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
;# 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 subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 ;# use O distances for storage 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 xmm4, xmm3 addps xmm4, xmm5 ;# rsq in xmm4 movaps xmm0, xmm4 mulps xmm0, [rsp + nb233_krf] movaps [rsp + nb233_krsqM], xmm0 rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [rsp + nb233_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [rsp + nb233_half] subps xmm1, xmm5 ;# constant 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv, xmm4=rsq #; LJ table interaction mulps xmm4, xmm0 mulps xmm4, [rsp + nb233_tsc] ;# rtab cvttps2pi mm6, xmm4 cvtpi2ps xmm6, mm6 subss xmm4, xmm6 movss xmm1, xmm4 ;# xmm1=eps movss xmm2, xmm1 mulss xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 3 movd mm0, eax mov rsi, [rbp + nb233_VFtab] movd eax, mm6 ;# dispersion movlps xmm5, [rsi + rax*4] movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [rsi + rax*4 + 8] movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# dispersion table ready, in xmm4-xmm7 mulss xmm6, xmm1 ;# xmm6=Geps mulss xmm7, xmm2 ;# xmm7=Heps2 addss xmm5, xmm6 addss xmm5, xmm7 ;# xmm5=Fp mulss xmm7, [rsp + nb233_two] ;# two*Heps2 addss xmm7, xmm6 addss xmm7, xmm5 ;# xmm7=FF mulss xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV movss xmm4, [rsp + nb233_c6] mulss xmm7, xmm4 ;# fijD mulss xmm5, xmm4 ;# Vvdw6 mulss xmm7, [rsp + nb233_tsc] ;# put scalar force on stack Update Vvdwtot directly addss xmm5, [rsp + nb233_Vvdwtot] movss [rsp + nb233_fstmp], xmm7 movss [rsp + nb233_Vvdwtot], xmm5 ;# repulsion movlps xmm5, [rsi + rax*4 + 16] movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [rsi + rax*4 + 24] movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# table ready, in xmm4-xmm7 mulss xmm6, xmm1 ;# xmm6=Geps mulss xmm7, xmm2 ;# xmm7=Heps2 addss xmm5, xmm6 addss xmm5, xmm7 ;# xmm5=Fp mulss xmm7, [rsp + nb233_two] ;# two*Heps2 addss xmm7, xmm6 addss xmm7, xmm5 ;# xmm7=FF mulss xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV movss xmm4, [rsp + nb233_c12] mulss xmm7, xmm4 ;# fijR mulss xmm5, xmm4 ;# Vvdw12 mulss xmm7, [rsp + nb233_tsc] addss xmm7, [rsp + nb233_fstmp] movss [rsp + nb233_fstmp], xmm7 addss xmm5, [rsp + nb233_Vvdwtot] movss [rsp + nb233_Vvdwtot], xmm5 movd eax, mm0 movaps xmm4, xmm0 movaps xmm1, xmm0 movaps xmm5, xmm0 mulps xmm4, xmm4 ;# xmm1=rinv, xmm4=rinvsq movaps xmm3, [rsp + nb233_krsqM] addps xmm5, xmm3 ;# xmm0=rinv+ krsq subps xmm5, [rsp + nb233_crf] ;# xmm0=rinv+ krsq-crf mulps xmm3, [rsp + nb233_two] subps xmm1, xmm3 ;# xmm1=rinv-2*krsq movaps xmm2, xmm5 mulps xmm2, [rsp + nb233_qqM] ;# xmm2=vcoul mulps xmm1, [rsp + nb233_qqM] ;# xmm1=coul part of fs mulps xmm1, xmm0 subss xmm1, [rsp + nb233_fstmp] mulps xmm1, xmm0 movaps xmm4, xmm1 addps xmm2, [rsp + nb233_vctot] movaps [rsp + nb233_vctot], xmm2 movaps xmm0, [rsp + nb233_dxO] movaps xmm1, [rsp + nb233_dyO] movaps xmm2, [rsp + nb233_dzO] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# xmm0-xmm2 now contains tx-tz (partial force) movss xmm3, [rsp + nb233_fixO] movss xmm4, [rsp + nb233_fiyO] movss xmm5, [rsp + nb233_fizO] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [rsp + nb233_fixO], xmm3 movss [rsp + nb233_fiyO], xmm4 movss [rsp + nb233_fizO], xmm5 ;# updated the O force now do the H's movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 0x39 ;# shift right shufps xmm4, xmm4, 0x39 shufps xmm5, xmm5, 0x39 addss xmm3, [rsp + nb233_fixH1] addss xmm4, [rsp + nb233_fiyH1] addss xmm5, [rsp + nb233_fizH1] movss [rsp + nb233_fixH1], xmm3 movss [rsp + nb233_fiyH1], xmm4 movss [rsp + nb233_fizH1], xmm5 ;# updated the H1 force shufps xmm3, xmm3, 0x39 shufps xmm4, xmm4, 0x39 shufps xmm5, xmm5, 0x39 addss xmm3, [rsp + nb233_fixH2] addss xmm4, [rsp + nb233_fiyH2] addss xmm5, [rsp + nb233_fizH2] movss [rsp + nb233_fixH2], xmm3 movss [rsp + nb233_fiyH2], xmm4 movss [rsp + nb233_fizH2], xmm5 ;# updated the H2 force mov rdi, [rbp + nb233_faction] shufps xmm3, xmm3, 0x39 shufps xmm4, xmm4, 0x39 shufps xmm5, xmm5, 0x39 addss xmm3, [rsp + nb233_fixM] addss xmm4, [rsp + nb233_fiyM] addss xmm5, [rsp + nb233_fizM] movss [rsp + nb233_fixM], xmm3 movss [rsp + nb233_fiyM], xmm4 movss [rsp + nb233_fizM], xmm5 ;# updated the M force ;# the fj's - move in from mem start by acc. tx/ty/tz in xmm0, xmm1 movlps xmm6, [rdi + rax*4] movss xmm7, [rdi + rax*4 + 8] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm3, xmm0 addps xmm4, xmm1 addps xmm5, xmm2 movaps xmm0, xmm3 movaps xmm1, xmm4 movaps xmm2, xmm5 shufps xmm3, xmm3, 0x39 ;# shift right shufps xmm4, xmm4, 0x39 shufps xmm5, xmm5, 0x39 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 unpcklps xmm0, xmm1 ;# x,y sum in xmm0, 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 + nb233_innerk] jz .nb233_updateouterdata jmp .nb233_odd_loop.nb233_updateouterdata: mov ecx, [rsp + nb233_ii3] mov rdi, [rbp + nb233_faction] mov rsi, [rbp + nb233_fshift] mov edx, [rsp + nb233_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb233_fixO] movaps xmm1, [rsp + nb233_fiyO] movaps xmm2, [rsp + nb233_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 ;# constant 00001000 ;# accumulate H1i forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb233_fixH1] movaps xmm1, [rsp + nb233_fiyH1] movaps xmm2, [rsp + nb233_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 ;# constant 00001000 addps xmm6, xmm0 ;# accumulate H2i forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb233_fixH2] movaps xmm1, [rsp + nb233_fiyH2] movaps xmm2, [rsp + nb233_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 ;# constant 00001000 addps xmm6, xmm0 ;# accumulate Mi forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb233_fixM] movaps xmm1, [rsp + nb233_fiyM] movaps xmm2, [rsp + nb233_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 ;# constant 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 + nb233_n] ;# get group index for i particle mov rdx, [rbp + nb233_gid] ;# base of gid[] mov edx, [rdx + rsi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [rsp + nb233_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 + nb233_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 + nb233_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 + nb233_Vvdw] addss xmm7, [rax + rdx*4] ;# move back to mem movss [rax + rdx*4], xmm7 ;# finish if last mov ecx, [rsp + nb233_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb233_outerend ;# not last, iterate outer loop once more! mov [rsp + nb233_n], esi jmp .nb233_outer.nb233_outerend: ;# check if more outer neighborlists remain mov ecx, [rsp + nb233_nri] ;# esi already loaded with n above sub ecx, esi jz .nb233_end ;# non-zero, do one more workunit jmp .nb233_threadloop.nb233_end: mov eax, [rsp + nb233_nouter] mov ebx, [rsp + nb233_ninner]⌨️ 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?