nb_kernel212_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,955 行 · 第 1/5 页
S
1,955 行
jnz _nb_kernel212_ia32_sse.nb212_single_loop jmp _nb_kernel212_ia32_sse.nb212_updateouterdata_nb_kernel212_ia32_sse.nb212_single_loop: movl nb212_innerjjnr(%esp),%edx ## pointer to jjnr[k] movl (%edx),%eax addl $4,nb212_innerjjnr(%esp) movl nb212_pos(%ebp),%esi leal (%eax,%eax,2),%eax ## fetch j coordinates xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 movss (%esi,%eax,4),%xmm3 ## jxO - - - movss 4(%esi,%eax,4),%xmm4 ## jyO - - - movss 8(%esi,%eax,4),%xmm5 ## jzO - - - movlps 12(%esi,%eax,4),%xmm6 ## xmm6 = jxH1 jyH1 - - movss 20(%esi,%eax,4),%xmm7 ## xmm7 = jzH1 - - - movhps 24(%esi,%eax,4),%xmm6 ## xmm6 = jxH1 jyH1 jxH2 jyH2 movss 32(%esi,%eax,4),%xmm2 ## xmm2 = jzH2 - - - ## have all coords, time for some shuffling. shufps $216,%xmm6,%xmm6 ## constant 11011000 ;# xmm6 = jxH1 jxH2 jyH1 jyH2 unpcklps %xmm2,%xmm7 ## xmm7 = jzH1 jzH2 - - movaps nb212_ixO(%esp),%xmm0 movaps nb212_iyO(%esp),%xmm1 movaps nb212_izO(%esp),%xmm2 movlhps %xmm6,%xmm3 ## xmm3 = jxO 0 jxH1 jxH2 shufps $228,%xmm6,%xmm4 ## constant 11100100 ;# xmm4 = jyO 0 jyH1 jyH2 shufps $68,%xmm7,%xmm5 ## constant 01000100 ;# xmm5 = jzO 0 jzH1 jzH2 ## store all j coordinates in jO movaps %xmm3,nb212_jxO(%esp) movaps %xmm4,nb212_jyO(%esp) movaps %xmm5,nb212_jzO(%esp) subps %xmm3,%xmm0 subps %xmm4,%xmm1 subps %xmm5,%xmm2 movaps %xmm0,nb212_dxOO(%esp) movaps %xmm1,nb212_dyOO(%esp) movaps %xmm2,nb212_dzOO(%esp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 addps %xmm1,%xmm0 addps %xmm2,%xmm0 ## have rsq in xmm0 movaps %xmm0,%xmm6 ## do invsqrt rsqrtps %xmm0,%xmm1 mulps nb212_krf(%esp),%xmm6 ## xmm6=krsq movaps %xmm1,%xmm2 movaps %xmm6,%xmm7 mulps %xmm1,%xmm1 movaps nb212_three(%esp),%xmm3 mulps %xmm0,%xmm1 subps %xmm1,%xmm3 mulps %xmm2,%xmm3 mulps nb212_half(%esp),%xmm3 ## rinv iO - j water addps %xmm3,%xmm6 ## xmm6=rinv+ krsq mulps nb212_two(%esp),%xmm7 subps nb212_crf(%esp),%xmm6 ## xmm6=rinv+ krsq-crf xorps %xmm1,%xmm1 movaps %xmm3,%xmm0 subps %xmm7,%xmm3 ## xmm3=rinv-2*krsq xorps %xmm4,%xmm4 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm4 (temporary) movss nb212_qqOO(%esp),%xmm4 movss %xmm0,%xmm1 movhps nb212_qqOH(%esp),%xmm4 mulss %xmm0,%xmm1 mulps %xmm4,%xmm6 ## vcoul mulps %xmm4,%xmm3 ## coul part of fs mulss %xmm0,%xmm1 ## xmm1(0)=rinvsix movaps %xmm1,%xmm2 ## zero everything else in xmm2 mulss %xmm2,%xmm2 ## xmm2=rinvtwelve mulss nb212_c6(%esp),%xmm1 mulss nb212_c12(%esp),%xmm2 movaps %xmm2,%xmm4 subss %xmm1,%xmm4 ## Vvdwtot=Vvdw12-Vvdw6 addps nb212_Vvdwtot(%esp),%xmm4 mulss nb212_six(%esp),%xmm1 mulss nb212_twelve(%esp),%xmm2 movaps %xmm4,nb212_Vvdwtot(%esp) subss %xmm1,%xmm2 ## fsD+ fsR addps %xmm3,%xmm2 ## fsC+ fsD+ fsR addps nb212_vctot(%esp),%xmm6 mulps %xmm2,%xmm0 ## total fscal movaps %xmm6,nb212_vctot(%esp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb212_dxOO(%esp),%xmm0 mulps nb212_dyOO(%esp),%xmm1 mulps nb212_dzOO(%esp),%xmm2 ## initial update for j forces xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb212_fjxO(%esp) movaps %xmm4,nb212_fjyO(%esp) movaps %xmm5,nb212_fjzO(%esp) addps nb212_fixO(%esp),%xmm0 addps nb212_fiyO(%esp),%xmm1 addps nb212_fizO(%esp),%xmm2 movaps %xmm0,nb212_fixO(%esp) movaps %xmm1,nb212_fiyO(%esp) movaps %xmm2,nb212_fizO(%esp) ## done with i O Now do i H1 & H2 simultaneously first get i particle coords: movaps nb212_ixH1(%esp),%xmm0 movaps nb212_iyH1(%esp),%xmm1 movaps nb212_izH1(%esp),%xmm2 movaps nb212_ixH2(%esp),%xmm3 movaps nb212_iyH2(%esp),%xmm4 movaps nb212_izH2(%esp),%xmm5 subps nb212_jxO(%esp),%xmm0 subps nb212_jyO(%esp),%xmm1 subps nb212_jzO(%esp),%xmm2 subps nb212_jxO(%esp),%xmm3 subps nb212_jyO(%esp),%xmm4 subps nb212_jzO(%esp),%xmm5 movaps %xmm0,nb212_dxH1O(%esp) movaps %xmm1,nb212_dyH1O(%esp) movaps %xmm2,nb212_dzH1O(%esp) movaps %xmm3,nb212_dxH2O(%esp) movaps %xmm4,nb212_dyH2O(%esp) movaps %xmm5,nb212_dzH2O(%esp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm1,%xmm0 addps %xmm3,%xmm4 addps %xmm2,%xmm0 ## have rsqH1 in xmm0 addps %xmm5,%xmm4 ## have rsqH2 in xmm4 ## do invsqrt rsqrtps %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movaps %xmm1,%xmm2 movaps %xmm5,%xmm6 mulps %xmm1,%xmm1 mulps %xmm5,%xmm5 movaps nb212_three(%esp),%xmm3 movaps %xmm3,%xmm7 mulps %xmm0,%xmm1 mulps %xmm4,%xmm5 subps %xmm1,%xmm3 subps %xmm5,%xmm7 mulps %xmm2,%xmm3 mulps %xmm6,%xmm7 mulps nb212_half(%esp),%xmm3 ## rinv H1 - j water mulps nb212_half(%esp),%xmm7 ## rinv H2 - j water mulps nb212_krf(%esp),%xmm0 ## krsq mulps nb212_krf(%esp),%xmm4 ## krsq ## assemble charges in xmm6 xorps %xmm6,%xmm6 movss nb212_qqOH(%esp),%xmm6 movhps nb212_qqHH(%esp),%xmm6 movaps %xmm0,%xmm1 movaps %xmm4,%xmm5 addps %xmm3,%xmm0 ## krsq+ rinv addps %xmm7,%xmm4 ## krsq+ rinv subps nb212_crf(%esp),%xmm0 subps nb212_crf(%esp),%xmm4 mulps nb212_two(%esp),%xmm1 mulps nb212_two(%esp),%xmm5 mulps %xmm6,%xmm0 ## vcoul mulps %xmm6,%xmm4 ## vcoul addps %xmm0,%xmm4 addps nb212_vctot(%esp),%xmm4 movaps %xmm4,nb212_vctot(%esp) movaps %xmm3,%xmm0 movaps %xmm7,%xmm4 mulps %xmm3,%xmm3 mulps %xmm7,%xmm7 subps %xmm1,%xmm0 subps %xmm5,%xmm4 mulps %xmm6,%xmm0 mulps %xmm6,%xmm4 mulps %xmm3,%xmm0 ## fscal mulps %xmm4,%xmm7 ## fscal movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb212_dxH1O(%esp),%xmm0 mulps nb212_dyH1O(%esp),%xmm1 mulps nb212_dzH1O(%esp),%xmm2 ## update forces H1 - j water movaps nb212_fjxO(%esp),%xmm3 movaps nb212_fjyO(%esp),%xmm4 movaps nb212_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb212_fjxO(%esp) movaps %xmm4,nb212_fjyO(%esp) movaps %xmm5,nb212_fjzO(%esp) addps nb212_fixH1(%esp),%xmm0 addps nb212_fiyH1(%esp),%xmm1 addps nb212_fizH1(%esp),%xmm2 movaps %xmm0,nb212_fixH1(%esp) movaps %xmm1,nb212_fiyH1(%esp) movaps %xmm2,nb212_fizH1(%esp) ## do forces H2 - j water movaps %xmm7,%xmm0 movaps %xmm7,%xmm1 movaps %xmm7,%xmm2 mulps nb212_dxH2O(%esp),%xmm0 mulps nb212_dyH2O(%esp),%xmm1 mulps nb212_dzH2O(%esp),%xmm2 movaps nb212_fjxO(%esp),%xmm3 movaps nb212_fjyO(%esp),%xmm4 movaps nb212_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movl nb212_faction(%ebp),%esi movaps %xmm3,nb212_fjxO(%esp) movaps %xmm4,nb212_fjyO(%esp) movaps %xmm5,nb212_fjzO(%esp) addps nb212_fixH2(%esp),%xmm0 addps nb212_fiyH2(%esp),%xmm1 addps nb212_fizH2(%esp),%xmm2 movaps %xmm0,nb212_fixH2(%esp) movaps %xmm1,nb212_fiyH2(%esp) movaps %xmm2,nb212_fizH2(%esp) ## update j water forces from local variables movlps (%esi,%eax,4),%xmm0 movlps 12(%esi,%eax,4),%xmm1 movhps 24(%esi,%eax,4),%xmm1 movaps nb212_fjxO(%esp),%xmm3 movaps nb212_fjyO(%esp),%xmm4 movaps nb212_fjzO(%esp),%xmm5 movaps %xmm5,%xmm6 movaps %xmm5,%xmm7 shufps $2,%xmm6,%xmm6 ## constant 00000010 shufps $3,%xmm7,%xmm7 ## constant 00000011 addss 8(%esi,%eax,4),%xmm5 addss 20(%esi,%eax,4),%xmm6 addss 32(%esi,%eax,4),%xmm7 movss %xmm5,8(%esi,%eax,4) movss %xmm6,20(%esi,%eax,4) movss %xmm7,32(%esi,%eax,4) movaps %xmm3,%xmm5 unpcklps %xmm4,%xmm3 unpckhps %xmm4,%xmm5 addps %xmm3,%xmm0 addps %xmm5,%xmm1 movlps %xmm0,(%esi,%eax,4) movlps %xmm1,12(%esi,%eax,4) movhps %xmm1,24(%esi,%eax,4) decl nb212_innerk(%esp) jz _nb_kernel212_ia32_sse.nb212_updateouterdata jmp _nb_kernel212_ia32_sse.nb212_single_loop_nb_kernel212_ia32_sse.nb212_updateouterdata: movl nb212_ii3(%esp),%ecx movl nb212_faction(%ebp),%edi movl nb212_fshift(%ebp),%esi movl nb212_is3(%esp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb212_fixO(%esp),%xmm0 movaps nb212_fiyO(%esp),%xmm1 movaps nb212_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 nb212_fixH1(%esp),%xmm0 movaps nb212_fiyH1(%esp),%xmm1 movaps nb212_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 nb212_fixH2(%esp),%xmm0 movaps nb212_fiyH2(%esp),%xmm1 movaps nb212_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⌨️ 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?