nb_kernel131_x86_64_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,068 行 · 第 1/4 页
S
2,068 行
movapd [rsp + nb131_rinvO], xmm9 movapd [rsp + nb131_rinvH1], xmm10 movapd [rsp + nb131_rinvH2], xmm11 ;# table LJ interaction mulpd xmm0, xmm9 mulpd xmm0, [rsp + nb131_tsc] ;# rtab ;# truncate and convert to integers cvttpd2dq xmm1, xmm0 ;# convert back to float cvtdq2pd xmm2, xmm1 ;# multiply by 8 pslld xmm1, 3 ;# move to integer registers pshufd xmm13, xmm1, 1 movd r8d, xmm1 movd r10d, xmm13 ;# calculate eps subpd xmm0, xmm2 mov rsi, [rbp + nb131_VFtab] movlpd xmm4, [rsi + r8*8] movlpd xmm5, [rsi + r8*8 + 8] movlpd xmm6, [rsi + r8*8 + 16] movlpd xmm7, [rsi + r8*8 + 24] movlpd xmm8, [rsi + r8*8 + 32] movlpd xmm9, [rsi + r8*8 + 40] movlpd xmm10, [rsi + r8*8 + 48] movlpd xmm11, [rsi + r8*8 + 56] movhpd xmm4, [rsi + r10*8] movhpd xmm5, [rsi + r10*8 + 8] movhpd xmm6, [rsi + r10*8 + 16] movhpd xmm7, [rsi + r10*8 + 24] movhpd xmm8, [rsi + r10*8 + 32] movhpd xmm9, [rsi + r10*8 + 40] movhpd xmm10, [rsi + r10*8 + 48] movhpd xmm11, [rsi + r10*8 + 56] ;# dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulpd xmm7, xmm0 ;# Heps mulpd xmm11, xmm0 mulpd xmm6, xmm0 ;# Geps mulpd xmm10, xmm0 mulpd xmm7, xmm0 ;# Heps2 mulpd xmm11, xmm0 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, xmm0 ;# eps*Fp mulpd xmm9, xmm0 movapd xmm12, [rsp + nb131_c6] movapd xmm13, [rsp + nb131_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 + nb131_Vvdwtot] movapd [rsp + nb131_Vvdwtot], xmm5 mulpd xmm7, xmm12 ;# FF*c6 = fnb6 mulpd xmm11, xmm13 ;# FF*c12 = fnb12 addpd xmm7, xmm11 mulpd xmm7, [rsp + nb131_tsc] movapd xmm9, [rsp + nb131_rinvO] movapd xmm10, [rsp + nb131_rinvH1] movapd xmm11, [rsp + nb131_rinvH2] movapd xmm0, xmm9 movapd xmm1, xmm10 movapd xmm2, xmm11 mulpd xmm10, xmm10 mulpd xmm11, xmm11 mulpd xmm0, [rsp + nb131_qqO] mulpd xmm1, [rsp + nb131_qqH] mulpd xmm2, [rsp + nb131_qqH] mulpd xmm9, xmm0 mulpd xmm10, xmm1 mulpd xmm11, xmm2 subpd xmm9, xmm7 mulpd xmm9, [rsp + nb131_rinvO] addpd xmm0, [rsp + nb131_vctot] addpd xmm1, xmm2 addpd xmm0, xmm1 movapd [rsp + nb131_vctot], xmm0 ;# move j forces to xmm0-xmm2 mov rdi, [rbp + nb131_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 mulpd xmm7, [rsp + nb131_dxO] mulpd xmm8, [rsp + nb131_dyO] mulpd xmm9, [rsp + nb131_dzO] mulpd xmm10, [rsp + nb131_dxH1] mulpd xmm11, [rsp + nb131_dyH1] mulpd xmm12, [rsp + nb131_dzH1] mulpd xmm13, [rsp + nb131_dxH2] mulpd xmm14, [rsp + nb131_dyH2] mulpd xmm15, [rsp + nb131_dzH2] addpd xmm0, xmm7 addpd xmm1, xmm8 addpd xmm2, xmm9 addpd xmm7, [rsp + nb131_fixO] addpd xmm8, [rsp + nb131_fiyO] addpd xmm9, [rsp + nb131_fizO] addpd xmm0, xmm10 addpd xmm1, xmm11 addpd xmm2, xmm12 addpd xmm10, [rsp + nb131_fixH1] addpd xmm11, [rsp + nb131_fiyH1] addpd xmm12, [rsp + nb131_fizH1] addpd xmm0, xmm13 addpd xmm1, xmm14 addpd xmm2, xmm15 addpd xmm13, [rsp + nb131_fixH2] addpd xmm14, [rsp + nb131_fiyH2] addpd xmm15, [rsp + nb131_fizH2] movapd [rsp + nb131_fixO], xmm7 movapd [rsp + nb131_fiyO], xmm8 movapd [rsp + nb131_fizO], xmm9 movapd [rsp + nb131_fixH1], xmm10 movapd [rsp + nb131_fiyH1], xmm11 movapd [rsp + nb131_fizH1], xmm12 movapd [rsp + nb131_fixH2], xmm13 movapd [rsp + nb131_fiyH2], xmm14 movapd [rsp + nb131_fizH2], 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 + nb131_innerk], 2 jl .nb131_checksingle jmp .nb131_unroll_loop.nb131_checksingle: mov edx, [rsp + nb131_innerk] and edx, 1 jnz .nb131_dosingle jmp .nb131_updateouterdata.nb131_dosingle: mov rdx, [rsp + nb131_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] add qword ptr [rsp + nb131_innerjjnr], 8 ;# advance pointer (unrolled 2) mov rsi, [rbp + nb131_charge] ;# base of charge[] movsd xmm3, [rsi + rax*8] movapd xmm4, xmm3 mulsd xmm3, [rsp + nb131_iqO] mulsd xmm4, [rsp + nb131_iqH] movapd [rsp + nb131_qqO], xmm3 movapd [rsp + nb131_qqH], xmm4 mov rsi, [rbp + nb131_type] mov r8d, [rsi + rax*4] mov rsi, [rbp + nb131_vdwparam] shl r8d, 1 mov edi, [rsp + nb131_ntia] add r8d, edi movsd xmm6, [rsi + r8*8] ;# c6a movsd xmm7, [rsi + r8*8 + 8] ;# c12a movapd [rsp + nb131_c6], xmm6 movapd [rsp + nb131_c12], xmm7 mov rsi, [rbp + nb131_pos] ;# base of pos[] lea rax, [rax + rax*2] ;# replace jnr with j3 ;# move j coordinates to local temp variables movsd xmm0, [rsi + rax*8] movsd xmm1, [rsi + rax*8 + 8] movsd xmm2, [rsi + rax*8 + 16] ;# xmm0 = jx ;# xmm1 = jy ;# xmm2 = jz movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 movapd xmm6, xmm0 movapd xmm7, xmm1 movapd xmm8, xmm2 subsd xmm0, [rsp + nb131_ixO] subsd xmm1, [rsp + nb131_iyO] subsd xmm2, [rsp + nb131_izO] subsd xmm3, [rsp + nb131_ixH1] subsd xmm4, [rsp + nb131_iyH1] subsd xmm5, [rsp + nb131_izH1] subsd xmm6, [rsp + nb131_ixH2] subsd xmm7, [rsp + nb131_iyH2] subsd xmm8, [rsp + nb131_izH2] movapd [rsp + nb131_dxO], xmm0 movapd [rsp + nb131_dyO], xmm1 movapd [rsp + nb131_dzO], xmm2 mulsd xmm0, xmm0 mulsd xmm1, xmm1 mulsd xmm2, xmm2 movapd [rsp + nb131_dxH1], xmm3 movapd [rsp + nb131_dyH1], xmm4 movapd [rsp + nb131_dzH1], xmm5 mulsd xmm3, xmm3 mulsd xmm4, xmm4 mulsd xmm5, xmm5 movapd [rsp + nb131_dxH2], xmm6 movapd [rsp + nb131_dyH2], xmm7 movapd [rsp + nb131_dzH2], xmm8 mulsd xmm6, xmm6 mulsd xmm7, xmm7 mulsd xmm8, xmm8 addsd xmm0, xmm1 addsd xmm0, xmm2 addsd xmm3, xmm4 addsd xmm3, xmm5 addsd xmm6, xmm7 addsd xmm6, xmm8 ;# start doing invsqrt for j atoms cvtsd2ss xmm1, xmm0 cvtsd2ss xmm4, xmm3 cvtsd2ss xmm7, xmm6 rsqrtss xmm1, xmm1 rsqrtss xmm4, xmm4 rsqrtss xmm7, xmm7 cvtss2sd xmm1, xmm1 cvtss2sd xmm4, xmm4 cvtss2sd xmm7, xmm7 movapd xmm2, xmm1 movapd xmm5, xmm4 movapd xmm8, xmm7 mulsd xmm1, xmm1 ;# lu*lu mulsd xmm4, xmm4 ;# lu*lu mulsd xmm7, xmm7 ;# lu*lu movapd xmm9, [rsp + nb131_three] movapd xmm10, xmm9 movapd xmm11, xmm9 mulsd xmm1, xmm0 ;# rsq*lu*lu mulsd xmm4, xmm3 ;# rsq*lu*lu mulsd xmm7, xmm6 ;# rsq*lu*lu subsd xmm9, xmm1 subsd xmm10, xmm4 subsd xmm11, xmm7 ;# 3-rsq*lu*lu mulsd xmm9, xmm2 mulsd xmm10, xmm5 mulsd xmm11, xmm8 ;# lu*(3-rsq*lu*lu) movapd xmm15, [rsp + nb131_half] mulsd xmm9, xmm15 ;# first iteration for rinvO mulsd xmm10, xmm15 ;# first iteration for rinvH1 mulsd xmm11, xmm15 ;# first iteration for rinvH2 ;# second iteration step movapd xmm2, xmm9 movapd xmm5, xmm10 movapd xmm8, xmm11 mulsd xmm2, xmm2 ;# lu*lu mulsd xmm5, xmm5 ;# lu*lu mulsd xmm8, xmm8 ;# lu*lu movapd xmm1, [rsp + nb131_three] movapd xmm4, xmm1 movapd xmm7, xmm1 mulsd xmm2, xmm0 ;# rsq*lu*lu mulsd xmm5, xmm3 ;# rsq*lu*lu mulsd xmm8, xmm6 ;# rsq*lu*lu subsd xmm1, xmm2 subsd xmm4, xmm5 subsd xmm7, xmm8 ;# 3-rsq*lu*lu mulsd xmm9, xmm1 mulsd xmm10, xmm4 mulsd xmm11, xmm7 ;# lu*(3-rsq*lu*lu) movapd xmm15, [rsp + nb131_half] mulsd xmm9, xmm15 ;# rinvO mulsd xmm10, xmm15 ;# rinvH1 mulsd xmm11, xmm15 ;# rinvH2 ;# O interactions ;# rsq in xmm0,xmm3,xmm6 ;# rinv in xmm9, xmm10, xmm11 movapd [rsp + nb131_rsqO], xmm0 movapd [rsp + nb131_rsqH1], xmm3 movapd [rsp + nb131_rsqH2], xmm6 movapd [rsp + nb131_rinvO], xmm9 movapd [rsp + nb131_rinvH1], xmm10 movapd [rsp + nb131_rinvH2], xmm11 ;# table LJ interaction mulsd xmm0, xmm9 mulsd xmm0, [rsp + nb131_tsc] ;# rtab ;# truncate and convert to integers cvttsd2si r8d, xmm0 ;# convert back to float cvtsi2sd xmm2, r8d ;# mult. by 8 shl r8d, 3 ;# calculate eps subsd xmm0, xmm2 mov rsi, [rbp + nb131_VFtab] movsd xmm4, [rsi + r8*8] movsd xmm5, [rsi + r8*8 + 8] movsd xmm6, [rsi + r8*8 + 16] movsd xmm7, [rsi + r8*8 + 24] movsd xmm8, [rsi + r8*8 + 32] movsd xmm9, [rsi + r8*8 + 40] movsd xmm10, [rsi + r8*8 + 48] movsd xmm11, [rsi + r8*8 + 56] ;# dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulsd xmm7, xmm0 ;# Heps mulsd xmm11, xmm0 mulsd xmm6, xmm0 ;# Geps mulsd xmm10, xmm0 mulsd xmm7, xmm0 ;# Heps2 mulsd xmm11, xmm0 addsd xmm5, xmm6 ;# F+Geps addsd xmm9, xmm10 addsd xmm5, xmm7 ;# F+Geps+Heps2 = Fp addsd xmm9, xmm11 addsd xmm7, xmm7 ;# 2*Heps2 addsd xmm11, xmm11 addsd xmm7, xmm6 ;# 2*Heps2+Geps addsd xmm11, xmm10 addsd xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps addsd xmm11, xmm9 mulsd xmm5, xmm0 ;# eps*Fp mulsd xmm9, xmm0 movapd xmm12, [rsp + nb131_c6] movapd xmm13, [rsp + nb131_c12] addsd xmm5, xmm4 ;# VV addsd xmm9, xmm8 mulsd xmm5, xmm12 ;# VV*c6 = vnb6 mulsd xmm9, xmm13 ;# VV*c12 = vnb12 addsd xmm5, xmm9 addsd xmm5, [rsp + nb131_Vvdwtot] movsd [rsp + nb131_Vvdwtot], xmm5 mulsd xmm7, xmm12 ;# FF*c6 = fnb6 mulsd xmm11, xmm13 ;# FF*c12 = fnb12 addsd xmm7, xmm11 mulsd xmm7, [rsp + nb131_tsc] movapd xmm9, [rsp + nb131_rinvO] movapd xmm10, [rsp + nb131_rinvH1] movapd xmm11, [rsp + nb131_rinvH2] movapd xmm0, xmm9 movapd xmm1, xmm10 movapd xmm2, xmm11 mulsd xmm10, xmm10 mulsd xmm11, xmm11 mulsd xmm0, [rsp + nb131_qqO] mulsd xmm1, [rsp + nb131_qqH] mulsd xmm2, [rsp + nb131_qqH] mulsd xmm9, xmm0 mulsd xmm10, xmm1 mulsd xmm11, xmm2 subsd xmm9, xmm7 mulsd xmm9, [rsp + nb131_rinvO] addsd xmm0, [rsp + nb131_vctot] addsd xmm1, xmm2 addsd xmm0, xmm1 movsd [rsp + nb131_vctot], xmm0 ;# move j forces to xmm0-xmm2 mov rdi, [rbp + nb131_faction] movsd xmm0, [rdi + rax*8] movsd xmm1, [rdi + rax*8 + 8] movsd xmm2, [rdi + rax*8 + 16] movapd xmm7, xmm9 movapd xmm8, xmm9 movapd xmm13, xmm11 movapd xmm14, xmm11 movapd xmm15, xmm11 movapd xmm11, xmm10 movapd xmm12, xmm10 mulsd xmm7, [rsp + nb131_dxO] mulsd xmm8, [rsp + nb131_dyO] mulsd xmm9, [rsp + nb131_dzO] mulsd xmm10, [rsp + nb131_dxH1] mulsd xmm11, [rsp + nb131_dyH1] mulsd xmm12, [rsp + nb131_dzH1] mulsd xmm13, [rsp + nb131_dxH2] mulsd xmm14, [rsp + nb131_dyH2] mulsd xmm15, [rsp + nb131_dzH2] addsd xmm0, xmm7 addsd xmm1, xmm8 addsd xmm2, xmm9 addsd xmm7, [rsp + nb131_fixO] addsd xmm8, [rsp + nb131_fiyO] addsd xmm9, [rsp + nb131_fizO] addsd xmm0, xmm10 addsd xmm1, xmm11 addsd xmm2, xmm12 addsd xmm10, [rsp + nb131_fixH1] addsd xmm11, [rsp + nb131_fiyH1] addsd xmm12, [rsp + nb131_fizH1] addsd xmm0, xmm13 addsd xmm1, xmm14 addsd xmm2, xmm15 addsd xmm13, [rsp + nb131_fixH2] addsd xmm14, [rsp + nb131_fiyH2] addsd xmm15, [rsp + nb131_fizH2] movsd [rsp + nb131_fixO], xmm7 movsd [rsp + nb131_fiyO], xmm8 movsd [rsp + nb131_fizO], xmm9 movsd [rsp + nb131_fixH1], xmm10 movsd [rsp + nb131_fiyH1], xmm11 movsd [rsp + nb131_fizH1], xmm12 movsd [rsp + nb131_fixH2], xmm13 movsd [rsp + nb131_fiyH2], xmm14 movsd [rsp + nb131_fizH2], xmm15 ;# store back j forces from xmm0-xmm2 movsd [rdi + rax*8], xmm0 movsd [rdi + rax*8 + 8], xmm1 movsd [rdi + rax*8 + 16], xmm2.nb131_updateouterdata: mov ecx, [rsp + nb131_ii3] mov rdi, [rbp + nb131_faction] mov rsi, [rbp + nb131_fshift] mov edx, [rsp + nb131_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movapd xmm0, [rsp + nb131_fixO] movapd xmm1, [rsp + nb131_fiyO] movapd xmm2, [rsp + nb131_fizO] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addsd xmm0, xmm3 addsd xmm1, xmm4 addsd xmm2, xmm5 ;# sum is in low xmm0-xmm2 movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, 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?