nb_kernel214_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,984 行 · 第 1/5 页
S
1,984 行
unpckhps %xmm3,%xmm1 movaps %xmm4,%xmm2 movaps %xmm0,%xmm3 unpcklps %xmm5,%xmm4 unpckhps %xmm5,%xmm2 unpcklps %xmm1,%xmm0 unpckhps %xmm1,%xmm3 ## results are now in xmm4, xmm2, xmm0, xmm3 ## load the corresponding j forces from memory movlps 16(%edi,%eax,4),%xmm1 movlps 16(%edi,%ebx,4),%xmm5 movlps 16(%edi,%ecx,4),%xmm6 movlps 16(%edi,%edx,4),%xmm7 movhps 24(%edi,%eax,4),%xmm1 movhps 24(%edi,%ebx,4),%xmm5 movhps 24(%edi,%ecx,4),%xmm6 movhps 24(%edi,%edx,4),%xmm7 ## add addps %xmm4,%xmm1 addps %xmm2,%xmm5 addps %xmm0,%xmm6 addps %xmm3,%xmm7 ## store back movlps %xmm1,16(%edi,%eax,4) movlps %xmm5,16(%edi,%ebx,4) movlps %xmm6,16(%edi,%ecx,4) movlps %xmm7,16(%edi,%edx,4) movhps %xmm1,24(%edi,%eax,4) movhps %xmm5,24(%edi,%ebx,4) movhps %xmm6,24(%edi,%ecx,4) movhps %xmm7,24(%edi,%edx,4) ## step 3 : transpose fjzH2, fjxM, fjyM, fjzM. xmm4 is scratch movaps nb214_fjzH2(%esp),%xmm0 movaps nb214_fjxM(%esp),%xmm1 movaps nb214_fjyM(%esp),%xmm2 movaps nb214_fjzM(%esp),%xmm3 movaps %xmm0,%xmm4 movaps %xmm1,%xmm5 unpcklps %xmm2,%xmm4 unpcklps %xmm3,%xmm5 unpckhps %xmm2,%xmm0 unpckhps %xmm3,%xmm1 movaps %xmm4,%xmm2 movaps %xmm0,%xmm3 unpcklps %xmm5,%xmm4 unpckhps %xmm5,%xmm2 unpcklps %xmm1,%xmm0 unpckhps %xmm1,%xmm3 ## results are now in xmm0, xmm1, xmm2, xmm3 ## load the corresponding j forces from memory movlps 32(%edi,%eax,4),%xmm1 movlps 32(%edi,%ebx,4),%xmm5 movlps 32(%edi,%ecx,4),%xmm6 movlps 32(%edi,%edx,4),%xmm7 movhps 40(%edi,%eax,4),%xmm1 movhps 40(%edi,%ebx,4),%xmm5 movhps 40(%edi,%ecx,4),%xmm6 movhps 40(%edi,%edx,4),%xmm7 ## add addps %xmm4,%xmm1 addps %xmm2,%xmm5 addps %xmm0,%xmm6 addps %xmm3,%xmm7 ## store back movlps %xmm1,32(%edi,%eax,4) movlps %xmm5,32(%edi,%ebx,4) movlps %xmm6,32(%edi,%ecx,4) movlps %xmm7,32(%edi,%edx,4) movhps %xmm1,40(%edi,%eax,4) movhps %xmm5,40(%edi,%ebx,4) movhps %xmm6,40(%edi,%ecx,4) movhps %xmm7,40(%edi,%edx,4) ## should we do one more iteration? subl $4,nb214_innerk(%esp) jl _nb_kernel214_ia32_sse.nb214_single_check jmp _nb_kernel214_ia32_sse.nb214_unroll_loop_nb_kernel214_ia32_sse.nb214_single_check: addl $4,nb214_innerk(%esp) jnz _nb_kernel214_ia32_sse.nb214_single_loop jmp _nb_kernel214_ia32_sse.nb214_updateouterdata_nb_kernel214_ia32_sse.nb214_single_loop: movl nb214_innerjjnr(%esp),%edx ## pointer to jjnr[k] movl (%edx),%eax addl $4,nb214_innerjjnr(%esp) movl nb214_pos(%ebp),%esi leal (%eax,%eax,2),%eax ## fetch j coordinates movlps (%esi,%eax,4),%xmm3 ## Ox Oy movlps 16(%esi,%eax,4),%xmm4 ## H1y H1z movlps 32(%esi,%eax,4),%xmm5 ## H2z Mx movhps 8(%esi,%eax,4),%xmm3 ## Ox Oy Oz H1x movhps 24(%esi,%eax,4),%xmm4 ## H1y H1z H2x H2y movhps 40(%esi,%eax,4),%xmm5 ## H2z Mx My Mz ## transpose movaps %xmm4,%xmm0 movaps %xmm3,%xmm1 movaps %xmm4,%xmm2 movaps %xmm3,%xmm6 shufps $18,%xmm5,%xmm4 ## (00010010) h2x - Mx - shufps $193,%xmm0,%xmm3 ## (11000001) Oy - H1y - shufps $35,%xmm5,%xmm2 ## (00100011) H2y - My - shufps $18,%xmm0,%xmm1 ## (00010010) Oz - H1z - ## xmm6: Ox - - H1x xmm5: H2z - - Mz shufps $140,%xmm4,%xmm6 ## (10001100) Ox H1x H2x Mx shufps $136,%xmm2,%xmm3 ## (10001000) Oy H1y H2y My shufps $200,%xmm5,%xmm1 ## (11001000) Oz H1z H2z Mz ## store all j coordinates in jO movaps %xmm6,nb214_jxO(%esp) movaps %xmm3,nb214_jyO(%esp) movaps %xmm1,nb214_jzO(%esp) ## do O and M in parallel movaps nb214_ixO(%esp),%xmm0 movaps nb214_iyO(%esp),%xmm1 movaps nb214_izO(%esp),%xmm2 movaps nb214_ixM(%esp),%xmm3 movaps nb214_iyM(%esp),%xmm4 movaps nb214_izM(%esp),%xmm5 subps nb214_jxO(%esp),%xmm0 subps nb214_jyO(%esp),%xmm1 subps nb214_jzO(%esp),%xmm2 subps nb214_jxO(%esp),%xmm3 subps nb214_jyO(%esp),%xmm4 subps nb214_jzO(%esp),%xmm5 movaps %xmm0,nb214_dxOO(%esp) movaps %xmm1,nb214_dyOO(%esp) movaps %xmm2,nb214_dzOO(%esp) movaps %xmm3,nb214_dxMM(%esp) movaps %xmm4,nb214_dyMM(%esp) movaps %xmm5,nb214_dzMM(%esp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 addps %xmm1,%xmm0 addps %xmm2,%xmm0 ## have rsq in xmm0 mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm3,%xmm4 addps %xmm5,%xmm4 ## have rsq in xmm4 ## Save M data movaps %xmm4,nb214_rsqMM(%esp) ## do 1/x for O and 1/sqrt(x) for M rcpss %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movss nb214_two(%esp),%xmm2 movaps %xmm5,%xmm6 mulss %xmm1,%xmm0 mulps %xmm5,%xmm5 subss %xmm0,%xmm2 movaps nb214_three(%esp),%xmm7 mulss %xmm1,%xmm2 ## constant 1/r2 mulps %xmm4,%xmm5 movss %xmm2,%xmm0 subps %xmm5,%xmm7 mulss %xmm2,%xmm2 mulps %xmm6,%xmm7 mulss %xmm0,%xmm2 ## constant 1/r6 mulps nb214_half(%esp),%xmm7 ## rinv iH1 - j water movss %xmm2,%xmm1 movaps %xmm7,nb214_rinvMM(%esp) mulss %xmm2,%xmm2 ## constant 1/r12 mulss nb214_c6(%esp),%xmm1 mulss nb214_c12(%esp),%xmm2 movss %xmm2,%xmm3 subss %xmm1,%xmm3 addss nb214_Vvdwtot(%esp),%xmm3 movss %xmm3,nb214_Vvdwtot(%esp) mulss nb214_six(%esp),%xmm1 mulss nb214_twelve(%esp),%xmm2 subss %xmm1,%xmm2 mulss %xmm2,%xmm0 ## fscal movss %xmm0,%xmm1 movss %xmm0,%xmm2 mulss nb214_dxOO(%esp),%xmm0 mulss nb214_dyOO(%esp),%xmm1 mulss nb214_dzOO(%esp),%xmm2 xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movaps %xmm3,nb214_fjxO(%esp) movaps %xmm4,nb214_fjyO(%esp) movaps %xmm5,nb214_fjzO(%esp) addss nb214_fixO(%esp),%xmm0 addss nb214_fiyO(%esp),%xmm1 addss nb214_fizO(%esp),%xmm2 movss %xmm0,nb214_fixO(%esp) movss %xmm1,nb214_fiyO(%esp) movss %xmm2,nb214_fizO(%esp) ## do M coulomb interaction movaps nb214_rinvMM(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb214_krf(%esp),%xmm5 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb214_qqMH(%esp),%xmm3 movhps nb214_qqMM(%esp),%xmm3 shufps $193,%xmm3,%xmm3 ## constant 11000001 mulps nb214_rsqMM(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq subps nb214_crf(%esp),%xmm6 mulps %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps nb214_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps %xmm3,%xmm7 ## xmm7 = coul part of fscal addps nb214_vctot(%esp),%xmm6 movaps %xmm6,nb214_vctot(%esp) mulps %xmm7,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb214_dxMM(%esp),%xmm0 mulps nb214_dyMM(%esp),%xmm1 mulps nb214_dzMM(%esp),%xmm2 ## update forces M - j water movaps nb214_fjxO(%esp),%xmm3 movaps nb214_fjyO(%esp),%xmm4 movaps nb214_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb214_fjxO(%esp) movaps %xmm4,nb214_fjyO(%esp) movaps %xmm5,nb214_fjzO(%esp) addps nb214_fixM(%esp),%xmm0 addps nb214_fiyM(%esp),%xmm1 addps nb214_fizM(%esp),%xmm2 movaps %xmm0,nb214_fixM(%esp) movaps %xmm1,nb214_fiyM(%esp) movaps %xmm2,nb214_fizM(%esp) ## i H1 & H2 simultaneously first get i particle coords: movaps nb214_ixH1(%esp),%xmm0 movaps nb214_iyH1(%esp),%xmm1 movaps nb214_izH1(%esp),%xmm2 movaps nb214_ixH2(%esp),%xmm3 movaps nb214_iyH2(%esp),%xmm4 movaps nb214_izH2(%esp),%xmm5 subps nb214_jxO(%esp),%xmm0 subps nb214_jyO(%esp),%xmm1 subps nb214_jzO(%esp),%xmm2 subps nb214_jxO(%esp),%xmm3 subps nb214_jyO(%esp),%xmm4 subps nb214_jzO(%esp),%xmm5 movaps %xmm0,nb214_dxH1H1(%esp) movaps %xmm1,nb214_dyH1H1(%esp) movaps %xmm2,nb214_dzH1H1(%esp) movaps %xmm3,nb214_dxH2H2(%esp) movaps %xmm4,nb214_dyH2H2(%esp) movaps %xmm5,nb214_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,nb214_rsqH1H1(%esp) movaps %xmm4,nb214_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 nb214_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 nb214_half(%esp),%xmm3 ## rinvH1H1 mulps nb214_half(%esp),%xmm7 ## rinvH2H2 movaps %xmm3,nb214_rinvH1H1(%esp) movaps %xmm7,nb214_rinvH2H2(%esp) ## Do H1 coulomb interaction movaps nb214_rinvH1H1(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb214_krf(%esp),%xmm5 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb214_qqHH(%esp),%xmm3 movhps nb214_qqMH(%esp),%xmm3 shufps $193,%xmm3,%xmm3 ## constant 11000001 mulps nb214_rsqH1H1(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq subps nb214_crf(%esp),%xmm6 mulps %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps nb214_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps %xmm3,%xmm7 ## xmm7 = coul part of fscal addps nb214_vctot(%esp),%xmm6 movaps %xmm6,nb214_vctot(%esp) mulps %xmm7,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb214_dxH1H1(%esp),%xmm0 mulps nb214_dyH1H1(%esp),%xmm1 mulps nb214_dzH1H1(%esp),%xmm2 ## update forces H1 - j water movaps nb214_fjxO(%esp),%xmm3 movaps nb214_fjyO(%esp),%xmm4 movaps nb214_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb214_fjxO(%esp) movaps %xmm4,nb214_fjyO(%esp) movaps %xmm5,nb214_fjzO(%esp) addps nb214_fixH1(%esp),%xmm0 addps nb214_fiyH1(%esp),%xmm1 addps nb214_fizH1(%esp),%xmm2 movaps %xmm0,nb214_fixH1(%esp) movaps %xmm1,nb214_fiyH1(%esp) movaps %xmm2,nb214_fizH1(%esp) ## H2 Coulomb movaps nb214_rinvH2H2(%esp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb214_krf(%esp),%xmm5 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb214_qqHH(%esp),%xmm3 movhps nb214_qqMH(%esp),%xmm3 shufps $193,%xmm3,%xmm3 ## constant 11000001 mulps nb214_rsqH2H2(%esp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq subps nb214_crf(%esp),%xmm6 mulps %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps nb214_two(%esp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps %xmm3,%xmm7 ## xmm7 = coul part of fscal addps nb214_vctot(%esp),%xmm6 ## local vctot summation variable movaps %xmm6,nb214_vctot(%esp) mulps %xmm7,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb214_dxH2H2(%esp),%xmm0 mulps nb214_dyH2H2(%esp),%xmm1 mulps nb214_dzH2H2(%esp),%xmm2 ## update forces H2 - j water movaps nb214_fjxO(%esp),%xmm3 movaps nb214_fjyO(%esp),%xmm4 movaps nb214_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb214_fjxO(%esp) movaps %xmm4,nb214_fjyO(%esp) movaps %xmm5,nb214_fjzO(%esp)⌨️ 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?