nb_kernel231_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,071 行 · 第 1/5 页
S
2,071 行
movaps nb231_ixO(%esp),%xmm4 movaps nb231_iyO(%esp),%xmm5 movaps nb231_izO(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb231_dxO(%esp) movaps %xmm5,nb231_dyO(%esp) movaps %xmm6,nb231_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 nb231_ixH1(%esp),%xmm4 movaps nb231_iyH1(%esp),%xmm5 movaps nb231_izH1(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb231_dxH1(%esp) movaps %xmm5,nb231_dyH1(%esp) movaps %xmm6,nb231_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 nb231_ixH2(%esp),%xmm3 movaps nb231_iyH2(%esp),%xmm4 movaps nb231_izH2(%esp),%xmm5 ## calc dr subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 ## store dr movaps %xmm3,nb231_dxH2(%esp) movaps %xmm4,nb231_dyH2(%esp) movaps %xmm5,nb231_dzH2(%esp) ## square it mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm4,%xmm5 addps %xmm3,%xmm5 ## rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 movaps %xmm5,%xmm0 movaps %xmm6,%xmm1 movaps %xmm7,%xmm2 mulps nb231_krf(%esp),%xmm0 mulps nb231_krf(%esp),%xmm1 mulps nb231_krf(%esp),%xmm2 movaps %xmm0,nb231_krsqH2(%esp) movaps %xmm1,nb231_krsqH1(%esp) movaps %xmm2,nb231_krsqO(%esp) ## start with rsqO (still in xmm7) - seed to xmm2 rsqrtps %xmm7,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb231_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 nb231_half(%esp),%xmm4 movaps %xmm4,nb231_rinvO(%esp) ## rinvO in xmm0, rsqO in xmm7 ## rsqH1 - seed in xmm2 rsqrtps %xmm6,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb231_three(%esp),%xmm4 mulps %xmm6,%xmm2 ## rsq*lu*lu subps %xmm2,%xmm4 ## constant 30-rsq*lu*lu mulps %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulps nb231_half(%esp),%xmm4 movaps %xmm4,nb231_rinvH1(%esp) ## rinvH1 in xmm6 ## rsqH2 - seed in xmm2 rsqrtps %xmm5,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb231_three(%esp),%xmm4 mulps %xmm5,%xmm2 ## rsq*lu*lu subps %xmm2,%xmm4 ## constant 30-rsq*lu*lu mulps %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulps nb231_half(%esp),%xmm4 movaps %xmm4,nb231_rinvH2(%esp) ## rinvH2 in xmm5 ## do O table interactions - rsqO in xmm7. mulps nb231_rinvO(%esp),%xmm7 mulps nb231_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 nb231_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 nb231_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 nb231_c6(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijD mulps %xmm4,%xmm5 ## Vvdw6 mulps nb231_tsc(%esp),%xmm7 ## put scalar force on stack Update Vvdwtot directly addps nb231_Vvdwtot(%esp),%xmm5 movaps %xmm7,nb231_fstmp(%esp) movaps %xmm5,nb231_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 nb231_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 nb231_c12(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijR mulps %xmm4,%xmm5 ## Vvdw12 mulps nb231_tsc(%esp),%xmm7 addps nb231_fstmp(%esp),%xmm7 movaps %xmm7,nb231_fstmp(%esp) addps nb231_Vvdwtot(%esp),%xmm5 movaps %xmm5,nb231_Vvdwtot(%esp) movaps nb231_rinvO(%esp),%xmm2 movaps nb231_krsqO(%esp),%xmm1 movaps %xmm2,%xmm3 addps %xmm1,%xmm2 ## rinv+krsq subps nb231_crf(%esp),%xmm2 ## rinv+krsq-crf movaps %xmm3,%xmm0 subps %xmm1,%xmm3 subps %xmm1,%xmm3 ## rinv-2*krsq mulps nb231_qqO(%esp),%xmm2 mulps nb231_qqO(%esp),%xmm3 mulps %xmm0,%xmm3 subps nb231_fstmp(%esp),%xmm3 mulps %xmm0,%xmm3 addps nb231_vctot(%esp),%xmm2 movaps %xmm2,nb231_vctot(%esp) movaps nb231_dxO(%esp),%xmm0 movaps nb231_dyO(%esp),%xmm1 movaps nb231_dzO(%esp),%xmm2 mulps %xmm3,%xmm0 mulps %xmm3,%xmm1 mulps %xmm3,%xmm2 movd %mm0,%eax movd %mm1,%ebx movd %mm2,%ecx movd %mm3,%edx ## update O forces movaps nb231_fixO(%esp),%xmm3 movaps nb231_fiyO(%esp),%xmm4 movaps nb231_fizO(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb231_fixO(%esp) movaps %xmm4,nb231_fiyO(%esp) movaps %xmm7,nb231_fizO(%esp) ## update j forces with water O movaps %xmm0,nb231_fjx(%esp) movaps %xmm1,nb231_fjy(%esp) movaps %xmm2,nb231_fjz(%esp) ## H1 interactions movaps nb231_rinvH1(%esp),%xmm2 movaps nb231_krsqH1(%esp),%xmm1 movaps %xmm2,%xmm3 addps %xmm1,%xmm2 ## rinv+krsq subps nb231_crf(%esp),%xmm2 ## rinv+krsq-crf movaps %xmm3,%xmm0 subps %xmm1,%xmm3 subps %xmm1,%xmm3 ## rinv-2*krsq mulps %xmm0,%xmm0 mulps nb231_qqH(%esp),%xmm2 mulps nb231_qqH(%esp),%xmm3 mulps %xmm0,%xmm3 addps nb231_vctot(%esp),%xmm2 movaps %xmm2,nb231_vctot(%esp) movaps nb231_dxH1(%esp),%xmm0 movaps nb231_dyH1(%esp),%xmm1 movaps nb231_dzH1(%esp),%xmm2 mulps %xmm3,%xmm0 mulps %xmm3,%xmm1 mulps %xmm3,%xmm2 ## update H1 forces movaps nb231_fixH1(%esp),%xmm3 movaps nb231_fiyH1(%esp),%xmm4 movaps nb231_fizH1(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb231_fixH1(%esp) movaps %xmm4,nb231_fiyH1(%esp) movaps %xmm7,nb231_fizH1(%esp) ## update j forces with water H1 addps nb231_fjx(%esp),%xmm0 addps nb231_fjy(%esp),%xmm1 addps nb231_fjz(%esp),%xmm2 movaps %xmm0,nb231_fjx(%esp) movaps %xmm1,nb231_fjy(%esp) movaps %xmm2,nb231_fjz(%esp) ## H2 interactions movaps nb231_rinvH2(%esp),%xmm2 movaps nb231_krsqH2(%esp),%xmm1 movaps %xmm2,%xmm3 addps %xmm1,%xmm2 ## rinv+krsq subps nb231_crf(%esp),%xmm2 ## rinv+krsq-crf movaps %xmm3,%xmm0 subps %xmm1,%xmm3 subps %xmm1,%xmm3 ## rinv-2*krsq mulps %xmm0,%xmm0 mulps nb231_qqH(%esp),%xmm2 mulps nb231_qqH(%esp),%xmm3 mulps %xmm0,%xmm3 addps nb231_vctot(%esp),%xmm2 movaps %xmm2,nb231_vctot(%esp) movaps nb231_dxH2(%esp),%xmm0 movaps nb231_dyH2(%esp),%xmm1 movaps nb231_dzH2(%esp),%xmm2 mulps %xmm3,%xmm0 mulps %xmm3,%xmm1 mulps %xmm3,%xmm2 ## update H2 forces movaps nb231_fixH2(%esp),%xmm3 movaps nb231_fiyH2(%esp),%xmm4 movaps nb231_fizH2(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb231_fixH2(%esp) movaps %xmm4,nb231_fiyH2(%esp) movaps %xmm7,nb231_fizH2(%esp) movl nb231_faction(%ebp),%edi ## update j forces addps nb231_fjx(%esp),%xmm0 addps nb231_fjy(%esp),%xmm1 addps nb231_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,nb231_innerk(%esp) jl _nb_kernel231_ia32_sse.nb231_odd_inner jmp _nb_kernel231_ia32_sse.nb231_unroll_loop_nb_kernel231_ia32_sse.nb231_odd_inner: addl $4,nb231_innerk(%esp) jnz _nb_kernel231_ia32_sse.nb231_odd_loop jmp _nb_kernel231_ia32_sse.nb231_updateouterdata_nb_kernel231_ia32_sse.nb231_odd_loop: movl nb231_innerjjnr(%esp),%edx ## pointer to jjnr[k] movl (%edx),%eax addl $4,nb231_innerjjnr(%esp) xorps %xmm4,%xmm4 movss nb231_iqO(%esp),%xmm4 movl nb231_charge(%ebp),%esi movhps nb231_iqH(%esp),%xmm4 movss (%esi,%eax,4),%xmm3 ## charge in xmm3 shufps $0,%xmm3,%xmm3 mulps %xmm4,%xmm3 movaps %xmm3,nb231_qqO(%esp) ## use oxygen qq for storage xorps %xmm6,%xmm6 movl nb231_type(%ebp),%esi movl (%esi,%eax,4),%ebx movl nb231_vdwparam(%ebp),%esi shll %ebx addl nb231_ntia(%esp),%ebx movlps (%esi,%ebx,4),%xmm6 movaps %xmm6,%xmm7 shufps $252,%xmm6,%xmm6 ## constant 11111100 shufps $253,%xmm7,%xmm7 ## constant 11111101 movaps %xmm6,nb231_c6(%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?