nb_kernel213_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,957 行 · 第 1/5 页
S
1,957 行
movhps (%esi,%edx,4),%xmm7 movaps %xmm6,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm6 ## constant 11011101 movd %mm0,%eax movd %mm1,%ebx movd %mm2,%ecx movd %mm3,%edx movaps %xmm4,nb213_c6(%esp) movaps %xmm6,nb213_c12(%esp) movl nb213_pos(%ebp),%esi ## base of pos[] leal (%eax,%eax,2),%eax ## replace jnr with j3 leal (%ebx,%ebx,2),%ebx leal (%ecx,%ecx,2),%ecx ## replace jnr with j3 leal (%edx,%edx,2),%edx ## 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 nb213_ixO(%esp),%xmm4 movaps nb213_iyO(%esp),%xmm5 movaps nb213_izO(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb213_dxO(%esp) movaps %xmm5,nb213_dyO(%esp) movaps %xmm6,nb213_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 nb213_ixH1(%esp),%xmm4 movaps nb213_iyH1(%esp),%xmm5 movaps nb213_izH1(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb213_dxH1(%esp) movaps %xmm5,nb213_dyH1(%esp) movaps %xmm6,nb213_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 nb213_ixH2(%esp),%xmm3 movaps nb213_iyH2(%esp),%xmm4 movaps nb213_izH2(%esp),%xmm5 ## calc dr subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 ## store dr movaps %xmm3,nb213_dxH2(%esp) movaps %xmm4,nb213_dyH2(%esp) movaps %xmm5,nb213_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 nb213_iyM(%esp),%xmm3 movaps nb213_izM(%esp),%xmm4 subps %xmm1,%xmm3 subps %xmm2,%xmm4 movaps nb213_ixM(%esp),%xmm2 subps %xmm0,%xmm2 ## store dr movaps %xmm2,nb213_dxM(%esp) movaps %xmm3,nb213_dyM(%esp) movaps %xmm4,nb213_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 nb213_krf(%esp),%xmm0 mulps nb213_krf(%esp),%xmm1 mulps nb213_krf(%esp),%xmm2 movaps %xmm0,nb213_krsqM(%esp) movaps %xmm1,nb213_krsqH2(%esp) movaps %xmm2,nb213_krsqH1(%esp) ## rsqH1 - seed in xmm2 rsqrtps %xmm6,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb213_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 nb213_half(%esp),%xmm0 movaps %xmm0,nb213_rinvH1(%esp) ## rinvH1 ## rsqH2 - seed to xmm2 rsqrtps %xmm5,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb213_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 nb213_half(%esp),%xmm0 movaps %xmm0,nb213_rinvH2(%esp) ## rinvH2 ## rsqM - seed to xmm2 rsqrtps %xmm4,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb213_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 nb213_half(%esp),%xmm0 movaps %xmm0,nb213_rinvM(%esp) ## Do the O LJ-only interaction directly. rcpps %xmm7,%xmm2 movaps nb213_two(%esp),%xmm1 mulps %xmm2,%xmm7 subps %xmm7,%xmm1 mulps %xmm1,%xmm2 ## rinvsq movaps %xmm2,%xmm0 mulps %xmm2,%xmm0 ## r4 mulps %xmm2,%xmm0 ## r6 movaps %xmm0,%xmm1 mulps %xmm1,%xmm1 ## r12 mulps nb213_c6(%esp),%xmm0 mulps nb213_c12(%esp),%xmm1 movaps %xmm1,%xmm3 subps %xmm0,%xmm3 ## Vvdw12-Vvdw6 addps nb213_Vvdwtot(%esp),%xmm3 movaps %xmm3,nb213_Vvdwtot(%esp) mulps nb213_six(%esp),%xmm0 mulps nb213_twelve(%esp),%xmm1 subps %xmm0,%xmm1 mulps %xmm2,%xmm1 ## fscal movaps nb213_dxO(%esp),%xmm3 movaps nb213_dyO(%esp),%xmm4 movaps nb213_dzO(%esp),%xmm5 mulps %xmm1,%xmm3 mulps %xmm1,%xmm4 mulps %xmm1,%xmm5 ## tx in xmm3-xmm5 ## update O forces movaps nb213_fixO(%esp),%xmm0 movaps nb213_fiyO(%esp),%xmm1 movaps nb213_fizO(%esp),%xmm2 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 movaps %xmm0,nb213_fixO(%esp) movaps %xmm1,nb213_fiyO(%esp) movaps %xmm2,nb213_fizO(%esp) ## update j forces with water O movaps %xmm3,nb213_fjx(%esp) movaps %xmm4,nb213_fjy(%esp) movaps %xmm5,nb213_fjz(%esp) ## Do H1 interaction movaps nb213_rinvH1(%esp),%xmm7 movaps %xmm7,%xmm4 mulps %xmm4,%xmm4 ## xmm7=rinv, xmm4=rinvsq movaps %xmm7,%xmm0 movaps nb213_krsqH1(%esp),%xmm1 addps %xmm1,%xmm0 subps nb213_crf(%esp),%xmm0 ## xmm0=rinv+ krsq-crf mulps nb213_two(%esp),%xmm1 subps %xmm1,%xmm7 mulps nb213_qqH(%esp),%xmm0 mulps nb213_qqH(%esp),%xmm7 mulps %xmm7,%xmm4 ## total fs H1 in xmm4 addps nb213_vctot(%esp),%xmm0 movaps %xmm0,nb213_vctot(%esp) movaps nb213_dxH1(%esp),%xmm0 movaps nb213_dyH1(%esp),%xmm1 movaps nb213_dzH1(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## update H1 forces movaps nb213_fixH1(%esp),%xmm3 movaps nb213_fiyH1(%esp),%xmm4 movaps nb213_fizH1(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb213_fixH1(%esp) movaps %xmm4,nb213_fiyH1(%esp) movaps %xmm7,nb213_fizH1(%esp) ## update j forces with water H1 addps nb213_fjx(%esp),%xmm0 addps nb213_fjy(%esp),%xmm1 addps nb213_fjz(%esp),%xmm2 movaps %xmm0,nb213_fjx(%esp) movaps %xmm1,nb213_fjy(%esp) movaps %xmm2,nb213_fjz(%esp) ## Done with H1, do H2 interactions movaps nb213_rinvH2(%esp),%xmm7 movaps %xmm7,%xmm4 mulps %xmm4,%xmm4 ## xmm7=rinv, xmm4=rinvsq movaps %xmm7,%xmm0 movaps nb213_krsqH2(%esp),%xmm1 addps %xmm1,%xmm0 subps nb213_crf(%esp),%xmm0 ## xmm0=rinv+ krsq-crf mulps nb213_two(%esp),%xmm1 subps %xmm1,%xmm7 mulps nb213_qqH(%esp),%xmm0 mulps nb213_qqH(%esp),%xmm7 mulps %xmm7,%xmm4 ## total fs H2 in xmm4 addps nb213_vctot(%esp),%xmm0 movaps %xmm0,nb213_vctot(%esp) movaps nb213_dxH2(%esp),%xmm0 movaps nb213_dyH2(%esp),%xmm1 movaps nb213_dzH2(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## update H2 forces movaps nb213_fixH2(%esp),%xmm3 movaps nb213_fiyH2(%esp),%xmm4 movaps nb213_fizH2(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb213_fixH2(%esp) movaps %xmm4,nb213_fiyH2(%esp) movaps %xmm7,nb213_fizH2(%esp) addps nb213_fjx(%esp),%xmm0 addps nb213_fjy(%esp),%xmm1 addps nb213_fjz(%esp),%xmm2 movaps %xmm0,nb213_fjx(%esp) movaps %xmm1,nb213_fjy(%esp) movaps %xmm2,nb213_fjz(%esp) ## Done with H2, do M interactions movaps nb213_rinvM(%esp),%xmm7 movaps %xmm7,%xmm4 mulps %xmm4,%xmm4 ## xmm7=rinv, xmm4=rinvsq movaps %xmm7,%xmm0 movaps nb213_krsqM(%esp),%xmm1 addps %xmm1,%xmm0 subps nb213_crf(%esp),%xmm0 ## xmm0=rinv+ krsq-crf mulps nb213_two(%esp),%xmm1 subps %xmm1,%xmm7 mulps nb213_qqM(%esp),%xmm0 mulps nb213_qqM(%esp),%xmm7 mulps %xmm7,%xmm4 ## total fs M in xmm4 addps nb213_vctot(%esp),%xmm0 movaps %xmm0,nb213_vctot(%esp) movaps nb213_dxM(%esp),%xmm0 movaps nb213_dyM(%esp),%xmm1 movaps nb213_dzM(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## update M forces movaps nb213_fixM(%esp),%xmm3 movaps nb213_fiyM(%esp),%xmm4 movaps nb213_fizM(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb213_fixM(%esp) movaps %xmm4,nb213_fiyM(%esp) movaps %xmm7,nb213_fizM(%esp) movl nb213_faction(%ebp),%edi ## update j forces from stored values addps nb213_fjx(%esp),%xmm0 addps nb213_fjy(%esp),%xmm1 addps nb213_fjz(%esp),%xmm2 movlps (%edi,%eax,4),%xmm4 movlps (%edi,%ecx,4),%xmm7 movhps (%edi,%ebx,4),%xmm4 movhps (%edi,%edx,4),%xmm7 movaps %xmm4,%xmm3 shufps $136,%xmm7,%xmm3 ## constant 10001000 shufps $221,%xmm7,%xmm4 ## constant 11011101 ## xmm3 has fjx, xmm4 has fjy subps %xmm0,%xmm3 subps %xmm1,%xmm4 ## unpack them back for storing movaps %xmm3,%xmm7 unpcklps %xmm4,%xmm7 unpckhps %xmm4,%xmm3 movlps %xmm7,(%edi,%eax,4) movlps %xmm3,(%edi,%ecx,4) movhps %xmm7,(%edi,%ebx,4) movhps %xmm3,(%edi,%edx,4) ## finally z forces movss 8(%edi,%eax,4),%xmm0 movss 8(%edi,%ebx,4),%xmm1 movss 8(%edi,%ecx,4),%xmm3 movss 8(%edi,%edx,4),%xmm4 subss %xmm2,%xmm0 shufps $229,%xmm2,%xmm2 ## constant 11100101 subss %xmm2,%xmm1 shufps $234,%xmm2,%xmm2 ## constant 11101010 subss %xmm2,%xmm3 shufps $255,%xmm2,%xmm2 ## constant 11111111 subss %xmm2,%xmm4 movss %xmm0,8(%edi,%eax,4) movss %xmm1,8(%edi,%ebx,4) movss %xmm3,8(%edi,%ecx,4) movss %xmm4,8(%edi,%edx,4) ## should we do one more iteration? subl $4,nb213_innerk(%esp) jl _nb_kernel213_ia32_sse.nb213_odd_inner jmp _nb_kernel213_ia32_sse.nb213_unroll_loop_nb_kernel213_ia32_sse.nb213_odd_inner: addl $4,nb213_innerk(%esp) jnz _nb_kernel213_ia32_sse.nb213_odd_loop jmp _nb_kernel213_ia32_sse.nb213_updateouterdata_nb_kernel213_ia32_sse.nb213_odd_loop: movl nb213_innerjjnr(%esp),%edx ## pointer to jjnr[k] movl (%edx),%eax addl $4,nb213_innerjjnr(%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?