nb_kernel112_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,970 行 · 第 1/5 页
S
1,970 行
## done with i O Now do i H1 & H2 simultaneously first get i particle coords: movaps nb112_ixH1(%esp),%xmm0 movaps nb112_iyH1(%esp),%xmm1 movaps nb112_izH1(%esp),%xmm2 movaps nb112_ixH2(%esp),%xmm3 movaps nb112_iyH2(%esp),%xmm4 movaps nb112_izH2(%esp),%xmm5 subps nb112_jxO(%esp),%xmm0 subps nb112_jyO(%esp),%xmm1 subps nb112_jzO(%esp),%xmm2 subps nb112_jxO(%esp),%xmm3 subps nb112_jyO(%esp),%xmm4 subps nb112_jzO(%esp),%xmm5 movaps %xmm0,nb112_dxH1O(%esp) movaps %xmm1,nb112_dyH1O(%esp) movaps %xmm2,nb112_dzH1O(%esp) movaps %xmm3,nb112_dxH2O(%esp) movaps %xmm4,nb112_dyH2O(%esp) movaps %xmm5,nb112_dzH2O(%esp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm1,%xmm0 addps %xmm3,%xmm4 addps %xmm2,%xmm0 ## have rsqH1 in xmm0 addps %xmm5,%xmm4 ## have rsqH2 in xmm4 ## do invsqrt rsqrtps %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movaps %xmm1,%xmm2 ## do coulomb interaction movaps %xmm5,%xmm6 mulps %xmm1,%xmm1 mulps %xmm5,%xmm5 movaps nb112_three(%esp),%xmm3 movaps %xmm3,%xmm7 mulps %xmm0,%xmm1 mulps %xmm4,%xmm5 subps %xmm1,%xmm3 subps %xmm5,%xmm7 mulps %xmm2,%xmm3 mulps %xmm6,%xmm7 mulps nb112_half(%esp),%xmm3 ## rinv H1 - j water mulps nb112_half(%esp),%xmm7 ## rinv H2 - j water ## assemble charges in xmm6 xorps %xmm6,%xmm6 ## do coulomb interaction movaps %xmm3,%xmm0 movss nb112_qqOH(%esp),%xmm6 movaps %xmm7,%xmm4 movhps nb112_qqHH(%esp),%xmm6 mulps %xmm0,%xmm0 ## rinvsq mulps %xmm4,%xmm4 ## rinvsq mulps %xmm6,%xmm3 ## vcoul mulps %xmm6,%xmm7 ## vcoul movaps %xmm3,%xmm2 addps %xmm7,%xmm2 ## total vcoul mulps %xmm3,%xmm0 ## fscal addps nb112_vctot(%esp),%xmm2 mulps %xmm4,%xmm7 ## fscal movaps %xmm2,nb112_vctot(%esp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb112_dxH1O(%esp),%xmm0 mulps nb112_dyH1O(%esp),%xmm1 mulps nb112_dzH1O(%esp),%xmm2 ## update forces H1 - j water movaps nb112_fjxO(%esp),%xmm3 movaps nb112_fjyO(%esp),%xmm4 movaps nb112_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb112_fjxO(%esp) movaps %xmm4,nb112_fjyO(%esp) movaps %xmm5,nb112_fjzO(%esp) addps nb112_fixH1(%esp),%xmm0 addps nb112_fiyH1(%esp),%xmm1 addps nb112_fizH1(%esp),%xmm2 movaps %xmm0,nb112_fixH1(%esp) movaps %xmm1,nb112_fiyH1(%esp) movaps %xmm2,nb112_fizH1(%esp) ## do forces H2 - j water movaps %xmm7,%xmm0 movaps %xmm7,%xmm1 movaps %xmm7,%xmm2 mulps nb112_dxH2O(%esp),%xmm0 mulps nb112_dyH2O(%esp),%xmm1 mulps nb112_dzH2O(%esp),%xmm2 movaps nb112_fjxO(%esp),%xmm3 movaps nb112_fjyO(%esp),%xmm4 movaps nb112_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movl nb112_faction(%ebp),%esi movaps %xmm3,nb112_fjxO(%esp) movaps %xmm4,nb112_fjyO(%esp) movaps %xmm5,nb112_fjzO(%esp) addps nb112_fixH2(%esp),%xmm0 addps nb112_fiyH2(%esp),%xmm1 addps nb112_fizH2(%esp),%xmm2 movaps %xmm0,nb112_fixH2(%esp) movaps %xmm1,nb112_fiyH2(%esp) movaps %xmm2,nb112_fizH2(%esp) ## update j water forces from local variables movlps (%esi,%eax,4),%xmm0 movlps 12(%esi,%eax,4),%xmm1 movhps 24(%esi,%eax,4),%xmm1 movaps nb112_fjxO(%esp),%xmm3 movaps nb112_fjyO(%esp),%xmm4 movaps nb112_fjzO(%esp),%xmm5 movaps %xmm5,%xmm6 movaps %xmm5,%xmm7 shufps $2,%xmm6,%xmm6 ## constant 00000010 shufps $3,%xmm7,%xmm7 ## constant 00000011 addss 8(%esi,%eax,4),%xmm5 addss 20(%esi,%eax,4),%xmm6 addss 32(%esi,%eax,4),%xmm7 movss %xmm5,8(%esi,%eax,4) movss %xmm6,20(%esi,%eax,4) movss %xmm7,32(%esi,%eax,4) movaps %xmm3,%xmm5 unpcklps %xmm4,%xmm3 unpckhps %xmm4,%xmm5 addps %xmm3,%xmm0 addps %xmm5,%xmm1 movlps %xmm0,(%esi,%eax,4) movlps %xmm1,12(%esi,%eax,4) movhps %xmm1,24(%esi,%eax,4) decl nb112_innerk(%esp) jz _nb_kernel112_ia32_sse.nb112_updateouterdata jmp _nb_kernel112_ia32_sse.nb112_single_loop_nb_kernel112_ia32_sse.nb112_updateouterdata: movl nb112_ii3(%esp),%ecx movl nb112_faction(%ebp),%edi movl nb112_fshift(%ebp),%esi movl nb112_is3(%esp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb112_fixO(%esp),%xmm0 movaps nb112_fiyO(%esp),%xmm1 movaps nb112_fizO(%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) ## accumulate force in xmm6/xmm7 for fshift movaps %xmm0,%xmm6 movss %xmm2,%xmm7 movlhps %xmm1,%xmm6 shufps $8,%xmm6,%xmm6 ## constant 00001000 ## accumulate H1i forces in xmm0, xmm1, xmm2 movaps nb112_fixH1(%esp),%xmm0 movaps nb112_fiyH1(%esp),%xmm1 movaps nb112_fizH1(%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 12(%edi,%ecx,4),%xmm3 movss 16(%edi,%ecx,4),%xmm4 movss 20(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,12(%edi,%ecx,4) movss %xmm4,16(%edi,%ecx,4) movss %xmm5,20(%edi,%ecx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## constant 00001000 addps %xmm0,%xmm6 ## accumulate H2i forces in xmm0, xmm1, xmm2 movaps nb112_fixH2(%esp),%xmm0 movaps nb112_fiyH2(%esp),%xmm1 movaps nb112_fizH2(%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 24(%edi,%ecx,4),%xmm3 movss 28(%edi,%ecx,4),%xmm4 movss 32(%edi,%ecx,4),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,24(%edi,%ecx,4) movss %xmm4,28(%edi,%ecx,4) movss %xmm5,32(%edi,%ecx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## constant 00001000 addps %xmm0,%xmm6 ## increment fshift force movlps (%esi,%edx,4),%xmm3 movss 8(%esi,%edx,4),%xmm4 addps %xmm6,%xmm3 addss %xmm7,%xmm4 movlps %xmm3,(%esi,%edx,4) movss %xmm4,8(%esi,%edx,4) ## get n from stack movl nb112_n(%esp),%esi ## get group index for i particle movl nb112_gid(%ebp),%edx ## base of gid[] movl (%edx,%esi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb112_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 nb112_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 nb112_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 nb112_Vvdw(%ebp),%eax addss (%eax,%edx,4),%xmm7 ## move back to mem movss %xmm7,(%eax,%edx,4) ## finish if last movl nb112_nn1(%esp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel112_ia32_sse.nb112_outerend ## not last, iterate outer loop once more! movl %esi,nb112_n(%esp) jmp _nb_kernel112_ia32_sse.nb112_outer_nb_kernel112_ia32_sse.nb112_outerend: ## check if more outer neighborlists remain movl nb112_nri(%esp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel112_ia32_sse.nb112_end ## non-zero, do one more workunit jmp _nb_kernel112_ia32_sse.nb112_threadloop_nb_kernel112_ia32_sse.nb112_end: emms movl nb112_nouter(%esp),%eax movl nb112_ninner(%esp),%ebx movl nb112_outeriter(%ebp),%ecx movl nb112_inneriter(%ebp),%edx movl %eax,(%ecx) movl %ebx,(%edx) movl nb112_salign(%esp),%eax addl %eax,%esp addl $1512,%esp popl %edi popl %esi popl %edx popl %ecx popl %ebx popl %eax leave ret.globl nb_kernel112nf_ia32_sse.globl _nb_kernel112nf_ia32_ssenb_kernel112nf_ia32_sse: _nb_kernel112nf_ia32_sse: .set nb112nf_p_nri, 8.set nb112nf_iinr, 12.set nb112nf_jindex, 16.set nb112nf_jjnr, 20.set nb112nf_shift, 24.set nb112nf_shiftvec, 28.set nb112nf_fshift, 32.set nb112nf_gid, 36.set nb112nf_pos, 40.set nb112nf_faction, 44.set nb112nf_charge, 48.set nb112nf_p_facel, 52.set nb112nf_krf, 56.set nb112nf_crf, 60.set nb112nf_Vc, 64.set nb112nf_type, 68.set nb112nf_p_ntype, 72.set nb112nf_vdwparam, 76.set nb112nf_Vvdw, 80.set nb112nf_p_tabscale, 84.set nb112nf_VFtab, 88.set nb112nf_invsqrta, 92.set nb112nf_dvda, 96.set nb112nf_p_gbtabscale, 100.set nb112nf_GBtab, 104.set nb112nf_p_nthreads, 108.set nb112nf_count, 112.set nb112nf_mtx, 116.set nb112nf_outeriter, 120.set nb112nf_inneriter, 124.set nb112nf_work, 128 ## stack offsets for local variables ## bottom of stack is cache-aligned for sse use .set nb112nf_ixO, 0.set nb112nf_iyO, 16.set nb112nf_izO, 32.set nb112nf_ixH1, 48.set nb112nf_iyH1, 64⌨️ 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?