nb_kernel233_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,269 行 · 第 1/5 页
S
2,269 行
movaps xmm1, xmm4 shufps xmm2, xmm6, 136 ;# constant 10001000 shufps xmm0, xmm5, 136 ;# constant 10001000 shufps xmm1, xmm5, 221 ;# constant 11011101 ;# xmm0 = jx ;# xmm1 = jy ;# xmm2 = jz ;# O interaction ;# copy to xmm3-xmm5 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 subps xmm3, [rsp + nb233_ixO] subps xmm4, [rsp + nb233_iyO] subps xmm5, [rsp + nb233_izO] movaps [rsp + nb233_dxO], xmm3 movaps [rsp + nb233_dyO], xmm4 movaps [rsp + nb233_dzO], xmm5 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 addps xmm3, xmm4 addps xmm3, xmm5 ;# xmm3=rsq ;# calculate rinv=1/sqrt(rsq) rsqrtps xmm5, xmm3 movaps xmm15, xmm5 mulps xmm5, xmm5 movaps xmm4, [rsp + nb233_three] mulps xmm5, xmm3 ;# rsq*lu*lu subps xmm4, xmm5 ;# 30-rsq*lu*lu mulps xmm4, xmm15 mulps xmm4, [rsp + nb233_half] movaps xmm15, xmm4 mulps xmm3, xmm4 ;# xmm15=rinv ;# xmm3=r mulps xmm3, [rsp + nb233_tsc] ;# rtab ;# truncate and convert to integers cvttps2dq xmm5, xmm3 ;# convert back to float cvtdq2ps xmm4, xmm5 ;# multiply by 8 pslld xmm5, 3 ;# calculate eps subps xmm3, xmm4 ;# xmm3=eps ;# move to integer registers movhlps xmm6, xmm5 movd r8d, xmm5 movd r10d, xmm6 pshufd xmm5, xmm5, 1 pshufd xmm6, xmm6, 1 movd r9d, xmm5 movd r11d, xmm6 ;# xmm3=eps ;# xmm15=rinv mov rsi, [rbp + nb233_VFtab] ;# calculate LJ table movlps xmm5, [rsi + r8*4] movlps xmm9, [rsi + r8*4 + 16] movlps xmm7, [rsi + r10*4] movlps xmm11, [rsi + r10*4 + 16] movhps xmm5, [rsi + r9*4] movhps xmm9, [rsi + r9*4 + 16] movhps xmm7, [rsi + r11*4] movhps xmm11, [rsi + r11*4 + 16] movaps xmm4, xmm5 movaps xmm8, xmm9 shufps xmm4, xmm7, 136 ;# 10001000 shufps xmm8, xmm11, 136 ;# 10001000 shufps xmm5, xmm7, 221 ;# 11011101 shufps xmm9, xmm11, 221 ;# 11011101 movlps xmm7, [rsi + r8*4 + 8] movlps xmm11, [rsi + r8*4 + 24] movlps xmm13, [rsi + r10*4 + 8] movlps xmm14, [rsi + r10*4 + 24] movhps xmm7, [rsi + r9*4 + 8] movhps xmm11, [rsi + r9*4 + 24] movhps xmm13, [rsi + r11*4 + 8] movhps xmm14, [rsi + r11*4 + 24] movaps xmm6, xmm7 movaps xmm10, xmm11 shufps xmm6, xmm13, 136 ;# 10001000 shufps xmm10, xmm14, 136 ;# 10001000 shufps xmm7, xmm13, 221 ;# 11011101 shufps xmm11, xmm14, 221 ;# 11011101 ;# dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulps xmm7, xmm3 ;# Heps mulps xmm11, xmm3 mulps xmm6, xmm3 ;# Geps mulps xmm10, xmm3 mulps xmm7, xmm3 ;# Heps2 mulps xmm11, xmm3 addps xmm5, xmm6 ;# F+Geps addps xmm9, xmm10 addps xmm5, xmm7 ;# F+Geps+Heps2 = Fp addps xmm9, xmm11 addps xmm7, xmm7 ;# 2*Heps2 addps xmm11, xmm11 addps xmm7, xmm6 ;# 2*Heps2+Geps addps xmm11, xmm10 addps xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps addps xmm11, xmm9 mulps xmm5, xmm3 ;# eps*Fp mulps xmm9, xmm3 movaps xmm12, [rsp + nb233_c6] movaps xmm13, [rsp + nb233_c12] addps xmm5, xmm4 ;# VV addps xmm9, xmm8 mulps xmm5, xmm12 ;# VV*c6 = vnb6 mulps xmm9, xmm13 ;# VV*c12 = vnb12 addps xmm5, xmm9 addps xmm5, [rsp + nb233_Vvdwtot] movaps [rsp + nb233_Vvdwtot], xmm5 mulps xmm7, xmm12 ;# FF*c6 = fnb6 mulps xmm11, xmm13 ;# FF*c12 = fnb12 addps xmm7, xmm11 mulps xmm7, [rsp + nb233_tsc] mulps xmm7, xmm15 ;# -fscal xorps xmm9, xmm9 subps xmm9, xmm7 ;# fscal movaps xmm10, xmm9 movaps xmm11, xmm9 mulps xmm9, [rsp + nb233_dxO] ;# fx/fy/fz mulps xmm10, [rsp + nb233_dyO] mulps xmm11, [rsp + nb233_dzO] ;# save j force temporarily movaps [rsp + nb233_fjx], xmm9 movaps [rsp + nb233_fjy], xmm10 movaps [rsp + nb233_fjz], xmm11 ;# increment i O force addps xmm9, [rsp + nb233_fixO] addps xmm10, [rsp + nb233_fiyO] addps xmm11, [rsp + nb233_fizO] movaps [rsp + nb233_fixO], xmm9 movaps [rsp + nb233_fiyO], xmm10 movaps [rsp + nb233_fizO], xmm11 ;# finished O LJ interaction. ;# do H1, H2, and M interactions in parallel. ;# xmm0-xmm2 still contain j coordinates. movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 movaps xmm6, xmm0 movaps xmm7, xmm1 movaps xmm8, xmm2 subps xmm0, [rsp + nb233_ixH1] subps xmm1, [rsp + nb233_iyH1] subps xmm2, [rsp + nb233_izH1] subps xmm3, [rsp + nb233_ixH2] subps xmm4, [rsp + nb233_iyH2] subps xmm5, [rsp + nb233_izH2] subps xmm6, [rsp + nb233_ixM] subps xmm7, [rsp + nb233_iyM] subps xmm8, [rsp + nb233_izM] movaps [rsp + nb233_dxH1], xmm0 movaps [rsp + nb233_dyH1], xmm1 movaps [rsp + nb233_dzH1], xmm2 mulps xmm0, xmm0 mulps xmm1, xmm1 mulps xmm2, xmm2 movaps [rsp + nb233_dxH2], xmm3 movaps [rsp + nb233_dyH2], xmm4 movaps [rsp + nb233_dzH2], xmm5 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 movaps [rsp + nb233_dxM], xmm6 movaps [rsp + nb233_dyM], xmm7 movaps [rsp + nb233_dzM], xmm8 mulps xmm6, xmm6 mulps xmm7, xmm7 mulps xmm8, xmm8 addps xmm0, xmm1 addps xmm0, xmm2 addps xmm3, xmm4 addps xmm3, xmm5 addps xmm6, xmm7 addps xmm6, xmm8 ;# start doing invsqrt for j atoms rsqrtps xmm1, xmm0 rsqrtps xmm4, xmm3 rsqrtps xmm7, xmm6 movaps xmm2, xmm1 movaps xmm5, xmm4 movaps xmm8, xmm7 mulps xmm1, xmm1 ;# lu*lu mulps xmm4, xmm4 ;# lu*lu mulps xmm7, xmm7 ;# lu*lu movaps xmm9, [rsp + nb233_three] movaps xmm10, xmm9 movaps xmm11, xmm9 mulps xmm1, xmm0 ;# rsq*lu*lu mulps xmm4, xmm3 ;# rsq*lu*lu mulps xmm7, xmm6 ;# rsq*lu*lu subps xmm9, xmm1 subps xmm10, xmm4 subps xmm11, xmm7 ;# 3-rsq*lu*lu mulps xmm9, xmm2 mulps xmm10, xmm5 mulps xmm11, xmm8 ;# lu*(3-rsq*lu*lu) movaps xmm4, [rsp + nb233_half] mulps xmm9, xmm4 ;# rinvH1 mulps xmm10, xmm4 ;# rinvH2 mulps xmm11, xmm4 ;# rinvM ;# interactions ;# rsq in xmm0,xmm3,xmm6 ;# rinv in xmm9, xmm10, xmm11 movaps xmm1, xmm9 ;# copy of rinv movaps xmm4, xmm10 movaps xmm7, xmm11 movaps xmm2, [rsp + nb233_krf] mulps xmm9, xmm9 ;# rinvsq mulps xmm10, xmm10 mulps xmm11, xmm11 mulps xmm0, xmm2 ;# k*rsq mulps xmm3, xmm2 mulps xmm6, xmm2 movaps xmm2, xmm0 ;# copy of k*rsq movaps xmm5, xmm3 movaps xmm8, xmm6 addps xmm2, xmm1 ;# rinv+krsq addps xmm5, xmm4 addps xmm8, xmm7 movaps xmm14, [rsp + nb233_crf] subps xmm2, xmm14 ;# rinv+krsq-crf subps xmm5, xmm14 subps xmm8, xmm14 movaps xmm12, [rsp + nb233_qqH] movaps xmm13, [rsp + nb233_qqM] mulps xmm2, xmm12 ;# voul=qq*(rinv+ krsq-crf) mulps xmm5, xmm12 ;# voul=qq*(rinv+ krsq-crf) mulps xmm8, xmm13 ;# voul=qq*(rinv+ krsq-crf) addps xmm0, xmm0 ;# 2*krsq addps xmm3, xmm3 addps xmm6, xmm6 subps xmm1, xmm0 ;# rinv-2*krsq subps xmm4, xmm3 subps xmm7, xmm6 mulps xmm1, xmm12 ;# (rinv-2*krsq)*qq mulps xmm4, xmm12 mulps xmm7, xmm13 addps xmm2, [rsp + nb233_vctot] addps xmm5, xmm8 addps xmm2, xmm5 movaps xmm15, xmm2 movaps [rsp + nb233_vctot], xmm2 mulps xmm1, xmm9 ;# fscal mulps xmm4, xmm10 mulps xmm7, xmm11 ;# move j forces to local temp variables mov rdi, [rbp + nb233_faction] movlps xmm9, [rdi + rax*4] ;# jxa jya - - movlps xmm10, [rdi + rcx*4] ;# jxc jyc - - movhps xmm9, [rdi + rbx*4] ;# jxa jya jxb jyb movhps xmm10, [rdi + rdx*4] ;# jxc jyc jxd jyd movss xmm11, [rdi + rax*4 + 8] ;# jza - - - movss xmm12, [rdi + rcx*4 + 8] ;# jzc - - - movss xmm6, [rdi + rbx*4 + 8] ;# jzb movss xmm8, [rdi + rdx*4 + 8] ;# jzd movlhps xmm11, xmm6 ;# jza - jzb - movlhps xmm12, xmm8 ;# jzc - jzd - shufps xmm11, xmm12, 136 ;# 10001000 => jza jzb jzc jzd ;# xmm9: jxa jya jxb jyb ;# xmm10: jxc jyc jxd jyd ;# xmm11: jza jzb jzc jzd movaps xmm0, xmm1 movaps xmm2, xmm1 movaps xmm3, xmm4 movaps xmm5, xmm4 movaps xmm6, xmm7 movaps xmm8, xmm7 mulps xmm0, [rsp + nb233_dxH1] mulps xmm1, [rsp + nb233_dyH1] mulps xmm2, [rsp + nb233_dzH1] mulps xmm3, [rsp + nb233_dxH2] mulps xmm4, [rsp + nb233_dyH2] mulps xmm5, [rsp + nb233_dzH2] mulps xmm6, [rsp + nb233_dxM] mulps xmm7, [rsp + nb233_dyM] mulps xmm8, [rsp + nb233_dzM] ;# fetch forces from O interaction movaps xmm13, [rsp + nb233_fjx] movaps xmm14, [rsp + nb233_fjy] addps xmm11, [rsp + nb233_fjz] addps xmm13, xmm0 addps xmm14, xmm1 addps xmm11, xmm2 addps xmm0, [rsp + nb233_fixH1] addps xmm1, [rsp + nb233_fiyH1] addps xmm2, [rsp + nb233_fizH1] addps xmm13, xmm3 addps xmm14, xmm4 addps xmm11, xmm5 addps xmm3, [rsp + nb233_fixH2] addps xmm4, [rsp + nb233_fiyH2] addps xmm5, [rsp + nb233_fizH2] addps xmm13, xmm6 addps xmm14, xmm7 addps xmm11, xmm8 addps xmm6, [rsp + nb233_fixM] addps xmm7, [rsp + nb233_fiyM] addps xmm8, [rsp + nb233_fizM] movaps [rsp + nb233_fixH1], xmm0 movaps [rsp + nb233_fiyH1], xmm1 movaps [rsp + nb233_fizH1], xmm2 movaps [rsp + nb233_fixH2], xmm3 movaps [rsp + nb233_fiyH2], xmm4 movaps [rsp + nb233_fizH2], xmm5 movaps [rsp + nb233_fixM], xmm6 movaps [rsp + nb233_fiyM], xmm7 movaps [rsp + nb233_fizM], xmm8 ;# xmm9 = fjx ;# xmm10 = fjy ;# xmm11 = fjz movaps xmm15, xmm13 unpcklps xmm13, xmm14 unpckhps xmm15, xmm14 addps xmm9, xmm13 addps xmm10, xmm15 movhlps xmm12, xmm11 ;# fjzc fjzd movlps [rdi + rax*4], xmm9 movhps [rdi + rbx*4], xmm9 movlps [rdi + rcx*4], xmm10 movhps [rdi + rdx*4], xmm10 movss [rdi + rax*4 + 8], xmm11 movss [rdi + rcx*4 + 8], xmm12 shufps xmm11, xmm11, 1 shufps xmm12, xmm12, 1 movss [rdi + rbx*4 + 8], xmm11 movss [rdi + rdx*4 + 8], xmm12 ;# should we do one more iteration? sub dword ptr [rsp + nb233_innerk], 4 jl .nb233_odd_inner jmp .nb233_unroll_loop.nb233_odd_inner: add dword ptr [rsp + nb233_innerk], 4 jnz .nb233_odd_loop jmp .nb233_updateouterdata.nb233_odd_loop: mov rdx, [rsp + nb233_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] add qword ptr [rsp + nb233_innerjjnr], 4 xorps xmm4, xmm4 ;# clear reg. movss xmm4, [rsp + nb233_iqM] mov rsi, [rbp + nb233_charge] movhps xmm4, [rsp + nb233_iqH] ;# [qM 0 qH qH] shufps xmm4, xmm4, 41 ;# [0 qH qH qM] movss xmm3, [rsi + rax*4] ;# charge in xmm3 shufps xmm3, xmm3, 0 mulps xmm3, xmm4 movaps [rsp + nb233_qqM], xmm3 ;# use dummy qq for storage xorps xmm6, xmm6 mov rsi, [rbp + nb233_type] mov ebx, [rsi + rax*4] mov rsi, [rbp + nb233_vdwparam] shl ebx, 1 add ebx, [rsp + nb233_ntia] movlps xmm6, [rsi + rbx*4] movaps xmm7, xmm6 shufps xmm6, xmm6, 252 ;# constant 11111100 shufps xmm7, xmm7, 253 ;# constant 11111101 movaps [rsp + nb233_c6], xmm6 movaps [rsp + nb233_c12], xmm7 mov rsi, [rbp + nb233_pos] lea rax, [rax + rax*2] movss xmm0, [rsp + nb233_ixO] movss xmm1, [rsp + nb233_iyO] movss xmm2, [rsp + nb233_izO] movss xmm3, [rsp + nb233_ixH1] movss xmm4, [rsp + nb233_iyH1] movss xmm5, [rsp + nb233_izH1] unpcklps xmm0, [rsp + nb233_ixH2] ;# ixO ixH2 - - unpcklps xmm1, [rsp + nb233_iyH2] ;# iyO iyH2 - - unpcklps xmm2, [rsp + nb233_izH2] ;# izO izH2 - - unpcklps xmm3, [rsp + nb233_ixM] ;# ixH1 ixM - - unpcklps xmm4, [rsp + nb233_iyM] ;# iyH1 iyM - - unpcklps xmm5, [rsp + nb233_izM] ;# izH1 izM - - unpcklps xmm0, xmm3 ;# ixO ixH1 ixH2 ixM unpcklps xmm1, xmm4 ;# same for y unpcklps xmm2, xmm5 ;# same for z ⌨️ 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?