nb_kernel100_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,206 行 · 第 1/3 页
S
1,206 行
movss (%esi,%eax,4),%xmm3 movss (%esi,%ebx,4),%xmm6 shufps $0,%xmm6,%xmm3 shufps $8,%xmm3,%xmm3 ## constant 00001000 ;# xmm3(0,1) has the charges leal (%eax,%eax,2),%eax leal (%ebx,%ebx,2),%ebx ## move coordinates to xmm0-xmm2 movlps (%edi,%eax,4),%xmm1 movss 8(%edi,%eax,4),%xmm2 movhps (%edi,%ebx,4),%xmm1 movss 8(%edi,%ebx,4),%xmm0 mulps nb100_iq(%esp),%xmm3 xorps %xmm7,%xmm7 movlhps %xmm7,%xmm3 shufps $0,%xmm0,%xmm2 movaps %xmm1,%xmm0 shufps $136,%xmm2,%xmm2 ## constant 10001000 shufps $136,%xmm0,%xmm0 ## constant 10001000 shufps $221,%xmm1,%xmm1 ## constant 11011101 movl nb100_faction(%ebp),%edi ## move nb100_ix-iz to xmm4-xmm6 xorps %xmm7,%xmm7 movaps nb100_ix(%esp),%xmm4 movaps nb100_iy(%esp),%xmm5 movaps nb100_iz(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb100_dx(%esp) movaps %xmm5,nb100_dy(%esp) movaps %xmm6,nb100_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 nb100_three(%esp),%xmm1 mulps %xmm4,%xmm5 ## rsq*lu*lu movaps nb100_half(%esp),%xmm0 subps %xmm5,%xmm1 ## constant 30-rsq*lu*lu mulps %xmm2,%xmm1 mulps %xmm1,%xmm0 ## xmm0=rinv movaps %xmm0,%xmm4 mulps %xmm4,%xmm4 ## xmm4=rinvsq movaps nb100_vctot(%esp),%xmm5 mulps %xmm0,%xmm3 ## xmm3=vcoul mulps %xmm3,%xmm4 ## xmm4=fscal addps %xmm3,%xmm5 movaps nb100_dx(%esp),%xmm0 movaps nb100_dy(%esp),%xmm1 movaps nb100_dz(%esp),%xmm2 movaps %xmm5,nb100_vctot(%esp) mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movaps nb100_fix(%esp),%xmm3 movaps nb100_fiy(%esp),%xmm4 movaps nb100_fiz(%esp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb100_fix(%esp) movaps %xmm4,nb100_fiy(%esp) movaps %xmm5,nb100_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_kernel100_ia32_sse.nb100_checksingle: movl nb100_innerk(%esp),%edx andl $1,%edx jnz _nb_kernel100_ia32_sse.nb100_dosingle jmp _nb_kernel100_ia32_sse.nb100_updateouterdata_nb_kernel100_ia32_sse.nb100_dosingle: movl nb100_charge(%ebp),%esi movl nb100_pos(%ebp),%edi movl nb100_innerjjnr(%esp),%ecx movl (%ecx),%eax movss (%esi,%eax,4),%xmm3 ## xmm3(0) has the charge 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 mulps nb100_iq(%esp),%xmm3 xorps %xmm7,%xmm7 movaps nb100_ix(%esp),%xmm4 movaps nb100_iy(%esp),%xmm5 movaps nb100_iz(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb100_dx(%esp) movaps %xmm5,nb100_dy(%esp) movaps %xmm6,nb100_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 nb100_three(%esp),%xmm1 mulps %xmm4,%xmm5 ## rsq*lu*lu movaps nb100_half(%esp),%xmm0 subps %xmm5,%xmm1 ## constant 30-rsq*lu*lu mulps %xmm2,%xmm1 mulps %xmm1,%xmm0 ## xmm0=rinv movaps %xmm0,%xmm4 mulps %xmm4,%xmm4 ## xmm4=rinvsq movl nb100_faction(%ebp),%edi movaps nb100_vctot(%esp),%xmm5 mulps %xmm0,%xmm3 ## xmm3=vcoul mulps %xmm3,%xmm4 ## xmm4=fscal addss %xmm3,%xmm5 movaps nb100_dx(%esp),%xmm0 movaps nb100_dy(%esp),%xmm1 movaps nb100_dz(%esp),%xmm2 movaps %xmm5,nb100_vctot(%esp) mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movaps nb100_fix(%esp),%xmm3 movaps nb100_fiy(%esp),%xmm4 movaps nb100_fiz(%esp),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movaps %xmm3,nb100_fix(%esp) movaps %xmm4,nb100_fiy(%esp) movaps %xmm5,nb100_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_kernel100_ia32_sse.nb100_updateouterdata: movl nb100_ii3(%esp),%ecx movl nb100_faction(%ebp),%edi movl nb100_fshift(%ebp),%esi movl nb100_is3(%esp),%edx ## accumulate i forces in xmm0, xmm1, xmm2 movaps nb100_fix(%esp),%xmm0 movaps nb100_fiy(%esp),%xmm1 movaps nb100_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 nb100_n(%esp),%esi ## get group index for i particle movl nb100_gid(%ebp),%edx ## base of gid[] movl (%edx,%esi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb100_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 nb100_Vc(%ebp),%eax addss (%eax,%edx,4),%xmm7 ## move back to mem movss %xmm7,(%eax,%edx,4) ## finish if last movl nb100_nn1(%esp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel100_ia32_sse.nb100_outerend ## not last, iterate outer loop once more! movl %esi,nb100_n(%esp) jmp _nb_kernel100_ia32_sse.nb100_outer_nb_kernel100_ia32_sse.nb100_outerend: ## check if more outer neighborlists remain movl nb100_nri(%esp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel100_ia32_sse.nb100_end ## non-zero, do one more workunit jmp _nb_kernel100_ia32_sse.nb100_threadloop_nb_kernel100_ia32_sse.nb100_end: emms movl nb100_nouter(%esp),%eax movl nb100_ninner(%esp),%ebx movl nb100_outeriter(%ebp),%ecx movl nb100_inneriter(%ebp),%edx movl %eax,(%ecx) movl %ebx,(%edx) movl nb100_salign(%esp),%eax addl %eax,%esp addl $252,%esp popl %edi popl %esi popl %edx popl %ecx popl %ebx popl %eax leave ret.globl nb_kernel100nf_ia32_sse.globl _nb_kernel100nf_ia32_ssenb_kernel100nf_ia32_sse: _nb_kernel100nf_ia32_sse: .set nb100nf_p_nri, 8.set nb100nf_iinr, 12.set nb100nf_jindex, 16.set nb100nf_jjnr, 20.set nb100nf_shift, 24.set nb100nf_shiftvec, 28.set nb100nf_fshift, 32.set nb100nf_gid, 36.set nb100nf_pos, 40.set nb100nf_faction, 44.set nb100nf_charge, 48.set nb100nf_p_facel, 52.set nb100nf_p_krf, 56.set nb100nf_p_crf, 60.set nb100nf_Vc, 64.set nb100nf_type, 68.set nb100nf_p_ntype, 72.set nb100nf_vdwparam, 76.set nb100nf_Vvdw, 80.set nb100nf_p_tabscale, 84.set nb100nf_VFtab, 88.set nb100nf_invsqrta, 92.set nb100nf_dvda, 96.set nb100nf_p_gbtabscale, 100.set nb100nf_GBtab, 104.set nb100nf_p_nthreads, 108.set nb100nf_count, 112.set nb100nf_mtx, 116.set nb100nf_outeriter, 120.set nb100nf_inneriter, 124.set nb100nf_work, 128 ## stack offsets for local variables ## bottom of stack is cache-aligned for sse use .set nb100nf_ix, 0.set nb100nf_iy, 16.set nb100nf_iz, 32.set nb100nf_iq, 48.set nb100nf_vctot, 64.set nb100nf_half, 80.set nb100nf_three, 96.set nb100nf_is3, 112.set nb100nf_ii3, 116.set nb100nf_innerjjnr, 120.set nb100nf_innerk, 124.set nb100nf_n, 128.set nb100nf_nn1, 132.set nb100nf_nri, 136.set nb100nf_facel, 140.set nb100nf_nouter, 144.set nb100nf_ninner, 148.set nb100nf_salign, 152 pushl %ebp movl %esp,%ebp pushl %eax pushl %ebx pushl %ecx pushl %edx pushl %esi pushl %edi subl $156,%esp ## local stack space movl %esp,%eax andl $0xf,%eax subl %eax,%esp movl %eax,nb100nf_salign(%esp) emms ## Move args passed by reference to stack movl nb100nf_p_nri(%ebp),%ecx movl nb100nf_p_facel(%ebp),%esi movl (%ecx),%ecx movl (%esi),%esi movl %ecx,nb100nf_nri(%esp) movl %esi,nb100nf_facel(%esp) ## zero iteration counters movl $0,%eax⌨️ 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?