm8, [rsp + nb133_dyH1] mulpd xmm9, [rsp + nb133_dzH1] mulpd xmm10, [rsp + nb133_dxH2] mulpd xmm11, [rsp + nb133_dyH2] mulpd xmm12, [rsp + nb133_dzH2] mulpd xmm13, [rsp + nb133_dxM] mulpd xmm14, [rsp + nb133_dyM] mulpd xmm15, [rsp + nb133_dzM] addpd xmm0, xmm7 addpd xmm1, xmm8 addpd xmm2, xmm9 addpd xmm7, [rsp + nb133_fixH1] addpd xmm8, [rsp + nb133_fiyH1] addpd xmm9, [rsp + nb133_fizH1] addpd xmm0, xmm10 addpd xmm1, xmm11 addpd xmm2, xmm12 addpd xmm10, [rsp + nb133_fixH2] addpd xmm11, [rsp + nb133_fiyH2] addpd xmm12, [rsp + nb133_fizH2] addpd xmm0, xmm13 addpd xmm1, xmm14 addpd xmm2, xmm15 addpd xmm13, [rsp + nb133_fixM] addpd xmm14, [rsp + nb133_fiyM] addpd xmm15, [rsp + nb133_fizM] movapd [rsp + nb133_fixH1], xmm7 movapd [rsp + nb133_fiyH1], xmm8 movapd [rsp + nb133_fizH1], xmm9 movapd [rsp + nb133_fixH2], xmm10 movapd [rsp + nb133_fiyH2], xmm11 movapd [rsp + nb133_fizH2], xmm12 movapd [rsp + nb133_fixM], xmm13 movapd [rsp + nb133_fiyM], xmm14 movapd [rsp + nb133_fizM], xmm15 ;# store back j forces from xmm0-xmm2 movlpd [rdi + rax*8], xmm0 movlpd [rdi + rax*8 + 8], xmm1 movlpd [rdi + rax*8 + 16], xmm2 movhpd [rdi + rbx*8], xmm0 movhpd [rdi + rbx*8 + 8], xmm1 movhpd [rdi + rbx*8 + 16], xmm2 ;# should we do one more iteration? sub dword ptr [rsp + nb133_innerk], 2 jl .nb133_checksingle jmp .nb133_unroll_loop.nb133_checksingle: mov edx, [rsp + nb133_innerk] and edx, 1 jnz .nb133_dosingle jmp .nb133_updateouterdata.nb133_dosingle: mov rdx, [rsp + nb133_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] add qword ptr [rsp + nb133_innerjjnr], 4 mov rsi, [rbp + nb133_charge] ;# base of charge[] xorpd xmm3, xmm3 movlpd xmm3, [rsi + rax*8] movapd xmm4, xmm3 mulsd xmm3, [rsp + nb133_iqM] mulsd xmm4, [rsp + nb133_iqH] movapd [rsp + nb133_qqM], xmm3 movapd [rsp + nb133_qqH], xmm4 mov rsi, [rbp + nb133_type] mov r8d, [rsi + rax*4] mov rsi, [rbp + nb133_vdwparam] shl r8d, 1 mov edi, [rsp + nb133_ntia] add r8d, edi movlpd xmm6, [rsi + r8*8] ;# c6a movhpd xmm6, [rsi + r8*8 + 8] ;# c6a c12a xorpd xmm7, xmm7 movapd xmm4, xmm6 unpcklpd xmm4, xmm7 unpckhpd xmm6, xmm7 movapd [rsp + nb133_c6], xmm4 movapd [rsp + nb133_c12], xmm6 mov rsi, [rbp + nb133_pos] ;# base of pos[] lea rax, [rax + rax*2] ;# replace jnr with j3 ;# move coordinates to xmm0-xmm2 and xmm4-xmm6 movlpd xmm4, [rsi + rax*8] movlpd xmm5, [rsi + rax*8 + 8] movlpd xmm6, [rsi + rax*8 + 16] movapd xmm0, xmm4 movapd xmm1, xmm5 movapd xmm2, xmm6 ;# calc dr subsd xmm4, [rsp + nb133_ixO] subsd xmm5, [rsp + nb133_iyO] subsd xmm6, [rsp + nb133_izO] ;# store dr movapd [rsp + nb133_dxO], xmm4 movapd [rsp + nb133_dyO], xmm5 movapd [rsp + 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 [rsp + nb133_rsqO], xmm7 ;# move j coords to xmm4-xmm6 movapd xmm4, xmm0 movapd xmm5, xmm1 movapd xmm6, xmm2 ;# calc dr subsd xmm4, [rsp + nb133_ixH1] subsd xmm5, [rsp + nb133_iyH1] subsd xmm6, [rsp + nb133_izH1] ;# store dr movapd [rsp + nb133_dxH1], xmm4 movapd [rsp + nb133_dyH1], xmm5 movapd [rsp + nb133_dzH1], xmm6 ;# square it mulsd xmm4,xmm4 mulsd xmm5,xmm5 mulsd xmm6,xmm6 addsd xmm6, xmm5 addsd xmm6, xmm4 ;# rsqH1 in xmm6 ;# move j coords to xmm3-xmm5 movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 ;# calc dr subsd xmm3, [rsp + nb133_ixH2] subsd xmm4, [rsp + nb133_iyH2] subsd xmm5, [rsp + nb133_izH2] ;# store dr movapd [rsp + nb133_dxH2], xmm3 movapd [rsp + nb133_dyH2], xmm4 movapd [rsp + nb133_dzH2], xmm5 ;# square it mulsd xmm3,xmm3 mulsd xmm4,xmm4 mulsd xmm5,xmm5 addsd xmm5, xmm4 addsd xmm5, xmm3 ;# move j coords to xmm4-xmm2 movapd xmm4, xmm0 movapd xmm3, xmm1 ;# xmm2 already contains z ;# calc dr subsd xmm4, [rsp + nb133_ixM] subsd xmm3, [rsp + nb133_iyM] subsd xmm2, [rsp + nb133_izM] ;# store dr movapd [rsp + nb133_dxM], xmm4 movapd [rsp + nb133_dyM], xmm3 movapd [rsp + nb133_dzM], xmm2 ;# 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 ;# start with rsqH1 - put seed in xmm2 cvtsd2ss xmm2, xmm6 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm1, [rsp + nb133_three] mulsd xmm2, xmm6 ;# rsq*lu*lu subsd xmm1, xmm2 ;# 30-rsq*lu*lu mulsd xmm1, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm1, [rsp + nb133_half] ;# iter1 ( new lu) movapd xmm3, xmm1 mulsd xmm1, xmm1 ;# lu*lu mulsd xmm6, xmm1 ;# rsq*lu*lu movapd xmm1, [rsp + nb133_three] subsd xmm1, xmm6 ;# 3-rsq*lu*lu mulsd xmm1, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm1, [rsp + nb133_half] ;# rinv movapd [rsp + nb133_rinvH1], xmm1 ;# rsqH2 - seed in xmm2 cvtsd2ss xmm2, xmm5 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2⌨️ 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?