nb_kernel213_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,957 行 · 第 1/5 页
S
1,957 行
xorps %xmm4,%xmm4 ## clear reg. movss nb213_iqM(%esp),%xmm4 movl nb213_charge(%ebp),%esi movhps nb213_iqH(%esp),%xmm4 ## [qM 0 qH qH] shufps $41,%xmm4,%xmm4 ## [0 qH qH qM] movss (%esi,%eax,4),%xmm3 ## charge in xmm3 shufps $0,%xmm3,%xmm3 mulps %xmm4,%xmm3 movaps %xmm3,nb213_qqM(%esp) ## use dummy qq for storage xorps %xmm6,%xmm6 movl nb213_type(%ebp),%esi movl (%esi,%eax,4),%ebx movl nb213_vdwparam(%ebp),%esi shll %ebx addl nb213_ntia(%esp),%ebx movlps (%esi,%ebx,4),%xmm6 movaps %xmm6,%xmm7 shufps $252,%xmm6,%xmm6 ## constant 11111100 shufps $253,%xmm7,%xmm7 ## constant 11111101 movaps %xmm6,nb213_c6(%esp) movaps %xmm7,nb213_c12(%esp) movl nb213_pos(%ebp),%esi leal (%eax,%eax,2),%eax movss nb213_ixO(%esp),%xmm3 movss nb213_iyO(%esp),%xmm4 movss nb213_izO(%esp),%xmm5 movss nb213_ixH1(%esp),%xmm0 movss nb213_iyH1(%esp),%xmm1 movss nb213_izH1(%esp),%xmm2 unpcklps nb213_ixH2(%esp),%xmm3 ## ixO ixH2 - - unpcklps nb213_iyH2(%esp),%xmm4 ## iyO iyH2 - - unpcklps nb213_izH2(%esp),%xmm5 ## izO izH2 - - unpcklps nb213_ixM(%esp),%xmm0 ## ixH1 ixM - - unpcklps nb213_iyM(%esp),%xmm1 ## iyH1 iyM - - unpcklps nb213_izM(%esp),%xmm2 ## izH1 izM - - unpcklps %xmm0,%xmm3 ## ixO ixH1 ixH2 ixM unpcklps %xmm1,%xmm4 ## same for y unpcklps %xmm2,%xmm5 ## same for z ## move j coords to xmm0-xmm2 movss (%esi,%eax,4),%xmm0 movss 4(%esi,%eax,4),%xmm1 movss 8(%esi,%eax,4),%xmm2 shufps $0,%xmm0,%xmm0 shufps $0,%xmm1,%xmm1 shufps $0,%xmm2,%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 ## use O distances for storage movaps %xmm3,nb213_dxO(%esp) movaps %xmm4,nb213_dyO(%esp) movaps %xmm5,nb213_dzO(%esp) mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm3,%xmm4 addps %xmm5,%xmm4 ## rsq in xmm4 movaps %xmm4,%xmm0 mulps nb213_krf(%esp),%xmm0 movaps %xmm0,nb213_krsqM(%esp) rsqrtps %xmm4,%xmm5 ## lookup seed in xmm5 movaps %xmm5,%xmm2 mulps %xmm5,%xmm5 movaps nb213_three(%esp),%xmm1 mulps %xmm4,%xmm5 ## rsq*lu*lu movaps nb213_half(%esp),%xmm0 subps %xmm5,%xmm1 ## constant 30-rsq*lu*lu mulps %xmm2,%xmm1 mulps %xmm1,%xmm0 ## xmm0=rinv movaps %xmm0,%xmm4 movaps %xmm0,%xmm1 movaps %xmm0,%xmm5 mulps %xmm4,%xmm4 ## xmm1=rinv, xmm4=rinvsq movaps nb213_krsqM(%esp),%xmm3 addps %xmm3,%xmm5 ## xmm0=rinv+ krsq subps nb213_crf(%esp),%xmm5 ## xmm0=rinv+ krsq-crf mulps nb213_two(%esp),%xmm3 subps %xmm3,%xmm1 ## xmm1=rinv-2*krsq movaps %xmm5,%xmm7 mulps nb213_qqM(%esp),%xmm7 ## xmm0=vcoul mulps nb213_qqM(%esp),%xmm1 ## xmm1=coul part of fs movaps %xmm1,%xmm6 addps nb213_vctot(%esp),%xmm7 movaps %xmm7,nb213_vctot(%esp) movaps %xmm0,%xmm1 mulps %xmm1,%xmm1 movaps %xmm1,%xmm2 mulss %xmm1,%xmm1 mulss %xmm2,%xmm1 ## xmm1=rinvsix xorps %xmm4,%xmm4 movss %xmm1,%xmm4 mulss %xmm4,%xmm4 ## xmm4=rinvtwelve mulss nb213_c6(%esp),%xmm1 mulss nb213_c12(%esp),%xmm4 movaps %xmm4,%xmm3 subss %xmm1,%xmm3 ## xmm3=Vvdw12-Vvdw6 mulss nb213_six(%esp),%xmm1 mulss nb213_twelve(%esp),%xmm4 subss %xmm1,%xmm4 addss nb213_Vvdwtot(%esp),%xmm3 movss %xmm3,nb213_Vvdwtot(%esp) addps %xmm6,%xmm4 mulps %xmm0,%xmm4 mulps %xmm0,%xmm4 ## fscal movaps nb213_dxO(%esp),%xmm0 movaps nb213_dyO(%esp),%xmm1 movaps nb213_dzO(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 now contains tx-tz (partial force) movss nb213_fixO(%esp),%xmm3 movss nb213_fiyO(%esp),%xmm4 movss nb213_fizO(%esp),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,nb213_fixO(%esp) movss %xmm4,nb213_fiyO(%esp) movss %xmm5,nb213_fizO(%esp) ## updated the O force now do the H's movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $0x39,%xmm3,%xmm3 ## shift right shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss nb213_fixH1(%esp),%xmm3 addss nb213_fiyH1(%esp),%xmm4 addss nb213_fizH1(%esp),%xmm5 movss %xmm3,nb213_fixH1(%esp) movss %xmm4,nb213_fiyH1(%esp) movss %xmm5,nb213_fizH1(%esp) ## updated the H1 force shufps $0x39,%xmm3,%xmm3 shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss nb213_fixH2(%esp),%xmm3 addss nb213_fiyH2(%esp),%xmm4 addss nb213_fizH2(%esp),%xmm5 movss %xmm3,nb213_fixH2(%esp) movss %xmm4,nb213_fiyH2(%esp) movss %xmm5,nb213_fizH2(%esp) ## updated the H2 force movl nb213_faction(%ebp),%edi shufps $0x39,%xmm3,%xmm3 shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss nb213_fixM(%esp),%xmm3 addss nb213_fiyM(%esp),%xmm4 addss nb213_fizM(%esp),%xmm5 movss %xmm3,nb213_fixM(%esp) movss %xmm4,nb213_fiyM(%esp) movss %xmm5,nb213_fizM(%esp) ## updated the M force ## the fj's - move in from mem start by acc. tx/ty/tz in xmm0, xmm1 movlps (%edi,%eax,4),%xmm6 movss 8(%edi,%eax,4),%xmm7 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,%xmm0 movaps %xmm4,%xmm1 movaps %xmm5,%xmm2 shufps $0x39,%xmm3,%xmm3 ## shift right shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 unpcklps %xmm1,%xmm0 ## x,y sum in xmm0, z sum in xmm2 subps %xmm0,%xmm6 subss %xmm2,%xmm7 movlps %xmm6,(%edi,%eax,4) movss %xmm7,8(%edi,%eax,4) decl nb213_innerk(%esp) jz _nb_kernel213_ia32_sse.nb213_updateouterdata jmp _nb_kernel213_ia32_sse.nb213_odd_loop_nb_kernel213_ia32_sse.nb213_updateouterdata: movl nb213_ii3(%esp),%ecx movl nb213_faction(%ebp),%edi movl nb213_fshift(%ebp),%esi movl nb213_is3(%esp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb213_fixO(%esp),%xmm0 movaps nb213_fiyO(%esp),%xmm1 movaps nb213_fizO(%esp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss (%edi,%ecx,4),%xmm3 movss 4(%edi,%ecx,4),%xmm4 movss 8(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,(%edi,%ecx,4) movss %xmm4,4(%edi,%ecx,4) movss %xmm5,8(%edi,%ecx,4) ## accumulate force in xmm6/xmm7 for fshift movaps %xmm0,%xmm6 movss %xmm2,%xmm7 movlhps %xmm1,%xmm6 shufps $8,%xmm6,%xmm6 ## constant 00001000 ## accumulate H1i forces in xmm0, xmm1, xmm2 movaps nb213_fixH1(%esp),%xmm0 movaps nb213_fiyH1(%esp),%xmm1 movaps nb213_fizH1(%esp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss 12(%edi,%ecx,4),%xmm3 movss 16(%edi,%ecx,4),%xmm4 movss 20(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,12(%edi,%ecx,4) movss %xmm4,16(%edi,%ecx,4) movss %xmm5,20(%edi,%ecx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## constant 00001000 addps %xmm0,%xmm6 ## accumulate H2i forces in xmm0, xmm1, xmm2 movaps nb213_fixH2(%esp),%xmm0 movaps nb213_fiyH2(%esp),%xmm1 movaps nb213_fizH2(%esp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss 24(%edi,%ecx,4),%xmm3 movss 28(%edi,%ecx,4),%xmm4 movss 32(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,24(%edi,%ecx,4) movss %xmm4,28(%edi,%ecx,4) movss %xmm5,32(%edi,%ecx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## constant 00001000 addps %xmm0,%xmm6 ## accumulate Mi forces in xmm0, xmm1, xmm2 movaps nb213_fixM(%esp),%xmm0 movaps nb213_fiyM(%esp),%xmm1 movaps nb213_fizM(%esp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss 36(%edi,%ecx,4),%xmm3 movss 40(%edi,%ecx,4),%xmm4 movss 44(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,36(%edi,%ecx,4) movss %xmm4,40(%edi,%ecx,4) movss %xmm5,44(%edi,%ecx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## constant 00001000 addps %xmm0,%xmm6 ## increment fshift force movlps (%esi,%edx,4),%xmm3 movss 8(%esi,%edx,4),%xmm4 addps %xmm6,%xmm3 addss %xmm7,%xmm4 movlps %xmm3,(%esi,%edx,4) movss %xmm4,8(%esi,%edx,4) ## get n from stack movl nb213_n(%esp),%esi ## get group index for i particle movl nb213_gid(%ebp),%edx ## base of gid[] movl (%edx,%esi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb213_vctot(%esp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addps %xmm6,%xmm7 ## pos 0-1 in xmm7 have the sum now movaps %xmm7,%xmm6 shufps $1,%xmm6,%xmm6 addss %xmm6,%xmm7⌨️ 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?