nb_kernel333_x86_64_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,340 行 · 第 1/5 页
S
2,340 行
mulpd xmm3, xmm15 ;# xmm3=r ;# xmm15=rinv ;# xmm3=r mulpd xmm3, [rsp + nb333_tsc] ;# rtab ;# truncate and convert to integers cvttpd2pi mm6, xmm3 ;# convert back to float cvtpi2pd xmm4, mm6 ;# multiply by 4 pslld mm6, 2 ;# calculate eps subpd xmm3, xmm4 ;# xmm3=eps ;# move to integer registers movd r10d, mm6 psrlq mm6, 32 movd r11d, mm6 ;# multiply by 3 lea r10, [r10 + r10*2] lea r11, [r11 + r11*2] ;# xmm3=eps ;# xmm15=rinv mov rsi, [rbp + nb333_VFtab] ;# indices in r10, r11. Load dispersion and repulsion tables in parallel. movapd xmm4, [rsi + r10*8 + 32] ;# Y1d F1d movapd xmm12, [rsi + r11*8 + 32] ;# Y2d F2d movapd xmm8, [rsi + r10*8 + 64] ;# Y1r F1r movapd xmm13, [rsi + r11*8 + 64] ;# 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 + 48] ;# G1d H1d movapd xmm12, [rsi + r11*8 + 48] ;# G2d H2d movapd xmm10, [rsi + r10*8 + 80] ;# G1r H1r movapd xmm13, [rsi + r11*8 + 80] ;# 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 + nb333_c6] movapd xmm13, [rsp + nb333_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 + nb333_Vvdwtot] movapd [rsp + nb333_Vvdwtot], xmm5 mulpd xmm7, xmm12 ;# FF*c6 = fnb6 mulpd xmm11, xmm13 ;# FF*c12 = fnb12 addpd xmm7, xmm11 mulpd xmm7, [rsp + nb333_tsc] mulpd xmm7, xmm15 ;# -fscal xorpd xmm9, xmm9 subpd xmm9, xmm7 ;# fscal movapd xmm10, xmm9 movapd xmm11, xmm9 mulpd xmm9, [rsp + nb333_dxO] ;# fx/fy/fz mulpd xmm10, [rsp + nb333_dyO] mulpd xmm11, [rsp + nb333_dzO] ;# save j force temporarily movapd [rsp + nb333_fjx], xmm9 movapd [rsp + nb333_fjy], xmm10 movapd [rsp + nb333_fjz], xmm11 ;# increment i O force addpd xmm9, [rsp + nb333_fixO] addpd xmm10, [rsp + nb333_fiyO] addpd xmm11, [rsp + nb333_fizO] movapd [rsp + nb333_fixO], xmm9 movapd [rsp + nb333_fiyO], xmm10 movapd [rsp + nb333_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 + nb333_ixH1] subpd xmm1, [rsp + nb333_iyH1] subpd xmm2, [rsp + nb333_izH1] subpd xmm3, [rsp + nb333_ixH2] subpd xmm4, [rsp + nb333_iyH2] subpd xmm5, [rsp + nb333_izH2] subpd xmm6, [rsp + nb333_ixM] subpd xmm7, [rsp + nb333_iyM] subpd xmm8, [rsp + nb333_izM] movapd [rsp + nb333_dxH1], xmm0 movapd [rsp + nb333_dyH1], xmm1 movapd [rsp + nb333_dzH1], xmm2 mulpd xmm0, xmm0 mulpd xmm1, xmm1 mulpd xmm2, xmm2 movapd [rsp + nb333_dxH2], xmm3 movapd [rsp + nb333_dyH2], xmm4 movapd [rsp + nb333_dzH2], xmm5 mulpd xmm3, xmm3 mulpd xmm4, xmm4 mulpd xmm5, xmm5 movapd [rsp + nb333_dxM], xmm6 movapd [rsp + nb333_dyM], xmm7 movapd [rsp + nb333_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 + nb333_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 + nb333_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 + nb333_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 + nb333_half] mulpd xmm9, xmm15 ;# rinvH1 mulpd xmm10, xmm15 ;# rinvH2 mulpd xmm11, xmm15 ;# rinvM movapd [rsp + nb333_rinvH1], xmm9 movapd [rsp + nb333_rinvH2], xmm10 movapd [rsp + nb333_rinvM], xmm11 ;# interactions ;# rsq in xmm0,xmm3,xmm6 ;# rinv in xmm9, xmm10, xmm11 movapd xmm1, [rsp + nb333_tsc] mulpd xmm0, xmm9 ;# r mulpd xmm3, xmm10 mulpd xmm6, xmm11 mulpd xmm0, xmm1 ;# rtab mulpd xmm3, xmm1 mulpd xmm6, xmm1 ;# truncate and convert to integers cvttpd2dq xmm1, xmm0 cvttpd2dq xmm4, xmm3 cvttpd2dq xmm7, xmm6 ;# convert back to float cvtdq2pd xmm2, xmm1 cvtdq2pd xmm5, xmm4 cvtdq2pd xmm8, xmm7 ;# multiply by 4 pslld xmm1, 2 pslld xmm4, 2 pslld xmm7, 2 ;# multiply by three (copy, mult. by two, add back) movapd xmm10, xmm1 movapd xmm11, xmm4 movapd xmm12, xmm7 pslld xmm1, 1 pslld xmm4, 1 pslld xmm7, 1 paddd xmm1, xmm10 paddd xmm4, xmm11 paddd xmm7, xmm12 ;# move to integer registers pshufd xmm13, xmm1, 1 pshufd xmm14, xmm4, 1 pshufd xmm15, xmm7, 1 movd r8d, xmm1 movd r10d, xmm4 movd r12d, xmm7 movd r9d, xmm13 movd r11d, xmm14 movd r13d, xmm15 mov rsi, [rbp + nb333_VFtab] ;# calculate eps subpd xmm0, xmm2 subpd xmm3, xmm5 subpd xmm6, xmm8 movapd xmm12, xmm0 ;# epsH1 movapd xmm13, xmm3 ;# epsH2 movapd xmm14, xmm6 ;# epsM ;# Load LOTS of table data movlpd xmm0, [rsi + r8*8] movlpd xmm1, [rsi + r8*8 + 8] movlpd xmm2, [rsi + r8*8 + 16] movlpd xmm3, [rsi + r8*8 + 24] movlpd xmm4, [rsi + r10*8] movlpd xmm5, [rsi + r10*8 + 8] movlpd xmm6, [rsi + r10*8 + 16] movlpd xmm7, [rsi + r10*8 + 24] movlpd xmm8, [rsi + r12*8] movlpd xmm9, [rsi + r12*8 + 8] movlpd xmm10, [rsi + r12*8 + 16] movlpd xmm11, [rsi + r12*8 + 24] movhpd xmm0, [rsi + r9*8] movhpd xmm1, [rsi + r9*8 + 8] movhpd xmm2, [rsi + r9*8 + 16] movhpd xmm3, [rsi + r9*8 + 24] movhpd xmm4, [rsi + r11*8] movhpd xmm5, [rsi + r11*8 + 8] movhpd xmm6, [rsi + r11*8 + 16] movhpd xmm7, [rsi + r11*8 + 24] movhpd xmm8, [rsi + r13*8] movhpd xmm9, [rsi + r13*8 + 8] movhpd xmm10, [rsi + r13*8 + 16] movhpd xmm11, [rsi + r13*8 + 24] ;# table data ready in xmm0-xmm3 , xmm4-xmm7 , and xmm8-xmm11 mulpd xmm3, xmm12 ;# Heps mulpd xmm7, xmm13 mulpd xmm11, xmm14 mulpd xmm2, xmm12 ;# Geps mulpd xmm6, xmm13 mulpd xmm10, xmm14 mulpd xmm3, xmm12 ;# Heps2 mulpd xmm7, xmm13 mulpd xmm11, xmm14 addpd xmm1, xmm2 ;# F+Geps addpd xmm5, xmm6 addpd xmm9, xmm10 addpd xmm1, xmm3 ;# F+Geps+Heps2 = Fp addpd xmm5, xmm7 addpd xmm9, xmm11 addpd xmm3, xmm3 ;# 2*Heps2 addpd xmm7, xmm7 addpd xmm11, xmm11 addpd xmm3, xmm2 ;# 2*Heps2+Geps addpd xmm7, xmm6 addpd xmm11, xmm10 addpd xmm3, xmm1 ;# FF = Fp + 2*Heps2 + Geps addpd xmm7, xmm5 addpd xmm11, xmm9 mulpd xmm1, xmm12 ;# eps*Fp mulpd xmm5, xmm13 mulpd xmm9, xmm14 movapd xmm12, [rsp + nb333_qqH] movapd xmm13, [rsp + nb333_qqM] addpd xmm1, xmm0 ;# VV addpd xmm5, xmm4 addpd xmm9, xmm8 mulpd xmm1, xmm12 ;# VV*qq = vcoul mulpd xmm5, xmm12 mulpd xmm9, xmm13 mulpd xmm3, xmm12 ;# FF*qq = fij mulpd xmm7, xmm12 mulpd xmm11, xmm13 ;# accumulate vctot addpd xmm1, [rsp + nb333_vctot] addpd xmm5, xmm9 addpd xmm1, xmm5 movapd [rsp + nb333_vctot], xmm1 movapd xmm10, [rsp + nb333_tsc] mulpd xmm3, xmm10 ;# fscal mulpd xmm7, xmm10 mulpd xmm10, xmm11 xorpd xmm4, xmm4 xorpd xmm8, xmm8 xorpd xmm11, xmm11 subpd xmm4, xmm3 subpd xmm8, xmm7 subpd xmm11, xmm10 mulpd xmm4, [rsp + nb333_rinvH1] mulpd xmm8, [rsp + nb333_rinvH2] mulpd xmm11, [rsp + nb333_rinvM] ;# move j forces to xmm0-xmm2 mov rdi, [rbp + nb333_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 xmm3, xmm4 movapd xmm5, xmm4 movapd xmm7, xmm8 movapd xmm9, xmm8 movapd xmm10, xmm11 movapd xmm12, xmm11 ;# add forces from O interaction addpd xmm0, [rsp + nb333_fjx] addpd xmm1, [rsp + nb333_fjy] addpd xmm2, [rsp + nb333_fjz] mulpd xmm3, [rsp + nb333_dxH1] mulpd xmm4, [rsp + nb333_dyH1] mulpd xmm5, [rsp + nb333_dzH1] mulpd xmm7, [rsp + nb333_dxH2] mulpd xmm8, [rsp + nb333_dyH2] mulpd xmm9, [rsp + nb333_dzH2] mulpd xmm10, [rsp + nb333_dxM] mulpd xmm11, [rsp + nb333_dyM] mulpd xmm12, [rsp + nb333_dzM] addpd xmm0, xmm3 addpd xmm1, xmm4 addpd xmm2, xmm5 addpd xmm3, [rsp + nb333_fixH1] addpd xmm4, [rsp + nb333_fiyH1] addpd xmm5, [rsp + nb333_fizH1] addpd xmm0, xmm7 addpd xmm1, xmm8 addpd xmm2, xmm9 addpd xmm7, [rsp + nb333_fixH2] addpd xmm8, [rsp + nb333_fiyH2] addpd xmm9, [rsp + nb333_fizH2] addpd xmm0, xmm10 addpd xmm1, xmm11 addpd xmm2, xmm12 addpd xmm10, [rsp + nb333_fixM] addpd xmm11, [rsp + nb333_fiyM] addpd xmm12, [rsp + nb333_fizM] movapd [rsp + nb333_fixH1], xmm3 movapd [rsp + nb333_fiyH1], xmm4 movapd [rsp + nb333_fizH1], xmm5 movapd [rsp + nb333_fixH2], xmm7 movapd [rsp + nb333_fiyH2], xmm8 movapd [rsp + nb333_fizH2], xmm9 movapd [rsp + nb333_fixM], xmm10 movapd [rsp + nb333_fiyM], xmm11 movapd [rsp + nb333_fizM], xmm12 ;# 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 + nb333_innerk], 2 jl .nb333_checksingle jmp .nb333_unroll_loop.nb333_checksingle: mov edx, [rsp + nb333_innerk] and edx, 1 jnz .nb333_dosingle jmp .nb333_updateouterdata.nb333_dosingle: mov rdx, [rsp + nb333_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] mov ebx, [rdx + 4]⌨️ 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?