nb_kernel330_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,062 行 · 第 1/5 页
S
2,062 行
## now update f_i movaps nb330_fix(%esp),%xmm3 movaps nb330_fiy(%esp),%xmm4 movaps nb330_fiz(%esp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb330_fix(%esp) movaps %xmm4,nb330_fiy(%esp) movaps %xmm5,nb330_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_kernel330_ia32_sse.nb330_checksingle: movl nb330_innerk(%esp),%edx andl $1,%edx jnz _nb_kernel330_ia32_sse.nb330_dosingle jmp _nb_kernel330_ia32_sse.nb330_updateouterdata_nb_kernel330_ia32_sse.nb330_dosingle: movl nb330_charge(%ebp),%esi movl nb330_pos(%ebp),%edi movl nb330_innerjjnr(%esp),%ecx movl (%ecx),%eax xorps %xmm6,%xmm6 movss (%esi,%eax,4),%xmm6 ## xmm6(0) has the charge mulps nb330_iq(%esp),%xmm6 movaps %xmm6,nb330_qq(%esp) movl nb330_type(%ebp),%esi movl %eax,%ecx movl (%esi,%ecx,4),%ecx movl nb330_vdwparam(%ebp),%esi shll %ecx addl nb330_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,nb330_c6(%esp) movaps %xmm6,nb330_c12(%esp) 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 nb330_ix(%esp),%xmm4 movaps nb330_iy(%esp),%xmm5 movaps nb330_iz(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb330_dx(%esp) movaps %xmm5,nb330_dy(%esp) movaps %xmm6,nb330_dz(%esp) ## square it mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 mulps %xmm6,%xmm6 addps %xmm5,%xmm4 addps %xmm6,%xmm4 ## rsq in xmm4 rsqrtps %xmm4,%xmm5 ## lookup seed in xmm5 movaps %xmm5,%xmm2 mulps %xmm5,%xmm5 movaps nb330_three(%esp),%xmm1 mulps %xmm4,%xmm5 ## rsq*lu*lu movaps nb330_half(%esp),%xmm0 subps %xmm5,%xmm1 ## constant 30-rsq*lu*lu mulps %xmm2,%xmm1 mulps %xmm1,%xmm0 ## xmm0=rinv mulps %xmm0,%xmm4 ## xmm4=r mulps nb330_tsc(%esp),%xmm4 cvttps2pi %xmm4,%mm6 ## mm6 contain lu indices cvtpi2ps %mm6,%xmm6 subps %xmm6,%xmm4 movaps %xmm4,%xmm1 ## xmm1=eps movaps %xmm1,%xmm2 mulps %xmm2,%xmm2 ## xmm2=eps2 pslld $2,%mm6 movl nb330_VFtab(%ebp),%esi movd %mm6,%ebx leal (%ebx,%ebx,2),%ebx movlps (%esi,%ebx,4),%xmm4 movlps 8(%esi,%ebx,4),%xmm6 movaps %xmm4,%xmm5 movaps %xmm6,%xmm7 shufps $1,%xmm5,%xmm5 shufps $1,%xmm7,%xmm7 ## table ready in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb330_two(%esp),%xmm7 ## two*Heps2 movaps nb330_qq(%esp),%xmm3 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 ## at this point mm5 contains vcoul and mm3 fijC ## increment vcoul - then we can get rid of mm5 ## update vctot addss nb330_vctot(%esp),%xmm5 movss %xmm5,nb330_vctot(%esp) ## put scalar force on stack temporarily movaps %xmm3,nb330_fscal(%esp) ## dispersion movlps 16(%esi,%ebx,4),%xmm4 movlps 24(%esi,%ebx,4),%xmm6 movaps %xmm4,%xmm5 movaps %xmm6,%xmm7 shufps $1,%xmm5,%xmm5 shufps $1,%xmm7,%xmm7 ## table ready in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb330_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 nb330_c6(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijD mulps %xmm4,%xmm5 ## Vvdw6 addps nb330_fscal(%esp),%xmm7 ## add to fscal ## put scalar force on stack Update Vvdwtot directly addss nb330_Vvdwtot(%esp),%xmm5 movaps %xmm7,nb330_fscal(%esp) movss %xmm5,nb330_Vvdwtot(%esp) ## repulsion movlps 32(%esi,%ebx,4),%xmm4 movlps 40(%esi,%ebx,4),%xmm6 movaps %xmm4,%xmm5 movaps %xmm6,%xmm7 shufps $1,%xmm5,%xmm5 shufps $1,%xmm7,%xmm7 ## table ready in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb330_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 nb330_c12(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijR mulps %xmm4,%xmm5 ## Vvdw12 addps nb330_fscal(%esp),%xmm7 addss nb330_Vvdwtot(%esp),%xmm5 movss %xmm5,nb330_Vvdwtot(%esp) xorps %xmm4,%xmm4 mulps nb330_tsc(%esp),%xmm7 mulps %xmm0,%xmm7 subps %xmm7,%xmm4 movl nb330_faction(%ebp),%edi movaps nb330_dx(%esp),%xmm0 movaps nb330_dy(%esp),%xmm1 movaps nb330_dz(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movaps nb330_fix(%esp),%xmm3 movaps nb330_fiy(%esp),%xmm4 movaps nb330_fiz(%esp),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movaps %xmm3,nb330_fix(%esp) movaps %xmm4,nb330_fiy(%esp) movaps %xmm5,nb330_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_kernel330_ia32_sse.nb330_updateouterdata: movl nb330_ii3(%esp),%ecx movl nb330_faction(%ebp),%edi movl nb330_fshift(%ebp),%esi movl nb330_is3(%esp),%edx ## accumulate i forces in xmm0, xmm1, xmm2 movaps nb330_fix(%esp),%xmm0 movaps nb330_fiy(%esp),%xmm1 movaps nb330_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 nb330_n(%esp),%esi ## get group index for i particle movl nb330_gid(%ebp),%edx ## base of gid[] movl (%edx,%esi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb330_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 nb330_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 nb330_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 nb330_Vvdw(%ebp),%eax addss (%eax,%edx,4),%xmm7 ## move back to mem movss %xmm7,(%eax,%edx,4) ## finish if last movl nb330_nn1(%esp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel330_ia32_sse.nb330_outerend ## not last, iterate outer loop once more! movl %esi,nb330_n(%esp) jmp _nb_kernel330_ia32_sse.nb330_outer_nb_kernel330_ia32_sse.nb330_outerend: ## check if more outer neighborlists remain movl nb330_nri(%esp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel330_ia32_sse.nb330_end ## non-zero, do one more workunit jmp _nb_kernel330_ia32_sse.nb330_threadloop_nb_kernel330_ia32_sse.nb330_end: emms movl nb330_nouter(%esp),%eax movl nb330_ninner(%esp),%ebx movl nb330_outeriter(%ebp),%ecx movl nb330_inneriter(%ebp),%edx movl %eax,(%ecx) movl %ebx,(%edx) movl nb330_salign(%esp),%eax addl %eax,%esp addl $372,%esp popl %edi popl %esi popl %edx popl %ecx popl %ebx popl %eax leave ret.globl nb_kernel330nf_ia32_sse.globl _nb_kernel330nf_ia32_ssenb_kernel330nf_ia32_sse: _nb_kernel330nf_ia32_sse: .set nb330nf_p_nri, 8.set nb330nf_iinr, 12.set nb330nf_jindex, 16.set nb330nf_jjnr, 20.set nb330nf_shift, 24.set nb330nf_shiftvec, 28.set nb330nf_fshift, 32.set nb330nf_gid, 36.set nb330nf_pos, 40.set nb330nf_faction, 44.set nb330nf_charge, 48.set nb330nf_p_facel, 52.set nb330nf_argkrf, 56.set nb330nf_argcrf, 60.set nb330nf_Vc, 64.set nb330nf_type, 68.set nb330nf_p_ntype, 72.set nb330nf_vdwparam, 76.set nb330nf_Vvdw, 80.set nb330nf_p_tabscale, 84.set nb330nf_VFtab, 88.set nb330nf_invsqrta, 92.set nb330nf_dvda, 96.set nb330nf_p_gbtabscale, 100.set nb330nf_GBtab, 104.set nb330nf_p_nthreads, 108.set nb330nf_count, 112.set nb330nf_mtx, 116.set nb330nf_outeriter, 120.set nb330nf_inneriter, 124.set nb330nf_work, 128 ## stack offsets for local variables ## bottom of stack is cache-aligned for sse use .set nb330nf_ix, 0.set nb330nf_iy, 16⌨️ 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?