nb_kernel114_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,204 行 · 第 1/5 页
S
2,204 行
movaps [rsp + nb114_fjzO], xmm5 addss xmm0, [rsp + nb114_fixO] addss xmm1, [rsp + nb114_fiyO] addss xmm2, [rsp + nb114_fizO] movss [rsp + nb114_fixO], xmm0 movss [rsp + nb114_fiyO], xmm1 movss [rsp + nb114_fizO], xmm2 ;# do M coulomb interaction movaps xmm0, [rsp + nb114_rinvMM] movaps xmm4, xmm0 ;# xmm4=rinv mulps xmm0, xmm0 ;# xmm0=rinvsq ;# fetch charges to xmm3 (temporary) xorps xmm3, xmm3 movss xmm3, [rsp + nb114_qqMH] movhps xmm3, [rsp + nb114_qqMM] shufps xmm3, xmm3, 193 ;# 11000001 mulps xmm4, xmm3 ;# xmm4=voul mulps xmm0, xmm4 addps xmm4, [rsp + nb114_vctot] movaps [rsp + nb114_vctot], xmm4 movaps xmm1, xmm0 movaps xmm2, xmm0 mulps xmm0, [rsp + nb114_dxMM] mulps xmm1, [rsp + nb114_dyMM] mulps xmm2, [rsp + nb114_dzMM] ;# update forces M - j water movaps xmm3, [rsp + nb114_fjxO] movaps xmm4, [rsp + nb114_fjyO] movaps xmm5, [rsp + nb114_fjzO] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm5, xmm2 movaps [rsp + nb114_fjxO], xmm3 movaps [rsp + nb114_fjyO], xmm4 movaps [rsp + nb114_fjzO], xmm5 addps xmm0, [rsp + nb114_fixM] addps xmm1, [rsp + nb114_fiyM] addps xmm2, [rsp + nb114_fizM] movaps [rsp + nb114_fixM], xmm0 movaps [rsp + nb114_fiyM], xmm1 movaps [rsp + nb114_fizM], xmm2 ;# i H1 & H2 simultaneously first get i particle coords: movaps xmm0, [rsp + nb114_jxO] movaps xmm1, [rsp + nb114_jyO] movaps xmm2, [rsp + nb114_jzO] movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 subps xmm0, [rsp + nb114_ixH1] subps xmm1, [rsp + nb114_iyH1] subps xmm2, [rsp + nb114_izH1] subps xmm3, [rsp + nb114_ixH2] subps xmm4, [rsp + nb114_iyH2] subps xmm5, [rsp + nb114_izH2] movaps [rsp + nb114_dxH1H1], xmm0 movaps [rsp + nb114_dyH1H1], xmm1 movaps [rsp + nb114_dzH1H1], xmm2 movaps [rsp + nb114_dxH2H2], xmm3 movaps [rsp + nb114_dyH2H2], xmm4 movaps [rsp + nb114_dzH2H2], xmm5 mulps xmm0, xmm0 mulps xmm1, xmm1 mulps xmm2, xmm2 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 addps xmm0, xmm1 addps xmm4, xmm3 addps xmm0, xmm2 ;# have rsqH1 in xmm0 addps xmm4, xmm5 ;# have rsqH2 in xmm4 movaps [rsp + nb114_rsqH1H1], xmm0 movaps [rsp + nb114_rsqH2H2], xmm4 ;# start doing invsqrt use rsq values in xmm0, xmm4 rsqrtps xmm1, xmm0 rsqrtps xmm5, xmm4 movaps xmm2, xmm1 movaps xmm6, xmm5 mulps xmm1, xmm1 mulps xmm5, xmm5 movaps xmm3, [rsp + nb114_three] movaps xmm7, xmm3 mulps xmm1, xmm0 mulps xmm5, xmm4 subps xmm3, xmm1 subps xmm7, xmm5 mulps xmm3, xmm2 mulps xmm7, xmm6 mulps xmm3, [rsp + nb114_half] ;# rinvH1H1 mulps xmm7, [rsp + nb114_half] ;# rinvH2H2 movaps [rsp + nb114_rinvH1H1], xmm3 movaps [rsp + nb114_rinvH2H2], xmm7 ;# Do H1 coulomb interaction movaps xmm0, [rsp + nb114_rinvH1H1] movaps xmm4, xmm0 ;# xmm4=rinv mulps xmm0, xmm0 ;# xmm0=rinvsq ;# fetch charges to xmm3 (temporary) xorps xmm3, xmm3 movss xmm3, [rsp + nb114_qqHH] movhps xmm3, [rsp + nb114_qqMH] shufps xmm3, xmm3, 193 ;# 11000001 mulps xmm4, xmm3 ;# xmm4=voul mulps xmm0, xmm4 addps xmm4, [rsp + nb114_vctot] movaps [rsp + nb114_vctot], xmm4 movaps xmm1, xmm0 movaps xmm2, xmm0 mulps xmm0, [rsp + nb114_dxH1H1] mulps xmm1, [rsp + nb114_dyH1H1] mulps xmm2, [rsp + nb114_dzH1H1] ;# update forces H1 - j water movaps xmm3, [rsp + nb114_fjxO] movaps xmm4, [rsp + nb114_fjyO] movaps xmm5, [rsp + nb114_fjzO] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm5, xmm2 movaps [rsp + nb114_fjxO], xmm3 movaps [rsp + nb114_fjyO], xmm4 movaps [rsp + nb114_fjzO], xmm5 addps xmm0, [rsp + nb114_fixH1] addps xmm1, [rsp + nb114_fiyH1] addps xmm2, [rsp + nb114_fizH1] movaps [rsp + nb114_fixH1], xmm0 movaps [rsp + nb114_fiyH1], xmm1 movaps [rsp + nb114_fizH1], xmm2 ;# H2 Coulomb movaps xmm0, [rsp + nb114_rinvH2H2] movaps xmm4, xmm0 ;# xmm4=rinv mulps xmm0, xmm0 ;# xmm0=rinvsq ;# fetch charges to xmm3 (temporary) xorps xmm3, xmm3 movss xmm3, [rsp + nb114_qqHH] movhps xmm3, [rsp + nb114_qqMH] shufps xmm3, xmm3, 193 ;# 11000001 mulps xmm4, xmm3 ;# xmm4=voul mulps xmm0, xmm4 addps xmm4, [rsp + nb114_vctot] ;# local vctot summation variable movaps [rsp + nb114_vctot], xmm4 movaps xmm1, xmm0 movaps xmm2, xmm0 mulps xmm0, [rsp + nb114_dxH2H2] mulps xmm1, [rsp + nb114_dyH2H2] mulps xmm2, [rsp + nb114_dzH2H2] ;# update forces H2 - j water movaps xmm3, [rsp + nb114_fjxO] movaps xmm4, [rsp + nb114_fjyO] movaps xmm5, [rsp + nb114_fjzO] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm5, xmm2 movaps [rsp + nb114_fjxO], xmm3 movaps [rsp + nb114_fjyO], xmm4 movaps [rsp + nb114_fjzO], xmm5 addps xmm0, [rsp + nb114_fixH2] addps xmm1, [rsp + nb114_fiyH2] addps xmm2, [rsp + nb114_fizH2] movaps [rsp + nb114_fixH2], xmm0 movaps [rsp + nb114_fiyH2], xmm1 movaps [rsp + nb114_fizH2], xmm2 mov rsi, [rbp + nb114_faction] ;# update j water forces from local variables. ;# transpose back first movaps xmm0, [rsp + nb114_fjxO] ;# Ox H1x H2x Mx movaps xmm1, [rsp + nb114_fjyO] ;# Oy H1y H2y My movaps xmm2, [rsp + nb114_fjzO] ;# Oz H1z H2z Mz movaps xmm3, xmm0 movaps xmm4, xmm0 unpcklps xmm3, xmm1 ;# Ox Oy - - shufps xmm4, xmm2, 0x1 ;# h1x - Oz - movaps xmm5, xmm1 movaps xmm6, xmm0 unpcklps xmm5, xmm2 ;# - - H1y H1z unpckhps xmm6, xmm1 ;# h2x h2y - - unpckhps xmm1, xmm2 ;# - - My Mz shufps xmm2, xmm0, 0x32 ;# (00110010) h2z - Mx - shufps xmm3, xmm4, 36 ;# 00100100 ;# Ox Oy Oz H1x shufps xmm5, xmm6, 78 ;# 01001110 ;# h1y h1z h2x h2y shufps xmm2, xmm1, 232 ;# 11101000 ;# h2z mx my mz movlps xmm0, [rsi + rax*4] movlps xmm1, [rsi + rax*4 + 16] movlps xmm4, [rsi + rax*4 + 32] movhps xmm0, [rsi + rax*4 + 8] movhps xmm1, [rsi + rax*4 + 24] movhps xmm4, [rsi + rax*4 + 40] addps xmm0, xmm3 addps xmm1, xmm5 addps xmm4, xmm2 movlps [rsi + rax*4], xmm0 movlps [rsi + rax*4 + 16], xmm1 movlps [rsi + rax*4 + 32], xmm4 movhps [rsi + rax*4 + 8], xmm0 movhps [rsi + rax*4 + 24], xmm1 movhps [rsi + rax*4 + 40], xmm4 dec dword ptr [rsp + nb114_innerk] jz .nb114_updateouterdata jmp .nb114_single_loop.nb114_updateouterdata: mov ecx, [rsp + nb114_ii3] mov rdi, [rbp + nb114_faction] mov rsi, [rbp + nb114_fshift] mov edx, [rsp + nb114_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb114_fixO] movaps xmm1, [rsp + nb114_fiyO] movaps xmm2, [rsp + nb114_fizO] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [rdi + rcx*4] movss xmm4, [rdi + rcx*4 + 4] movss xmm5, [rdi + rcx*4 + 8] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [rdi + rcx*4], xmm3 movss [rdi + rcx*4 + 4], xmm4 movss [rdi + rcx*4 + 8], xmm5 ;# accumulate force in xmm6/xmm7 for fshift movaps xmm6, xmm0 movss xmm7, xmm2 movlhps xmm6, xmm1 shufps xmm6, xmm6, 8 ;# 00001000 ;# accumulate H1i forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb114_fixH1] movaps xmm1, [rsp + nb114_fiyH1] movaps xmm2, [rsp + nb114_fizH1] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [rdi + rcx*4 + 12] movss xmm4, [rdi + rcx*4 + 16] movss xmm5, [rdi + rcx*4 + 20] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [rdi + rcx*4 + 12], xmm3 movss [rdi + rcx*4 + 16], xmm4 movss [rdi + rcx*4 + 20], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addss xmm7, xmm2 movlhps xmm0, xmm1 shufps xmm0, xmm0, 8 ;# 00001000 addps xmm6, xmm0 ;# accumulate H2i forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb114_fixH2] movaps xmm1, [rsp + nb114_fiyH2] movaps xmm2, [rsp + nb114_fizH2] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [rdi + rcx*4 + 24] movss xmm4, [rdi + rcx*4 + 28] movss xmm5, [rdi + rcx*4 + 32] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [rdi + rcx*4 + 24], xmm3 movss [rdi + rcx*4 + 28], xmm4 movss [rdi + rcx*4 + 32], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addss xmm7, xmm2 movlhps xmm0, xmm1 shufps xmm0, xmm0, 8 ;# 00001000 addps xmm6, xmm0 ;# accumulate Mi forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb114_fixM] movaps xmm1, [rsp + nb114_fiyM] movaps xmm2, [rsp + nb114_fizM] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [rdi + rcx*4 + 36] movss xmm4, [rdi + rcx*4 + 40] movss xmm5, [rdi + rcx*4 + 44] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [rdi + rcx*4 + 36], xmm3 movss [rdi + rcx*4 + 40], xmm4 movss [rdi + rcx*4 + 44], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addss xmm7, xmm2 movlhps xmm0, xmm1 shufps xmm0, xmm0, 8 ;# 00001000 addps xmm6, xmm0 ;# increment fshift force movlps xmm3, [rsi + rdx*4] movss xmm4, [rsi + rdx*4 + 8] subps xmm3, xmm6 subss xmm4, xmm7 movlps [rsi + rdx*4], xmm3 movss [rsi + rdx*4 + 8], xmm4 ;# get n from stack mov esi, [rsp + nb114_n] ;# get group index for i particle mov rdx, [rbp + nb114_gid] ;# base of gid[] mov edx, [rdx + rsi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [rsp + nb114_vctot] ;# accumulate movhlps xmm6, xmm7 addps xmm7, xmm6 ;# pos 0-1 in xmm7 have the sum now movaps xmm6, xmm7 shufps xmm6, xmm6, 1 addss xmm7, xmm6 ;# add earlier value from mem mov rax, [rbp + nb114_Vc] addss xmm7, [rax + rdx*4] ;# move back to mem movss [rax + rdx*4], xmm7 ;# accumulate total lj energy and update it movaps xmm7, [rsp + nb114_Vvdwtot] ;# accumulate movhlps xmm6, xmm7 addps xmm7, xmm6 ;# pos 0-1 in xmm7 have the sum now movaps xmm6, xmm7 shufps xmm6, xmm6, 1 addss xmm7, xmm6 ;# add earlier value from mem mov rax, [rbp + nb114_Vvdw] addss xmm7, [rax + rdx*4] ;# move back to mem movss [rax + rdx*4], xmm7 ;# finish if last mov ecx, [rsp + nb114_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb114_outerend ;# not last, iterate outer loop once more! mov [rsp + nb114_n], esi jmp .nb114_outer.nb114_outerend: ;# check if more outer neighborlists remain⌨️ 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?