nb_kernel430_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,138 行 · 第 1/5 页
S
2,138 行
addps %xmm6,%xmm7 addps %xmm5,%xmm7 ## xmm7=FF mulps %xmm1,%xmm5 ## xmm5=eps*Fp addps %xmm4,%xmm5 ## xmm5=VV mulps %xmm3,%xmm5 ## vcoul=qq*VV mulps %xmm7,%xmm3 ## fijC=FF*qq ## get jnr from mm0/mm1 movd %mm0,%ecx movd %mm1,%edx movl nb430_dvda(%ebp),%esi ## Calculate dVda xorps %xmm7,%xmm7 mulps nb430_gbscale(%esp),%xmm3 movaps %xmm3,%xmm6 mulps nb430_r(%esp),%xmm6 addps %xmm5,%xmm6 addps nb430_vctot(%esp),%xmm5 movaps %xmm5,nb430_vctot(%esp) ## xmm6=(vcoul+fijC*r) subps %xmm6,%xmm7 movaps %xmm7,%xmm6 ## update dvdasum addps nb430_dvdasum(%esp),%xmm7 movaps %xmm7,nb430_dvdasum(%esp) ## update j atoms dvdaj movaps %xmm6,%xmm7 shufps $0x1,%xmm7,%xmm7 addss (%esi,%ecx,4),%xmm6 addss (%esi,%edx,4),%xmm7 movss %xmm6,(%esi,%ecx,4) movss %xmm7,(%esi,%edx,4) ## put scalar force on stack temporarily movaps %xmm3,nb430_fscal(%esp) movaps nb430_r(%esp),%xmm4 mulps nb430_tsc(%esp),%xmm4 cvttps2pi %xmm4,%mm6 cvtpi2ps %mm6,%xmm6 subps %xmm6,%xmm4 movaps %xmm4,%xmm1 ## xmm1=eps movaps %xmm1,%xmm2 mulps %xmm2,%xmm2 ## xmm2=eps2 pslld $3,%mm6 movl nb430_VFtab(%ebp),%esi movd %mm6,%ecx psrlq $32,%mm6 movd %mm6,%edx ## dispersion movaps (%esi,%ecx,4),%xmm4 movaps (%esi,%edx,4),%xmm7 ## transpose, using xmm3 for scratch movaps %xmm4,%xmm6 unpcklps %xmm7,%xmm4 ## Y1 Y2 F1 F2 unpckhps %xmm7,%xmm6 ## G1 G2 H1 H2 movhlps %xmm4,%xmm5 ## F1 F2 movhlps %xmm6,%xmm7 ## H1 H2 ## 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 nb430_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 nb430_c6(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijD mulps %xmm4,%xmm5 ## Vvdw6 mulps nb430_tsc(%esp),%xmm7 addps nb430_fscal(%esp),%xmm7 ## add to fscal ## put scalar force on stack Update Vvdwtot directly addps nb430_Vvdwtot(%esp),%xmm5 movaps %xmm7,nb430_fscal(%esp) movaps %xmm5,nb430_Vvdwtot(%esp) ## repulsion movaps 16(%esi,%ecx,4),%xmm4 movaps 16(%esi,%edx,4),%xmm7 ## transpose, using xmm3 for scratch movaps %xmm4,%xmm6 unpcklps %xmm7,%xmm4 ## Y1 Y2 F1 F2 unpckhps %xmm7,%xmm6 ## G1 G2 H1 H2 movhlps %xmm4,%xmm5 ## F1 F2 movhlps %xmm6,%xmm7 ## H1 H2 ## table ready, in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb430_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 nb430_c12(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijR mulps %xmm4,%xmm5 ## Vvdw12 mulps nb430_tsc(%esp),%xmm7 addps nb430_fscal(%esp),%xmm7 addps nb430_Vvdwtot(%esp),%xmm5 movaps %xmm5,nb430_Vvdwtot(%esp) xorps %xmm4,%xmm4 mulps %xmm0,%xmm7 subps %xmm7,%xmm4 movaps nb430_dx(%esp),%xmm0 movaps nb430_dy(%esp),%xmm1 movaps nb430_dz(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movaps nb430_fix(%esp),%xmm3 movaps nb430_fiy(%esp),%xmm4 movaps nb430_fiz(%esp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb430_fix(%esp) movaps %xmm4,nb430_fiy(%esp) movaps %xmm5,nb430_fiz(%esp) ## update the fj's movss (%edi,%eax,4),%xmm3 movss 4(%edi,%eax,4),%xmm4 movss 8(%edi,%eax,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,(%edi,%eax,4) movss %xmm4,4(%edi,%eax,4) movss %xmm5,8(%edi,%eax,4) shufps $225,%xmm0,%xmm0 ## constant 11100001 shufps $225,%xmm1,%xmm1 ## constant 11100001 shufps $225,%xmm2,%xmm2 ## constant 11100001 movss (%edi,%ebx,4),%xmm3 movss 4(%edi,%ebx,4),%xmm4 movss 8(%edi,%ebx,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,(%edi,%ebx,4) movss %xmm4,4(%edi,%ebx,4) movss %xmm5,8(%edi,%ebx,4)_nb_kernel430_ia32_sse.nb430_checksingle: movl nb430_innerk(%esp),%edx andl $1,%edx jnz _nb_kernel430_ia32_sse.nb430_dosingle jmp _nb_kernel430_ia32_sse.nb430_updateouterdata_nb_kernel430_ia32_sse.nb430_dosingle: movl nb430_charge(%ebp),%esi movl nb430_invsqrta(%ebp),%edx movl nb430_pos(%ebp),%edi movl nb430_innerjjnr(%esp),%ecx movl (%ecx),%eax xorps %xmm2,%xmm2 movaps %xmm2,%xmm6 movss (%edx,%eax,4),%xmm2 ## isaj mulss nb430_isai(%esp),%xmm2 movss %xmm2,nb430_isaprod(%esp) movss %xmm2,%xmm1 mulss nb430_gbtsc(%esp),%xmm1 movss %xmm1,nb430_gbscale(%esp) mulss nb430_iq(%esp),%xmm2 movss (%esi,%eax,4),%xmm6 ## xmm6(0) has the charge mulss %xmm2,%xmm6 movss %xmm6,nb430_qq(%esp) movl nb430_type(%ebp),%esi movl %eax,%ecx movl (%esi,%ecx,4),%ecx movl nb430_vdwparam(%ebp),%esi shll %ecx addl nb430_ntia(%esp),%ecx movlps (%esi,%ecx,4),%xmm6 movaps %xmm6,%xmm4 shufps $252,%xmm4,%xmm4 ## constant 11111100 shufps $253,%xmm6,%xmm6 ## constant 11111101 movss %xmm4,nb430_c6(%esp) movss %xmm6,nb430_c12(%esp) movd %eax,%mm0 leal (%eax,%eax,2),%eax ## move coordinates to xmm0-xmm2 movss (%edi,%eax,4),%xmm0 movss 4(%edi,%eax,4),%xmm1 movss 8(%edi,%eax,4),%xmm2 movss nb430_ix(%esp),%xmm4 movss nb430_iy(%esp),%xmm5 movss nb430_iz(%esp),%xmm6 ## calc dr subss %xmm0,%xmm4 subss %xmm1,%xmm5 subss %xmm2,%xmm6 ## store dr movaps %xmm4,nb430_dx(%esp) movaps %xmm5,nb430_dy(%esp) movaps %xmm6,nb430_dz(%esp) ## square it mulss %xmm4,%xmm4 mulss %xmm5,%xmm5 mulss %xmm6,%xmm6 addss %xmm5,%xmm4 addss %xmm6,%xmm4 ## rsq in xmm4 rsqrtss %xmm4,%xmm5 ## lookup seed in xmm5 movaps %xmm5,%xmm2 mulss %xmm5,%xmm5 movss nb430_three(%esp),%xmm1 mulss %xmm4,%xmm5 ## rsq*lu*lu movss nb430_half(%esp),%xmm0 subss %xmm5,%xmm1 ## constant 30-rsq*lu*lu mulss %xmm2,%xmm1 mulss %xmm1,%xmm0 ## xmm0=rinv mulss %xmm0,%xmm4 ## xmm4=r movss %xmm4,nb430_r(%esp) mulss nb430_gbscale(%esp),%xmm4 cvttss2si %xmm4,%ebx ## mm6 contain lu indices cvtsi2ss %ebx,%xmm6 subss %xmm6,%xmm4 movaps %xmm4,%xmm1 ## xmm1=eps movaps %xmm1,%xmm2 mulss %xmm2,%xmm2 ## xmm2=eps2 shll $2,%ebx movl nb430_GBtab(%ebp),%esi movaps (%esi,%ebx,4),%xmm4 movhlps %xmm4,%xmm6 movaps %xmm4,%xmm5 movaps %xmm6,%xmm7 shufps $1,%xmm5,%xmm5 shufps $1,%xmm7,%xmm7 ## table ready in xmm4-xmm7 mulss %xmm1,%xmm6 ## xmm6=Geps mulss %xmm2,%xmm7 ## xmm7=Heps2 addss %xmm6,%xmm5 addss %xmm7,%xmm5 ## xmm5=Fp mulss nb430_two(%esp),%xmm7 ## two*Heps2 movss nb430_qq(%esp),%xmm3 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV mulss %xmm3,%xmm5 ## vcoul=qq*VV mulss %xmm7,%xmm3 ## fijC=FF*qq movd %mm0,%ebx movl nb430_dvda(%ebp),%esi ## Calculate dVda xorps %xmm7,%xmm7 mulss nb430_gbscale(%esp),%xmm3 movaps %xmm3,%xmm6 mulss nb430_r(%esp),%xmm6 addss %xmm5,%xmm6 addss nb430_vctot(%esp),%xmm5 movss %xmm5,nb430_vctot(%esp) ## xmm6=(vcoul+fijC*r) subss %xmm6,%xmm7 movaps %xmm7,%xmm6 ## update dvdasum addss nb430_dvdasum(%esp),%xmm7 movaps %xmm7,nb430_dvdasum(%esp) ## update j atoms dvdaj addss (%esi,%ebx,4),%xmm6 movss %xmm6,(%esi,%ebx,4) ## put scalar force on stack temporarily movss %xmm3,nb430_fscal(%esp) movss nb430_r(%esp),%xmm4 mulps nb430_tsc(%esp),%xmm4 cvttss2si %xmm4,%ebx cvtsi2ss %ebx,%xmm6 subss %xmm6,%xmm4 movss %xmm4,%xmm1 ## xmm1=eps movss %xmm1,%xmm2 mulss %xmm2,%xmm2 ## xmm2=eps2 shll $3,%ebx movl nb430_VFtab(%ebp),%esi ## dispersion movaps (%esi,%ebx,4),%xmm4 movhlps %xmm4,%xmm6 movaps %xmm4,%xmm5 movaps %xmm6,%xmm7 shufps $1,%xmm5,%xmm5 shufps $1,%xmm7,%xmm7 ## table ready in xmm4-xmm7 mulss %xmm1,%xmm6 ## xmm6=Geps mulss %xmm2,%xmm7 ## xmm7=Heps2 addss %xmm6,%xmm5 addss %xmm7,%xmm5 ## xmm5=Fp mulss nb430_two(%esp),%xmm7 ## two*Heps2 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV movss nb430_c6(%esp),%xmm4 mulss %xmm4,%xmm7 ## fijD mulss %xmm4,%xmm5 ## Vvdw6 mulps nb430_tsc(%esp),%xmm7 addss nb430_fscal(%esp),%xmm7 ## add to fscal ## put scalar force on stack Update Vvdwtot directly addss nb430_Vvdwtot(%esp),%xmm5 movss %xmm7,nb430_fscal(%esp) movss %xmm5,nb430_Vvdwtot(%esp) ## repulsion movaps 16(%esi,%ebx,4),%xmm4 movhlps %xmm4,%xmm6 movaps %xmm4,%xmm5 movaps %xmm6,%xmm7 shufps $1,%xmm5,%xmm5 shufps $1,%xmm7,%xmm7 ## table ready in xmm4-xmm7 mulss %xmm1,%xmm6 ## xmm6=Geps mulss %xmm2,%xmm7 ## xmm7=Heps2 addss %xmm6,%xmm5 addss %xmm7,%xmm5 ## xmm5=Fp mulss nb430_two(%esp),%xmm7 ## two*Heps2 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV movss nb430_c12(%esp),%xmm4 mulss %xmm4,%xmm7 ## fijR mulss %xmm4,%xmm5 ## Vvdw12 mulps nb430_tsc(%esp),%xmm7 addss nb430_fscal(%esp),%xmm7 addss nb430_Vvdwtot(%esp),%xmm5 movss %xmm5,nb430_Vvdwtot(%esp) xorps %xmm4,%xmm4 mulss %xmm0,%xmm7 subss %xmm7,%xmm4 movl nb430_faction(%ebp),%edi movss nb430_dx(%esp),%xmm0 movss nb430_dy(%esp),%xmm1 movss nb430_dz(%esp),%xmm2 mulss %xmm4,%xmm0 mulss %xmm4,%xmm1 mulss %xmm4,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movss nb430_fix(%esp),%xmm3 movss nb430_fiy(%esp),%xmm4 movss nb430_fiz(%esp),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,nb430_fix(%esp) movss %xmm4,nb430_fiy(%esp) movss %xmm5,nb430_fiz(%esp) ## update fj movss (%edi,%eax,4),%xmm3 movss 4(%edi,%eax,4),%xmm4 movss 8(%edi,%eax,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,(%edi,%eax,4) movss %xmm4,4(%edi,%eax,4) movss %xmm5,8(%edi,%eax,4)_nb_kernel430_ia32_sse.nb430_updateouterdata: movl nb430_ii3(%esp),%ecx movl nb430_faction(%ebp),%edi movl nb430_fshift(%ebp),%esi movl nb430_is3(%esp),%edx ## accumulate i forces in xmm0, xmm1, xmm2 movaps nb430_fix(%esp),%xmm0 movaps nb430_fiy(%esp),%xmm1 movaps nb430_fiz(%esp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1⌨️ 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?