nb_kernel233_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,059 行 · 第 1/5 页
S
2,059 行
## move four coordinates to xmm0-xmm2 movlps (%esi,%eax,4),%xmm4 movlps (%esi,%ecx,4),%xmm5 movss 8(%esi,%eax,4),%xmm2 movss 8(%esi,%ecx,4),%xmm6 movhps (%esi,%ebx,4),%xmm4 movhps (%esi,%edx,4),%xmm5 movss 8(%esi,%ebx,4),%xmm0 movss 8(%esi,%edx,4),%xmm1 shufps $0,%xmm0,%xmm2 shufps $0,%xmm1,%xmm6 movaps %xmm4,%xmm0 movaps %xmm4,%xmm1 shufps $136,%xmm6,%xmm2 ## constant 10001000 shufps $136,%xmm5,%xmm0 ## constant 10001000 shufps $221,%xmm5,%xmm1 ## constant 11011101 ## move ixO-izO to xmm4-xmm6 movaps nb233_ixO(%esp),%xmm4 movaps nb233_iyO(%esp),%xmm5 movaps nb233_izO(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb233_dxO(%esp) movaps %xmm5,nb233_dyO(%esp) movaps %xmm6,nb233_dzO(%esp) ## square it mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 mulps %xmm6,%xmm6 addps %xmm5,%xmm4 addps %xmm6,%xmm4 movaps %xmm4,%xmm7 ## rsqO in xmm7 ## move ixH1-izH1 to xmm4-xmm6 movaps nb233_ixH1(%esp),%xmm4 movaps nb233_iyH1(%esp),%xmm5 movaps nb233_izH1(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb233_dxH1(%esp) movaps %xmm5,nb233_dyH1(%esp) movaps %xmm6,nb233_dzH1(%esp) ## square it mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 mulps %xmm6,%xmm6 addps %xmm5,%xmm6 addps %xmm4,%xmm6 ## rsqH1 in xmm6 ## move ixH2-izH2 to xmm3-xmm5 movaps nb233_ixH2(%esp),%xmm3 movaps nb233_iyH2(%esp),%xmm4 movaps nb233_izH2(%esp),%xmm5 ## calc dr subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 ## store dr movaps %xmm3,nb233_dxH2(%esp) movaps %xmm4,nb233_dyH2(%esp) movaps %xmm5,nb233_dzH2(%esp) ## square it mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm4,%xmm5 addps %xmm3,%xmm5 ## move ixM-izM to xmm2-xmm4 movaps nb233_iyM(%esp),%xmm3 movaps nb233_izM(%esp),%xmm4 subps %xmm1,%xmm3 subps %xmm2,%xmm4 movaps nb233_ixM(%esp),%xmm2 subps %xmm0,%xmm2 ## store dr movaps %xmm2,nb233_dxM(%esp) movaps %xmm3,nb233_dyM(%esp) movaps %xmm4,nb233_dzM(%esp) ## square it mulps %xmm2,%xmm2 mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 addps %xmm3,%xmm4 addps %xmm2,%xmm4 ## rsqM in xmm4, rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 movaps %xmm4,%xmm0 movaps %xmm5,%xmm1 movaps %xmm6,%xmm2 mulps nb233_krf(%esp),%xmm0 mulps nb233_krf(%esp),%xmm1 mulps nb233_krf(%esp),%xmm2 movaps %xmm0,nb233_krsqM(%esp) movaps %xmm1,nb233_krsqH2(%esp) movaps %xmm2,nb233_krsqH1(%esp) ## rsqH1 - seed in xmm2 rsqrtps %xmm6,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb233_three(%esp),%xmm0 mulps %xmm6,%xmm2 ## rsq*lu*lu subps %xmm2,%xmm0 ## constant 30-rsq*lu*lu mulps %xmm3,%xmm0 ## lu*(3-rsq*lu*lu) mulps nb233_half(%esp),%xmm0 movaps %xmm0,nb233_rinvH1(%esp) ## rinvH1 ## rsqH2 - seed to xmm2 rsqrtps %xmm5,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb233_three(%esp),%xmm0 mulps %xmm5,%xmm2 ## rsq*lu*lu subps %xmm2,%xmm0 ## constant 30-rsq*lu*lu mulps %xmm3,%xmm0 ## lu*(3-rsq*lu*lu) mulps nb233_half(%esp),%xmm0 movaps %xmm0,nb233_rinvH2(%esp) ## rinvH2 ## rsqM - seed to xmm2 rsqrtps %xmm4,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb233_three(%esp),%xmm0 mulps %xmm4,%xmm2 ## rsq*lu*lu subps %xmm2,%xmm0 ## constant 30-rsq*lu*lu mulps %xmm3,%xmm0 ## lu*(3-rsq*lu*lu) mulps nb233_half(%esp),%xmm0 movaps %xmm0,nb233_rinvM(%esp) ## Do the O LJ-only interaction directly. ## rsqO is in xmm7 rsqrtps %xmm7,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb233_three(%esp),%xmm4 mulps %xmm7,%xmm2 ## rsq*lu*lu subps %xmm2,%xmm4 ## constant 30-rsq*lu*lu mulps %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulps nb233_half(%esp),%xmm4 movaps %xmm4,%xmm0 ## xmm0=rinvO mulps %xmm0,%xmm7 mulps nb233_tsc(%esp),%xmm7 ## rtab movhlps %xmm7,%xmm5 cvttps2pi %xmm7,%mm6 cvttps2pi %xmm5,%mm7 ## mm6/mm7 contain lu indices cvtpi2ps %mm6,%xmm6 cvtpi2ps %mm7,%xmm5 movlhps %xmm5,%xmm6 subps %xmm6,%xmm7 movaps %xmm7,%xmm1 ## xmm1=eps movaps %xmm1,%xmm2 mulps %xmm2,%xmm2 ## xmm2=eps2 pslld $3,%mm6 pslld $3,%mm7 movd %eax,%mm0 movd %ebx,%mm1 movd %ecx,%mm2 movd %edx,%mm3 movl nb233_VFtab(%ebp),%esi movd %mm6,%eax psrlq $32,%mm6 movd %mm7,%ecx psrlq $32,%mm7 movd %mm6,%ebx movd %mm7,%edx ## dispersion movlps (%esi,%eax,4),%xmm5 movlps (%esi,%ecx,4),%xmm7 movhps (%esi,%ebx,4),%xmm5 movhps (%esi,%edx,4),%xmm7 ## got half dispersion table movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm5 ## constant 11011101 movlps 8(%esi,%eax,4),%xmm7 movlps 8(%esi,%ecx,4),%xmm3 movhps 8(%esi,%ebx,4),%xmm7 movhps 8(%esi,%edx,4),%xmm3 ## other half of dispersion table movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## constant 10001000 shufps $221,%xmm3,%xmm7 ## constant 11011101 ## dispersion table ready, in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb233_two(%esp),%xmm7 ## two*Heps2 addps %xmm6,%xmm7 addps %xmm5,%xmm7 ## xmm7=FF mulps %xmm1,%xmm5 ## xmm5=eps*Fp addps %xmm4,%xmm5 ## xmm5=VV movaps nb233_c6(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijD mulps %xmm4,%xmm5 ## Vvdw6 mulps nb233_tsc(%esp),%xmm7 ## put scalar force on stack Update Vvdwtot directly addps nb233_Vvdwtot(%esp),%xmm5 movaps %xmm7,nb233_fstmp(%esp) movaps %xmm5,nb233_Vvdwtot(%esp) ## repulsion movlps 16(%esi,%eax,4),%xmm5 movlps 16(%esi,%ecx,4),%xmm7 movhps 16(%esi,%ebx,4),%xmm5 movhps 16(%esi,%edx,4),%xmm7 ## got half repulsion table movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm5 ## constant 11011101 movlps 24(%esi,%eax,4),%xmm7 movlps 24(%esi,%ecx,4),%xmm3 movhps 24(%esi,%ebx,4),%xmm7 movhps 24(%esi,%edx,4),%xmm3 ## other half of repulsion table movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## constant 10001000 shufps $221,%xmm3,%xmm7 ## constant 11011101 ## table ready, in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb233_two(%esp),%xmm7 ## two*Heps2 addps %xmm6,%xmm7 addps %xmm5,%xmm7 ## xmm7=FF mulps %xmm1,%xmm5 ## xmm5=eps*Fp addps %xmm4,%xmm5 ## xmm5=VV movaps nb233_c12(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijR mulps %xmm4,%xmm5 ## Vvdw12 mulps nb233_tsc(%esp),%xmm7 addps nb233_fstmp(%esp),%xmm7 addps nb233_Vvdwtot(%esp),%xmm5 movaps %xmm5,nb233_Vvdwtot(%esp) movd %mm0,%eax movd %mm1,%ebx movd %mm2,%ecx movd %mm3,%edx xorps %xmm1,%xmm1 mulps %xmm0,%xmm7 subps %xmm7,%xmm1 movaps nb233_dxO(%esp),%xmm3 movaps nb233_dyO(%esp),%xmm4 movaps nb233_dzO(%esp),%xmm5 mulps %xmm1,%xmm3 mulps %xmm1,%xmm4 mulps %xmm1,%xmm5 ## tx in xmm3-xmm5 ## update O forces movaps nb233_fixO(%esp),%xmm0 movaps nb233_fiyO(%esp),%xmm1 movaps nb233_fizO(%esp),%xmm2 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 movaps %xmm0,nb233_fixO(%esp) movaps %xmm1,nb233_fiyO(%esp) movaps %xmm2,nb233_fizO(%esp) ## update j forces with water O movaps %xmm3,nb233_fjx(%esp) movaps %xmm4,nb233_fjy(%esp) movaps %xmm5,nb233_fjz(%esp) ## Do H1 interaction movaps nb233_rinvH1(%esp),%xmm7 movaps %xmm7,%xmm4 mulps %xmm4,%xmm4 ## xmm7=rinv, xmm4=rinvsq movaps %xmm7,%xmm0 movaps nb233_krsqH1(%esp),%xmm1 addps %xmm1,%xmm0 subps nb233_crf(%esp),%xmm0 ## xmm0=rinv+ krsq-crf mulps nb233_two(%esp),%xmm1 subps %xmm1,%xmm7 mulps nb233_qqH(%esp),%xmm0 mulps nb233_qqH(%esp),%xmm7 mulps %xmm7,%xmm4 ## total fs H1 in xmm4 addps nb233_vctot(%esp),%xmm0 movaps %xmm0,nb233_vctot(%esp) movaps nb233_dxH1(%esp),%xmm0 movaps nb233_dyH1(%esp),%xmm1 movaps nb233_dzH1(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## update H1 forces movaps nb233_fixH1(%esp),%xmm3 movaps nb233_fiyH1(%esp),%xmm4 movaps nb233_fizH1(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb233_fixH1(%esp) movaps %xmm4,nb233_fiyH1(%esp) movaps %xmm7,nb233_fizH1(%esp) ## update j forces with water H1 addps nb233_fjx(%esp),%xmm0 addps nb233_fjy(%esp),%xmm1 addps nb233_fjz(%esp),%xmm2 movaps %xmm0,nb233_fjx(%esp) movaps %xmm1,nb233_fjy(%esp) movaps %xmm2,nb233_fjz(%esp) ## Done with H1, do H2 interactions movaps nb233_rinvH2(%esp),%xmm7 movaps %xmm7,%xmm4 mulps %xmm4,%xmm4 ## xmm7=rinv, xmm4=rinvsq movaps %xmm7,%xmm0 movaps nb233_krsqH2(%esp),%xmm1 addps %xmm1,%xmm0 subps nb233_crf(%esp),%xmm0 ## xmm0=rinv+ krsq-crf mulps nb233_two(%esp),%xmm1 subps %xmm1,%xmm7 mulps nb233_qqH(%esp),%xmm0 mulps nb233_qqH(%esp),%xmm7 mulps %xmm7,%xmm4 ## total fs H2 in xmm4 addps nb233_vctot(%esp),%xmm0 movaps %xmm0,nb233_vctot(%esp) movaps nb233_dxH2(%esp),%xmm0 movaps nb233_dyH2(%esp),%xmm1 movaps nb233_dzH2(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## update H2 forces movaps nb233_fixH2(%esp),%xmm3 movaps nb233_fiyH2(%esp),%xmm4 movaps nb233_fizH2(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb233_fixH2(%esp) movaps %xmm4,nb233_fiyH2(%esp) movaps %xmm7,nb233_fizH2(%esp) addps nb233_fjx(%esp),%xmm0 addps nb233_fjy(%esp),%xmm1 addps nb233_fjz(%esp),%xmm2 movaps %xmm0,nb233_fjx(%esp) movaps %xmm1,nb233_fjy(%esp) movaps %xmm2,nb233_fjz(%esp) ## Done with H2, do M interactions movaps nb233_rinvM(%esp),%xmm7 movaps %xmm7,%xmm4 mulps %xmm4,%xmm4 ## xmm7=rinv, xmm4=rinvsq movaps %xmm7,%xmm0 movaps nb233_krsqM(%esp),%xmm1 addps %xmm1,%xmm0 subps nb233_crf(%esp),%xmm0 ## xmm0=rinv+ krsq-crf mulps nb233_two(%esp),%xmm1 subps %xmm1,%xmm7 mulps nb233_qqM(%esp),%xmm0 mulps nb233_qqM(%esp),%xmm7 mulps %xmm7,%xmm4 ## total fs M in xmm4 addps nb233_vctot(%esp),%xmm0 movaps %xmm0,nb233_vctot(%esp) movaps nb233_dxM(%esp),%xmm0 movaps nb233_dyM(%esp),%xmm1 movaps nb233_dzM(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## update M forces movaps nb233_fixM(%esp),%xmm3 movaps nb233_fiyM(%esp),%xmm4 movaps nb233_fizM(%esp),%xmm7 addps %xmm0,%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?