nb_kernel211_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,773 行 · 第 1/4 页
S
1,773 行
movaps xmm5, xmm2 movaps xmm6, xmm0 movaps xmm7, xmm1 movaps xmm8, xmm2 subps xmm0, [rsp + nb211_ixO] subps xmm1, [rsp + nb211_iyO] subps xmm2, [rsp + nb211_izO] subps xmm3, [rsp + nb211_ixH1] subps xmm4, [rsp + nb211_iyH1] subps xmm5, [rsp + nb211_izH1] subps xmm6, [rsp + nb211_ixH2] subps xmm7, [rsp + nb211_iyH2] subps xmm8, [rsp + nb211_izH2] movaps [rsp + nb211_dxO], xmm0 movaps [rsp + nb211_dyO], xmm1 movaps [rsp + nb211_dzO], xmm2 mulps xmm0, xmm0 mulps xmm1, xmm1 mulps xmm2, xmm2 movaps [rsp + nb211_dxH1], xmm3 movaps [rsp + nb211_dyH1], xmm4 movaps [rsp + nb211_dzH1], xmm5 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 movaps [rsp + nb211_dxH2], xmm6 movaps [rsp + nb211_dyH2], xmm7 movaps [rsp + nb211_dzH2], 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 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 + nb211_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 + nb211_half] mulps xmm9, xmm4 ;# rinvO mulps xmm10, xmm4 ;# rinvH1 mulps xmm11, xmm4 ;# rinvH2 ;# 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 + nb211_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 xmm12, xmm9 mulps xmm12, xmm12 ;# rinv4 mulps xmm12, xmm9 ;# rinv6 subps xmm2, [rsp + nb211_crf] ;# rinv+krsq-crf subps xmm5, [rsp + nb211_crf] subps xmm8, [rsp + nb211_crf] mulps xmm2, [rsp + nb211_qqO] ;# voul=qq*(rinv+ krsq-crf) mulps xmm5, [rsp + nb211_qqH] ;# voul=qq*(rinv+ krsq-crf) mulps xmm8, [rsp + nb211_qqH] ;# 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 movaps xmm13, xmm12 ;# rinv6 mulps xmm12, xmm12 ;# rinv12 mulps xmm13, [rsp + nb211_c6] mulps xmm12, [rsp + nb211_c12] movaps xmm14, xmm12 subps xmm14, xmm13 mulps xmm1, [rsp + nb211_qqO] ;# (rinv-2*krsq)*qq mulps xmm4, [rsp + nb211_qqH] mulps xmm7, [rsp + nb211_qqH] addps xmm2, [rsp + nb211_vctot] addps xmm5, xmm8 addps xmm2, xmm5 movaps [rsp + nb211_vctot], xmm2 addps xmm14, [rsp + nb211_Vvdwtot] mulps xmm13, [rsp + nb211_six] mulps xmm12, [rsp + nb211_twelve] movaps [rsp + nb211_Vvdwtot], xmm14 subps xmm12, xmm13 ;# LJ fscal addps xmm1, xmm12 mulps xmm1, xmm9 ;# fscal mulps xmm4, xmm10 mulps xmm7, xmm11 mov rdi, [rbp + nb211_faction] ;# move j forces to local temp variables 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 + nb211_dxO] mulps xmm1, [rsp + nb211_dyO] mulps xmm2, [rsp + nb211_dzO] mulps xmm3, [rsp + nb211_dxH1] mulps xmm4, [rsp + nb211_dyH1] mulps xmm5, [rsp + nb211_dzH1] mulps xmm6, [rsp + nb211_dxH2] mulps xmm7, [rsp + nb211_dyH2] mulps xmm8, [rsp + nb211_dzH2] movaps xmm13, xmm0 movaps xmm14, xmm1 addps xmm11, xmm2 addps xmm0, [rsp + nb211_fixO] addps xmm1, [rsp + nb211_fiyO] addps xmm2, [rsp + nb211_fizO] addps xmm13, xmm3 addps xmm14, xmm4 addps xmm11, xmm5 addps xmm3, [rsp + nb211_fixH1] addps xmm4, [rsp + nb211_fiyH1] addps xmm5, [rsp + nb211_fizH1] addps xmm13, xmm6 addps xmm14, xmm7 addps xmm11, xmm8 addps xmm6, [rsp + nb211_fixH2] addps xmm7, [rsp + nb211_fiyH2] addps xmm8, [rsp + nb211_fizH2] movaps [rsp + nb211_fixO], xmm0 movaps [rsp + nb211_fiyO], xmm1 movaps [rsp + nb211_fizO], xmm2 movaps [rsp + nb211_fixH1], xmm3 movaps [rsp + nb211_fiyH1], xmm4 movaps [rsp + nb211_fizH1], xmm5 movaps [rsp + nb211_fixH2], xmm6 movaps [rsp + nb211_fiyH2], xmm7 movaps [rsp + nb211_fizH2], 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 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 + nb211_innerk], 4 jl .nb211_odd_inner jmp .nb211_unroll_loop.nb211_odd_inner: add dword ptr [rsp + nb211_innerk], 4 jnz .nb211_odd_loop jmp .nb211_updateouterdata.nb211_odd_loop: mov rdx, [rsp + nb211_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] add qword ptr [rsp + nb211_innerjjnr], 4 xorps xmm4, xmm4 movss xmm4, [rsp + nb211_iqO] mov rsi, [rbp + nb211_charge] movhps xmm4, [rsp + nb211_iqH] movss xmm3, [rsi + rax*4] ;# charge in xmm3 shufps xmm3, xmm3, 0 mulps xmm3, xmm4 movaps [rsp + nb211_qqO], xmm3 ;# use oxygen qq for storage xorps xmm6, xmm6 mov rsi, [rbp + nb211_type] mov ebx, [rsi + rax*4] mov rsi, [rbp + nb211_vdwparam] shl ebx, 1 add ebx, [rsp + nb211_ntia] movlps xmm6, [rsi + rbx*4] movaps xmm7, xmm6 shufps xmm6, xmm6, 252 ;# 11111100 shufps xmm7, xmm7, 253 ;# 11111101 movaps [rsp + nb211_c6], xmm6 movaps [rsp + nb211_c12], xmm7 mov rsi, [rbp + nb211_pos] lea rax, [rax + rax*2] ;# move j coords to xmm0-xmm2 movss xmm3, [rsi + rax*4] movss xmm4, [rsi + rax*4 + 4] movss xmm5, [rsi + rax*4 + 8] shufps xmm3, xmm3, 0 shufps xmm4, xmm4, 0 shufps xmm5, xmm5, 0 movss xmm0, [rsp + nb211_ixO] movss xmm1, [rsp + nb211_iyO] movss xmm2, [rsp + nb211_izO] movlps xmm6, [rsp + nb211_ixH1] movlps xmm7, [rsp + nb211_ixH2] unpcklps xmm6, xmm7 movlhps xmm0, xmm6 movlps xmm6, [rsp + nb211_iyH1] movlps xmm7, [rsp + nb211_iyH2] unpcklps xmm6, xmm7 movlhps xmm1, xmm6 movlps xmm6, [rsp + nb211_izH1] movlps xmm7, [rsp + nb211_izH2] unpcklps xmm6, xmm7 movlhps xmm2, xmm6 subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 movaps [rsp + nb211_dxO], xmm3 movaps [rsp + nb211_dyO], xmm4 movaps [rsp + nb211_dzO], xmm5 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 addps xmm4, xmm3 addps xmm4, xmm5 ;# rsq in xmm4 movaps xmm0, xmm4 mulps xmm0, [rsp + nb211_krf] movaps [rsp + nb211_krsqO], xmm0 rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [rsp + nb211_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [rsp + nb211_half] subps xmm1, xmm5 ;# 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv ;# a little trick to avoid NaNs: ;# positions 0,2,and 3 are valid, but not 1. ;# If it contains NaN it doesnt help to mult by 0, ;# So we shuffle it and copy pos 0 to pos1! shufps xmm0, xmm0, 224 ;# 11100000 movaps xmm4, xmm0 mulps xmm4, xmm4 ;# xmm4=rinvsq movaps xmm1, xmm4 mulss xmm1, xmm4 mulss xmm1, xmm4 ;# xmm1=rinvsix movaps xmm2, xmm1 mulss xmm2, xmm2 ;# xmm2=rinvtwelve mulps xmm1, [rsp + nb211_c6] mulps xmm2, [rsp + nb211_c12] movaps xmm5, xmm2 subss xmm5, xmm1 ;# Vvdw=Vvdw12-Vvdw6 addps xmm5, [rsp + nb211_Vvdwtot] mulss xmm1, [rsp + nb211_six] mulss xmm2, [rsp + nb211_twelve] subss xmm2, xmm1 movaps xmm1, xmm0 ;# xmm1=rinv movaps xmm3, [rsp + nb211_krsqO] addps xmm0, xmm3 ;# xmm0=rinv+ krsq mulps xmm3, [rsp + nb211_two] subps xmm0, [rsp + nb211_crf] ;# xmm0=rinv+ krsq-crf subps xmm1, xmm3 ;# xmm1=rinv-2*krsq mulps xmm0, [rsp + nb211_qqO] ;# xmm0=vcoul mulps xmm1, [rsp + nb211_qqO] ;# xmm1=coul part of fs addps xmm2, xmm1 ;# total fs mulps xmm4, xmm2 ;# xmm4=total fscal addps xmm0, [rsp + nb211_vctot] movaps [rsp + nb211_vctot], xmm0 movaps xmm0, [rsp + nb211_dxO] movaps xmm1, [rsp + nb211_dyO] movaps xmm2, [rsp + nb211_dzO] movaps [rsp + nb211_Vvdwtot], xmm5 mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# xmm0-xmm2 contains tx-tz (partial force) movss xmm3, [rsp + nb211_fixO] movss xmm4, [rsp + nb211_fiyO] movss xmm5, [rsp + nb211_fizO] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [rsp + nb211_fixO], xmm3 movss [rsp + nb211_fiyO], xmm4 movss [rsp + nb211_fizO], xmm5 ;# updated the O force now do the H's movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 230 ;# 11100110 ;# shift right shufps xmm4, xmm4, 230 ;# 11100110 shufps xmm5, xmm5, 230 ;# 11100110 addss xmm3, [rsp + nb211_fixH1] addss xmm4, [rsp + nb211_fiyH1] addss xmm5, [rsp + nb211_fizH1] movss [rsp + nb211_fixH1], xmm3 movss [rsp + nb211_fiyH1], xmm4 movss [rsp + nb211_fizH1], xmm5 ;# updated the H1 force mov rdi, [rbp + nb211_faction] shufps xmm3, xmm3, 231 ;# 11100111 ;# shift right shufps xmm4, xmm4, 231 ;# 11100111 shufps xmm5, xmm5, 231 ;# 11100111 addss xmm3, [rsp + nb211_fixH2] addss xmm4, [rsp + nb211_fiyH2] addss xmm5, [rsp + nb211_fizH2] movss [rsp + nb211_fixH2], xmm3 movss [rsp + nb211_fiyH2], xmm4 movss [rsp + nb211_fizH2], xmm5 ;# updated the H2 force ;# the fj's - start by accumulating the tx/ty/tz force in xmm0, xmm1 xorps xmm5, xmm5 movaps xmm3, xmm0 movlps xmm6, [rdi + rax*4] movss xmm7, [rdi + rax*4 + 8] unpcklps xmm3, xmm1 movlhps xmm3, xmm5 unpckhps xmm0, xmm1 addps xmm0, xmm3 movhlps xmm3, xmm0 addps xmm0, xmm3 ;# x,y sum in xmm0 movhlps xmm1, xmm2 addss xmm2, xmm1 shufps xmm1, xmm1, 1 addss xmm2, xmm1 ;# z sum in xmm2 addps xmm6, xmm0 addss xmm7, xmm2 movlps [rdi + rax*4], xmm6 movss [rdi + rax*4 + 8], xmm7 dec dword ptr [rsp + nb211_innerk] jz .nb211_updateouterdata jmp .nb211_odd_loop.nb211_updateouterdata: mov ecx, [rsp + nb211_ii3] mov rdi, [rbp + nb211_faction] mov rsi, [rbp + nb211_fshift] mov edx, [rsp + nb211_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb211_fixO] movaps xmm1, [rsp + nb211_fiyO] movaps xmm2, [rsp + nb211_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 ⌨️ 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?