nb_kernel234_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,980 行 · 第 1/5 页
S
1,980 行
mulps nb234_rsqMM(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq subps nb234_crf(%esp),%xmm6 mulps %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps nb234_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps %xmm3,%xmm7 ## xmm7 = coul part of fscal addps nb234_vctot(%esp),%xmm6 movaps %xmm6,nb234_vctot(%esp) mulps %xmm7,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb234_dxMM(%esp),%xmm0 mulps nb234_dyMM(%esp),%xmm1 mulps nb234_dzMM(%esp),%xmm2 ## update forces M - j water movaps nb234_fjxO(%esp),%xmm3 movaps nb234_fjyO(%esp),%xmm4 movaps nb234_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb234_fjxO(%esp) movaps %xmm4,nb234_fjyO(%esp) movaps %xmm5,nb234_fjzO(%esp) addps nb234_fixM(%esp),%xmm0 addps nb234_fiyM(%esp),%xmm1 addps nb234_fizM(%esp),%xmm2 movaps %xmm0,nb234_fixM(%esp) movaps %xmm1,nb234_fiyM(%esp) movaps %xmm2,nb234_fizM(%esp) ## i H1 & H2 simultaneously first get i particle coords: movaps nb234_ixH1(%esp),%xmm0 movaps nb234_iyH1(%esp),%xmm1 movaps nb234_izH1(%esp),%xmm2 movaps nb234_ixH2(%esp),%xmm3 movaps nb234_iyH2(%esp),%xmm4 movaps nb234_izH2(%esp),%xmm5 subps nb234_jxO(%esp),%xmm0 subps nb234_jyO(%esp),%xmm1 subps nb234_jzO(%esp),%xmm2 subps nb234_jxO(%esp),%xmm3 subps nb234_jyO(%esp),%xmm4 subps nb234_jzO(%esp),%xmm5 movaps %xmm0,nb234_dxH1H1(%esp) movaps %xmm1,nb234_dyH1H1(%esp) movaps %xmm2,nb234_dzH1H1(%esp) movaps %xmm3,nb234_dxH2H2(%esp) movaps %xmm4,nb234_dyH2H2(%esp) movaps %xmm5,nb234_dzH2H2(%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 movaps %xmm0,nb234_rsqH1H1(%esp) movaps %xmm4,nb234_rsqH2H2(%esp) ## start doing invsqrt use rsq values in xmm0, xmm4 rsqrtps %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movaps %xmm1,%xmm2 movaps %xmm5,%xmm6 mulps %xmm1,%xmm1 mulps %xmm5,%xmm5 movaps nb234_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 nb234_half(%esp),%xmm3 ## rinvH1H1 mulps nb234_half(%esp),%xmm7 ## rinvH2H2 movaps %xmm3,nb234_rinvH1H1(%esp) movaps %xmm7,nb234_rinvH2H2(%esp) ## Do H1 coulomb interaction movaps nb234_rinvH1H1(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb234_krf(%esp),%xmm5 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb234_qqHH(%esp),%xmm3 movhps nb234_qqMH(%esp),%xmm3 shufps $193,%xmm3,%xmm3 ## constant 11000001 mulps nb234_rsqH1H1(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq subps nb234_crf(%esp),%xmm6 mulps %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps nb234_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps %xmm3,%xmm7 ## xmm7 = coul part of fscal addps nb234_vctot(%esp),%xmm6 movaps %xmm6,nb234_vctot(%esp) mulps %xmm7,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb234_dxH1H1(%esp),%xmm0 mulps nb234_dyH1H1(%esp),%xmm1 mulps nb234_dzH1H1(%esp),%xmm2 ## update forces H1 - j water movaps nb234_fjxO(%esp),%xmm3 movaps nb234_fjyO(%esp),%xmm4 movaps nb234_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb234_fjxO(%esp) movaps %xmm4,nb234_fjyO(%esp) movaps %xmm5,nb234_fjzO(%esp) addps nb234_fixH1(%esp),%xmm0 addps nb234_fiyH1(%esp),%xmm1 addps nb234_fizH1(%esp),%xmm2 movaps %xmm0,nb234_fixH1(%esp) movaps %xmm1,nb234_fiyH1(%esp) movaps %xmm2,nb234_fizH1(%esp) ## H2 Coulomb movaps nb234_rinvH2H2(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb234_krf(%esp),%xmm5 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb234_qqHH(%esp),%xmm3 movhps nb234_qqMH(%esp),%xmm3 shufps $193,%xmm3,%xmm3 ## constant 11000001 mulps nb234_rsqH2H2(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq subps nb234_crf(%esp),%xmm6 mulps %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps nb234_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps %xmm3,%xmm7 ## xmm7 = coul part of fscal addps nb234_vctot(%esp),%xmm6 ## local vctot summation variable movaps %xmm6,nb234_vctot(%esp) mulps %xmm7,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb234_dxH2H2(%esp),%xmm0 mulps nb234_dyH2H2(%esp),%xmm1 mulps nb234_dzH2H2(%esp),%xmm2 ## update forces H2 - j water movaps nb234_fjxO(%esp),%xmm3 movaps nb234_fjyO(%esp),%xmm4 movaps nb234_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb234_fjxO(%esp) movaps %xmm4,nb234_fjyO(%esp) movaps %xmm5,nb234_fjzO(%esp) addps nb234_fixH2(%esp),%xmm0 addps nb234_fiyH2(%esp),%xmm1 addps nb234_fizH2(%esp),%xmm2 movaps %xmm0,nb234_fixH2(%esp) movaps %xmm1,nb234_fiyH2(%esp) movaps %xmm2,nb234_fizH2(%esp) movl nb234_faction(%ebp),%esi ## update j water forces from local variables. ## transpose back first movaps nb234_fjxO(%esp),%xmm0 ## Ox H1x H2x Mx movaps nb234_fjyO(%esp),%xmm1 ## Oy H1y H2y My movaps nb234_fjzO(%esp),%xmm2 ## Oz H1z H2z Mz movaps %xmm0,%xmm3 movaps %xmm0,%xmm4 unpcklps %xmm1,%xmm3 ## Ox Oy - - shufps $0x1,%xmm2,%xmm4 ## h1x - Oz - movaps %xmm1,%xmm5 movaps %xmm0,%xmm6 unpcklps %xmm2,%xmm5 ## - - H1y H1z unpckhps %xmm1,%xmm6 ## h2x h2y - - unpckhps %xmm2,%xmm1 ## - - My Mz shufps $50,%xmm0,%xmm2 ## (00110010) h2z - Mx - shufps $36,%xmm4,%xmm3 ## constant 00100100 ;# Ox Oy Oz H1x shufps $78,%xmm6,%xmm5 ## constant 01001110 ;# h1y h1z h2x h2y shufps $232,%xmm1,%xmm2 ## constant 11101000 ;# h2z mx my mz movlps (%esi,%eax,4),%xmm0 movlps 16(%esi,%eax,4),%xmm1 movlps 32(%esi,%eax,4),%xmm4 movhps 8(%esi,%eax,4),%xmm0 movhps 24(%esi,%eax,4),%xmm1 movhps 40(%esi,%eax,4),%xmm4 addps %xmm3,%xmm0 addps %xmm5,%xmm1 addps %xmm2,%xmm4 movlps %xmm0,(%esi,%eax,4) movlps %xmm1,16(%esi,%eax,4) movlps %xmm4,32(%esi,%eax,4) movhps %xmm0,8(%esi,%eax,4) movhps %xmm1,24(%esi,%eax,4) movhps %xmm4,40(%esi,%eax,4) decl nb234_innerk(%esp) jz _nb_kernel234_ia32_sse.nb234_updateouterdata jmp _nb_kernel234_ia32_sse.nb234_single_loop_nb_kernel234_ia32_sse.nb234_updateouterdata: movl nb234_ii3(%esp),%ecx movl nb234_faction(%ebp),%edi movl nb234_fshift(%ebp),%esi movl nb234_is3(%esp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb234_fixO(%esp),%xmm0 movaps nb234_fiyO(%esp),%xmm1 movaps nb234_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 nb234_fixH1(%esp),%xmm0 movaps nb234_fiyH1(%esp),%xmm1 movaps nb234_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 nb234_fixH2(%esp),%xmm0 movaps nb234_fiyH2(%esp),%xmm1 movaps nb234_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 nb234_fixM(%esp),%xmm0 movaps nb234_fiyM(%esp),%xmm1 movaps nb234_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⌨️ 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?