nb_kernel410_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,024 行 · 第 1/5 页
S
2,024 行
## increment vcoul - then we can get rid of mm5 ## update vctot movaps %xmm4,%xmm6 mulps %xmm4,%xmm6 mulps %xmm4,%xmm6 ## xmm6=rinvsix movaps %xmm6,%xmm4 mulps %xmm4,%xmm4 ## xmm4=rinvtwelve mulps nb410_c6(%esp),%xmm6 mulps nb410_c12(%esp),%xmm4 movaps nb410_Vvdwtot(%esp),%xmm7 addps %xmm4,%xmm7 mulps nb410_twelve(%esp),%xmm4 subps %xmm6,%xmm7 mulps nb410_six(%esp),%xmm6 movaps %xmm7,nb410_Vvdwtot(%esp) subps %xmm6,%xmm4 mulps %xmm0,%xmm4 subps %xmm3,%xmm4 mulps %xmm0,%xmm4 movaps nb410_dx(%esp),%xmm0 movaps nb410_dy(%esp),%xmm1 movaps nb410_dz(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movaps nb410_fix(%esp),%xmm3 movaps nb410_fiy(%esp),%xmm4 movaps nb410_fiz(%esp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb410_fix(%esp) movaps %xmm4,nb410_fiy(%esp) movaps %xmm5,nb410_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_kernel410_ia32_sse.nb410_checksingle: movl nb410_innerk(%esp),%edx andl $1,%edx jnz _nb_kernel410_ia32_sse.nb410_dosingle jmp _nb_kernel410_ia32_sse.nb410_updateouterdata_nb_kernel410_ia32_sse.nb410_dosingle: movl nb410_charge(%ebp),%esi movl nb410_invsqrta(%ebp),%edx movl nb410_pos(%ebp),%edi movl nb410_innerjjnr(%esp),%ecx movl (%ecx),%eax xorps %xmm2,%xmm2 movaps %xmm2,%xmm6 movss (%edx,%eax,4),%xmm2 ## isaj mulss nb410_isai(%esp),%xmm2 movss %xmm2,nb410_isaprod(%esp) movss %xmm2,%xmm1 mulss nb410_gbtsc(%esp),%xmm1 movss %xmm1,nb410_gbscale(%esp) mulss nb410_iq(%esp),%xmm2 movss (%esi,%eax,4),%xmm6 ## xmm6(0) has the charge mulss %xmm2,%xmm6 movss %xmm6,nb410_qq(%esp) movl nb410_type(%ebp),%esi movl %eax,%ecx movl (%esi,%ecx,4),%ecx movl nb410_vdwparam(%ebp),%esi shll %ecx addl nb410_ntia(%esp),%ecx movlps (%esi,%ecx,4),%xmm6 movaps %xmm6,%xmm4 shufps $252,%xmm4,%xmm4 ## constant 11111100 shufps $253,%xmm6,%xmm6 ## constant 11111101 movaps %xmm4,nb410_c6(%esp) movaps %xmm6,nb410_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 movaps nb410_ix(%esp),%xmm4 movaps nb410_iy(%esp),%xmm5 movaps nb410_iz(%esp),%xmm6 ## calc dr subss %xmm0,%xmm4 subss %xmm1,%xmm5 subss %xmm2,%xmm6 ## store dr movss %xmm4,nb410_dx(%esp) movss %xmm5,nb410_dy(%esp) movss %xmm6,nb410_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 nb410_three(%esp),%xmm1 mulss %xmm4,%xmm5 ## rsq*lu*lu movss nb410_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,nb410_r(%esp) mulss nb410_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 nb410_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 nb410_two(%esp),%xmm7 ## two*Heps2 movss nb410_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 nb410_dvda(%ebp),%esi ## Calculate dVda xorps %xmm7,%xmm7 mulss nb410_gbscale(%esp),%xmm3 movaps %xmm3,%xmm6 mulss nb410_r(%esp),%xmm6 addss %xmm5,%xmm6 addss nb410_vctot(%esp),%xmm5 movss %xmm5,nb410_vctot(%esp) ## xmm6=(vcoul+fijC*r) subps %xmm6,%xmm7 movaps %xmm7,%xmm6 ## update dvdasum addps nb410_dvdasum(%esp),%xmm7 movaps %xmm7,nb410_dvdasum(%esp) ## update j atoms dvdaj addss (%esi,%ebx,4),%xmm6 movss %xmm6,(%esi,%ebx,4) ## L-J movaps %xmm0,%xmm4 mulss %xmm0,%xmm4 ## xmm4=rinvsq movaps %xmm4,%xmm6 mulss %xmm4,%xmm6 mulss %xmm4,%xmm6 ## xmm6=rinvsix movaps %xmm6,%xmm4 mulss %xmm4,%xmm4 ## xmm4=rinvtwelve mulss nb410_c6(%esp),%xmm6 mulss nb410_c12(%esp),%xmm4 movss nb410_Vvdwtot(%esp),%xmm7 addss %xmm4,%xmm7 mulss nb410_twelve(%esp),%xmm4 subss %xmm6,%xmm7 mulss nb410_six(%esp),%xmm6 movss %xmm7,nb410_Vvdwtot(%esp) subss %xmm6,%xmm4 mulss %xmm0,%xmm4 subss %xmm3,%xmm4 mulss %xmm0,%xmm4 movss nb410_dx(%esp),%xmm0 movss nb410_dy(%esp),%xmm1 movss nb410_dz(%esp),%xmm2 movl nb410_faction(%ebp),%edi mulss %xmm4,%xmm0 mulss %xmm4,%xmm1 mulss %xmm4,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movss nb410_fix(%esp),%xmm3 movss nb410_fiy(%esp),%xmm4 movss nb410_fiz(%esp),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,nb410_fix(%esp) movss %xmm4,nb410_fiy(%esp) movss %xmm5,nb410_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_kernel410_ia32_sse.nb410_updateouterdata: movl nb410_ii3(%esp),%ecx movl nb410_faction(%ebp),%edi movl nb410_fshift(%ebp),%esi movl nb410_is3(%esp),%edx ## accumulate i forces in xmm0, xmm1, xmm2 movaps nb410_fix(%esp),%xmm0 movaps nb410_fiy(%esp),%xmm1 movaps nb410_fiz(%esp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss (%edi,%ecx,4),%xmm3 movss 4(%edi,%ecx,4),%xmm4 movss 8(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,(%edi,%ecx,4) movss %xmm4,4(%edi,%ecx,4) movss %xmm5,8(%edi,%ecx,4) ## increment fshift force movss (%esi,%edx,4),%xmm3 movss 4(%esi,%edx,4),%xmm4 movss 8(%esi,%edx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,(%esi,%edx,4) movss %xmm4,4(%esi,%edx,4) movss %xmm5,8(%esi,%edx,4) ## get n from stack movl nb410_n(%esp),%esi ## get group index for i particle movl nb410_gid(%ebp),%edx ## base of gid[] movl (%edx,%esi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb410_vctot(%esp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addps %xmm6,%xmm7 ## pos 0-1 in xmm7 have the sum now movaps %xmm7,%xmm6 shufps $1,%xmm6,%xmm6 addss %xmm6,%xmm7 ## add earlier value from mem movl nb410_Vc(%ebp),%eax addss (%eax,%edx,4),%xmm7 ## move back to mem movss %xmm7,(%eax,%edx,4) ## accumulate total lj energy and update it movaps nb410_Vvdwtot(%esp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addps %xmm6,%xmm7 ## pos 0-1 in xmm7 have the sum now movaps %xmm7,%xmm6 shufps $1,%xmm6,%xmm6 addss %xmm6,%xmm7 ## add earlier value from mem movl nb410_Vvdw(%ebp),%eax addss (%eax,%edx,4),%xmm7 ## move back to mem movss %xmm7,(%eax,%edx,4) ## accumulate dVda and update it movaps nb410_dvdasum(%esp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addps %xmm6,%xmm7 ## pos 0-1 in xmm7 have the sum now movaps %xmm7,%xmm6 shufps $1,%xmm6,%xmm6 addss %xmm6,%xmm7 movl nb410_ii(%esp),%edx movl nb410_dvda(%ebp),%eax addss (%eax,%edx,4),%xmm7 movss %xmm7,(%eax,%edx,4) ## finish if last movl nb410_nn1(%esp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel410_ia32_sse.nb410_outerend ## not last, iterate outer loop once more! movl %esi,nb410_n(%esp) jmp _nb_kernel410_ia32_sse.nb410_outer_nb_kernel410_ia32_sse.nb410_outerend: ## check if more outer neighborlists remain movl nb410_nri(%esp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel410_ia32_sse.nb410_end ## non-zero, do one more workunit jmp _nb_kernel410_ia32_sse.nb410_threadloop_nb_kernel410_ia32_sse.nb410_end: emms movl nb410_nouter(%esp),%eax movl nb410_ninner(%esp),%ebx movl nb410_outeriter(%ebp),%ecx movl nb410_inneriter(%ebp),%edx movl %eax,(%ecx) movl %ebx,(%edx) movl nb410_salign(%esp),%eax addl %eax,%esp addl $504,%esp popl %edi popl %esi popl %edx popl %ecx popl %ebx popl %eax leave ret.globl nb_kernel410nf_ia32_sse.globl _nb_kernel410nf_ia32_ssenb_kernel410nf_ia32_sse: _nb_kernel410nf_ia32_sse: .set nb410nf_p_nri, 8.set nb410nf_iinr, 12⌨️ 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?