nb_kernel232_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,002 行 · 第 1/5 页
S
2,002 行
movaps nb232_rinvOO(%esp),%xmm0 movaps %xmm0,%xmm2 ## xmm7=rinv movaps nb232_krf(%esp),%xmm5 mulps nb232_rsqOO(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm6 addps %xmm2,%xmm6 ## xmm6=rinv+ krsq subps nb232_crf(%esp),%xmm6 mulps nb232_qqOO(%esp),%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps nb232_two(%esp),%xmm5 subps %xmm5,%xmm2 ## xmm2=rinv-2*krsq mulps nb232_qqOO(%esp),%xmm2 ## xmm2 = coul part of fscal addps nb232_vctot(%esp),%xmm6 ## local vctot summation variable mulps %xmm0,%xmm2 subps %xmm7,%xmm2 mulps %xmm0,%xmm2 movaps %xmm2,%xmm0 movaps %xmm2,%xmm1 movd %mm0,%eax movd %mm1,%ebx movd %mm2,%ecx movd %mm3,%edx xorps %xmm3,%xmm3 movaps %xmm3,%xmm4 movaps %xmm3,%xmm5 mulps nb232_dxOO(%esp),%xmm0 mulps nb232_dyOO(%esp),%xmm1 mulps nb232_dzOO(%esp),%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 addps nb232_fixO(%esp),%xmm0 addps nb232_fiyO(%esp),%xmm1 addps nb232_fizO(%esp),%xmm2 movaps %xmm3,nb232_fjxO(%esp) movaps %xmm4,nb232_fjyO(%esp) movaps %xmm5,nb232_fjzO(%esp) movaps %xmm0,nb232_fixO(%esp) movaps %xmm1,nb232_fiyO(%esp) movaps %xmm2,nb232_fizO(%esp) ## O-H1 interaction movaps nb232_rinvOH1(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb232_krf(%esp),%xmm5 movaps %xmm0,%xmm1 mulps nb232_rsqOH1(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm4 addps %xmm7,%xmm4 ## xmm4=rinv+ krsq mulps %xmm0,%xmm0 subps nb232_crf(%esp),%xmm4 mulps nb232_qqOH(%esp),%xmm4 ## xmm4=voul=qq*(rinv+ krsq) mulps nb232_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps nb232_qqOH(%esp),%xmm7 ## xmm7 = coul part of fscal addps %xmm4,%xmm6 ## add to local vctot mulps %xmm7,%xmm0 ## fsOH1 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 xorps %xmm3,%xmm3 movaps %xmm3,%xmm4 movaps %xmm3,%xmm5 mulps nb232_dxOH1(%esp),%xmm0 mulps nb232_dyOH1(%esp),%xmm1 mulps nb232_dzOH1(%esp),%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 addps nb232_fixO(%esp),%xmm0 addps nb232_fiyO(%esp),%xmm1 addps nb232_fizO(%esp),%xmm2 movaps %xmm3,nb232_fjxH1(%esp) movaps %xmm4,nb232_fjyH1(%esp) movaps %xmm5,nb232_fjzH1(%esp) movaps %xmm0,nb232_fixO(%esp) movaps %xmm1,nb232_fiyO(%esp) movaps %xmm2,nb232_fizO(%esp) ## O-H2 interaction movaps nb232_rinvOH2(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb232_krf(%esp),%xmm5 movaps %xmm0,%xmm1 mulps nb232_rsqOH2(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm4 addps %xmm7,%xmm4 ## xmm4=r inv+ krsq mulps %xmm0,%xmm0 subps nb232_crf(%esp),%xmm4 mulps nb232_qqOH(%esp),%xmm4 ## xmm4=voul=qq*(rinv+ krsq) mulps nb232_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps nb232_qqOH(%esp),%xmm7 ## xmm7 = coul part of fscal addps %xmm4,%xmm6 ## add to local vctot mulps %xmm7,%xmm0 ## fsOH2 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 xorps %xmm3,%xmm3 movaps %xmm3,%xmm4 movaps %xmm3,%xmm5 mulps nb232_dxOH2(%esp),%xmm0 mulps nb232_dyOH2(%esp),%xmm1 mulps nb232_dzOH2(%esp),%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 addps nb232_fixO(%esp),%xmm0 addps nb232_fiyO(%esp),%xmm1 addps nb232_fizO(%esp),%xmm2 movaps %xmm3,nb232_fjxH2(%esp) movaps %xmm4,nb232_fjyH2(%esp) movaps %xmm5,nb232_fjzH2(%esp) movaps %xmm0,nb232_fixO(%esp) movaps %xmm1,nb232_fiyO(%esp) movaps %xmm2,nb232_fizO(%esp) ## H1-O interaction movaps nb232_rinvH1O(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb232_krf(%esp),%xmm5 movaps %xmm0,%xmm1 mulps nb232_rsqH1O(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm4 addps %xmm7,%xmm4 ## xmm4=rinv+ krsq mulps %xmm0,%xmm0 subps nb232_crf(%esp),%xmm4 mulps nb232_qqOH(%esp),%xmm4 ## xmm4=voul=qq*(rinv+ krsq) mulps nb232_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps nb232_qqOH(%esp),%xmm7 ## xmm7 = coul part of fscal addps %xmm4,%xmm6 ## add to local vctot mulps %xmm7,%xmm0 ## fsOH2 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 movaps nb232_fjxO(%esp),%xmm3 movaps nb232_fjyO(%esp),%xmm4 movaps nb232_fjzO(%esp),%xmm5 mulps nb232_dxH1O(%esp),%xmm0 mulps nb232_dyH1O(%esp),%xmm1 mulps nb232_dzH1O(%esp),%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 addps nb232_fixH1(%esp),%xmm0 addps nb232_fiyH1(%esp),%xmm1 addps nb232_fizH1(%esp),%xmm2 movaps %xmm3,nb232_fjxO(%esp) movaps %xmm4,nb232_fjyO(%esp) movaps %xmm5,nb232_fjzO(%esp) movaps %xmm0,nb232_fixH1(%esp) movaps %xmm1,nb232_fiyH1(%esp) movaps %xmm2,nb232_fizH1(%esp) ## H1-H1 interaction movaps nb232_rinvH1H1(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb232_krf(%esp),%xmm5 movaps %xmm0,%xmm1 mulps nb232_rsqH1H1(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm4 addps %xmm7,%xmm4 ## xmm4=r inv+ krsq subps nb232_crf(%esp),%xmm4 mulps %xmm0,%xmm0 mulps nb232_qqHH(%esp),%xmm4 ## xmm4=voul=qq*(rinv+ krsq) mulps nb232_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps nb232_qqHH(%esp),%xmm7 ## xmm7 = coul part of fscal addps %xmm4,%xmm6 ## add to local vctot mulps %xmm7,%xmm0 ## fsOH2 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 movaps nb232_fjxH1(%esp),%xmm3 movaps nb232_fjyH1(%esp),%xmm4 movaps nb232_fjzH1(%esp),%xmm5 mulps nb232_dxH1H1(%esp),%xmm0 mulps nb232_dyH1H1(%esp),%xmm1 mulps nb232_dzH1H1(%esp),%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 addps nb232_fixH1(%esp),%xmm0 addps nb232_fiyH1(%esp),%xmm1 addps nb232_fizH1(%esp),%xmm2 movaps %xmm3,nb232_fjxH1(%esp) movaps %xmm4,nb232_fjyH1(%esp) movaps %xmm5,nb232_fjzH1(%esp) movaps %xmm0,nb232_fixH1(%esp) movaps %xmm1,nb232_fiyH1(%esp) movaps %xmm2,nb232_fizH1(%esp) ## H1-H2 interaction movaps nb232_rinvH1H2(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb232_krf(%esp),%xmm5 movaps %xmm0,%xmm1 mulps nb232_rsqH1H2(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm4 addps %xmm7,%xmm4 ## xmm4=r inv+ krsq mulps %xmm0,%xmm0 subps nb232_crf(%esp),%xmm4 mulps nb232_qqHH(%esp),%xmm4 ## xmm4=voul=qq*(rinv+ krsq) mulps nb232_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps nb232_qqHH(%esp),%xmm7 ## xmm7 = coul part of fscal addps %xmm4,%xmm6 ## add to local vctot mulps %xmm7,%xmm0 ## fsOH2 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 movaps nb232_fjxH2(%esp),%xmm3 movaps nb232_fjyH2(%esp),%xmm4 movaps nb232_fjzH2(%esp),%xmm5 mulps nb232_dxH1H2(%esp),%xmm0 mulps nb232_dyH1H2(%esp),%xmm1 mulps nb232_dzH1H2(%esp),%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 addps nb232_fixH1(%esp),%xmm0 addps nb232_fiyH1(%esp),%xmm1 addps nb232_fizH1(%esp),%xmm2 movaps %xmm3,nb232_fjxH2(%esp) movaps %xmm4,nb232_fjyH2(%esp) movaps %xmm5,nb232_fjzH2(%esp) movaps %xmm0,nb232_fixH1(%esp) movaps %xmm1,nb232_fiyH1(%esp) movaps %xmm2,nb232_fizH1(%esp) ## H2-O interaction movaps nb232_rinvH2O(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb232_krf(%esp),%xmm5 movaps %xmm0,%xmm1 mulps nb232_rsqH2O(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm4 addps %xmm7,%xmm4 ## xmm4=r inv+ krsq subps nb232_crf(%esp),%xmm4 mulps %xmm0,%xmm0 mulps nb232_qqOH(%esp),%xmm4 ## xmm4=voul=qq*(rinv+ krsq) mulps nb232_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps nb232_qqOH(%esp),%xmm7 ## xmm7 = coul part of fscal addps %xmm4,%xmm6 ## add to local vctot mulps %xmm7,%xmm0 ## fsOH2 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 movaps nb232_fjxO(%esp),%xmm3 movaps nb232_fjyO(%esp),%xmm4 movaps nb232_fjzO(%esp),%xmm5 mulps nb232_dxH2O(%esp),%xmm0 mulps nb232_dyH2O(%esp),%xmm1 mulps nb232_dzH2O(%esp),%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 addps nb232_fixH2(%esp),%xmm0 addps nb232_fiyH2(%esp),%xmm1 addps nb232_fizH2(%esp),%xmm2 movaps %xmm3,nb232_fjxO(%esp) movaps %xmm4,nb232_fjyO(%esp) movaps %xmm5,nb232_fjzO(%esp) movaps %xmm0,nb232_fixH2(%esp) movaps %xmm1,nb232_fiyH2(%esp) movaps %xmm2,nb232_fizH2(%esp) ## H2-H1 interaction movaps nb232_rinvH2H1(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb232_krf(%esp),%xmm5 movaps %xmm0,%xmm1 mulps nb232_rsqH2H1(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm4 addps %xmm7,%xmm4 ## xmm4=r inv+ krsq subps nb232_crf(%esp),%xmm4 mulps %xmm0,%xmm0 mulps nb232_qqHH(%esp),%xmm4 ## xmm4=voul=qq*(rinv+ krsq) mulps nb232_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps nb232_qqHH(%esp),%xmm7 ## xmm7 = coul part of fscal addps %xmm4,%xmm6 ## add to local vctot mulps %xmm7,%xmm0 ## fsOH2 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 movaps nb232_fjxH1(%esp),%xmm3 movaps nb232_fjyH1(%esp),%xmm4 movaps nb232_fjzH1(%esp),%xmm5 mulps nb232_dxH2H1(%esp),%xmm0 mulps nb232_dyH2H1(%esp),%xmm1 mulps nb232_dzH2H1(%esp),%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 addps nb232_fixH2(%esp),%xmm0 addps nb232_fiyH2(%esp),%xmm1 addps nb232_fizH2(%esp),%xmm2 movaps %xmm3,nb232_fjxH1(%esp) movaps %xmm4,nb232_fjyH1(%esp) movaps %xmm5,nb232_fjzH1(%esp) movaps %xmm0,nb232_fixH2(%esp) movaps %xmm1,nb232_fiyH2(%esp) movaps %xmm2,nb232_fizH2(%esp) ## H2-H2 interaction movaps nb232_rinvH2H2(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb232_krf(%esp),%xmm5 movaps %xmm0,%xmm1 mulps nb232_rsqH2H2(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm4 addps %xmm7,%xmm4 ## xmm4=r inv+ krsq subps nb232_crf(%esp),%xmm4 mulps %xmm0,%xmm0 mulps nb232_qqHH(%esp),%xmm4 ## xmm4=voul=qq*(rinv+ krsq) mulps nb232_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps nb232_qqHH(%esp),%xmm7 ## xmm7 = coul part of fscal addps %xmm4,%xmm6 ## add to local vctot mulps %xmm7,%xmm0 ## fsOH2 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 movaps %xmm0,%xmm1 movaps %xmm6,nb232_vctot(%esp) movaps %xmm0,%xmm2 movaps nb232_fjxH2(%esp),%xmm3 movaps nb232_fjyH2(%esp),%xmm4 movaps nb232_fjzH2(%esp),%xmm5 mulps nb232_dxH2H2(%esp),%xmm0 mulps nb232_dyH2H2(%esp),%xmm1 mulps nb232_dzH2H2(%esp),%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 addps nb232_fixH2(%esp),%xmm0 addps nb232_fiyH2(%esp),%xmm1 addps nb232_fizH2(%esp),%xmm2 movaps %xmm3,nb232_fjxH2(%esp) movaps %xmm4,nb232_fjyH2(%esp) movaps %xmm5,nb232_fjzH2(%esp) movaps %xmm0,nb232_fixH2(%esp) movaps %xmm1,nb232_fiyH2(%esp) movaps %xmm2,nb232_fizH2(%esp) movl nb232_faction(%ebp),%edi ## Did all interactions - now update j forces ## At this stage forces are still on the stack, in positions: ## fjxO, fjyO, fjzO, ... , fjzH2. ## Each position is a quadruplet of forces for the four ## corresponding j waters, so we need to transpose them before ## adding to the memory positions. ## ## This _used_ to be a simple transpose, but the resulting high number ## of unaligned 128-bit load/stores might trigger a possible hardware ## bug on Athlon and Opteron chips, so I have worked around it ## to use 64-bit load/stores instead. The performance hit should be ## very modest, since the 128-bit unaligned memory instructions were ## slow anyway. ## 4 j waters with three atoms each - first do Oxygen X & Y forces for 4 j particles movaps nb232_fjxO(%esp),%xmm0 ## xmm0= fjxOa fjxOb fjxOc fjxOd movaps nb232_fjyO(%esp),%xmm2 ## xmm1= fjyOa fjyOb fjyOc fjyOd movlps (%edi,%eax,4),%xmm3 movlps (%edi,%ecx,4),%xmm4 movaps %xmm0,%xmm1 unpcklps %xmm2,%xmm0 ## xmm0= fjxOa fjyOa fjxOb fjyOb unpckhps %xmm2,%xmm1 ## xmm1= fjxOc fjyOc fjxOd fjyOd movhps (%edi,%ebx,4),%xmm3 movhps (%edi,%edx,4),%xmm4 addps %xmm0,%xmm3 addps %xmm1,%xmm4 movlps %xmm3,(%edi,%eax,4) movlps %xmm4,(%edi,%ecx,4) movhps %xmm3,(%edi,%ebx,4) movhps %xmm4,(%edi,%edx,4) ## Oxygen Z & first hydrogen X forces for 4 j particles movaps nb232_fjzO(%esp),%xmm0 ## xmm0= fjzOa fjzOb fjzOc fjzOd movaps nb232_fjxH1(%esp),%xmm2 ## xmm1= fjxH1a fjxH1b fjxH1c fjxH1d movlps 8(%edi,%eax,4),%xmm3 movlps 8(%edi,%ecx,4),%xmm4 movaps %xmm0,%xmm1 unpcklps %xmm2,%xmm0 ## xmm0= fjzOa fjxH1a fjzOb fjxH1b unpckhps %xmm2,%xmm1 ## xmm1= fjzOc fjxH1c fjzOd fjxH1d movhps 8(%edi,%ebx,4),%xmm3 movhps 8(%edi,%edx,4),%xmm4 addps %xmm0,%xmm3 addps %xmm1,%xmm4⌨️ 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?