nb_kernel210_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,569 行 · 第 1/4 页
S
1,569 行
mulps %xmm3,%xmm6 ## xmm6=vcoul=qq*(rinv+ krsq) mulps nb210_two(%esp),%xmm7 mulps nb210_c6(%esp),%xmm1 mulps nb210_c12(%esp),%xmm2 movaps %xmm2,%xmm5 subps %xmm1,%xmm5 ## Vvdw=Vvdw12-Vvdw6 addps nb210_Vvdwtot(%esp),%xmm5 mulps nb210_six(%esp),%xmm1 mulps nb210_twelve(%esp),%xmm2 subps %xmm1,%xmm2 subps %xmm7,%xmm0 mulps %xmm0,%xmm3 addps %xmm3,%xmm2 mulps %xmm2,%xmm4 ## xmm4=total fscal addps nb210_vctot(%esp),%xmm6 movaps nb210_dx(%esp),%xmm0 movaps nb210_dy(%esp),%xmm1 movaps nb210_dz(%esp),%xmm2 movaps %xmm6,nb210_vctot(%esp) movaps %xmm5,nb210_Vvdwtot(%esp) movl nb210_faction(%ebp),%edi mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movaps nb210_fix(%esp),%xmm3 movaps nb210_fiy(%esp),%xmm4 movaps nb210_fiz(%esp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb210_fix(%esp) movaps %xmm4,nb210_fiy(%esp) movaps %xmm5,nb210_fiz(%esp) ## the fj's - start by accumulating x & y forces from memory movlps (%edi,%eax,4),%xmm4 movlps (%edi,%ecx,4),%xmm6 movhps (%edi,%ebx,4),%xmm4 movhps (%edi,%edx,4),%xmm6 movaps %xmm4,%xmm3 shufps $136,%xmm6,%xmm3 ## constant 10001000 shufps $221,%xmm6,%xmm4 ## constant 11011101 ## now xmm3-xmm5 contains fjx, fjy, fjz subps %xmm0,%xmm3 subps %xmm1,%xmm4 ## unpack them back so we can store them - first x & y in xmm3/xmm4 movaps %xmm3,%xmm6 unpcklps %xmm4,%xmm6 unpckhps %xmm4,%xmm3 ## xmm6(l)=x & y for j1, (h) for j2 ## xmm3(l)=x & y for j3, (h) for j4 movlps %xmm6,(%edi,%eax,4) movlps %xmm3,(%edi,%ecx,4) movhps %xmm6,(%edi,%ebx,4) movhps %xmm3,(%edi,%edx,4) ## and the z forces movss 8(%edi,%eax,4),%xmm4 movss 8(%edi,%ebx,4),%xmm5 movss 8(%edi,%ecx,4),%xmm6 movss 8(%edi,%edx,4),%xmm7 subss %xmm2,%xmm4 shufps $229,%xmm2,%xmm2 ## constant 11100101 subss %xmm2,%xmm5 shufps $234,%xmm2,%xmm2 ## constant 11101010 subss %xmm2,%xmm6 shufps $255,%xmm2,%xmm2 ## constant 11111111 subss %xmm2,%xmm7 movss %xmm4,8(%edi,%eax,4) movss %xmm5,8(%edi,%ebx,4) movss %xmm6,8(%edi,%ecx,4) movss %xmm7,8(%edi,%edx,4) ## should we do one more iteration? subl $4,nb210_innerk(%esp) jl _nb_kernel210_ia32_sse.nb210_finish_inner jmp _nb_kernel210_ia32_sse.nb210_unroll_loop_nb_kernel210_ia32_sse.nb210_finish_inner: ## check if at least two particles remain addl $4,nb210_innerk(%esp) movl nb210_innerk(%esp),%edx andl $2,%edx jnz _nb_kernel210_ia32_sse.nb210_dopair jmp _nb_kernel210_ia32_sse.nb210_checksingle_nb_kernel210_ia32_sse.nb210_dopair: movl nb210_charge(%ebp),%esi movl nb210_innerjjnr(%esp),%ecx movl (%ecx),%eax movl 4(%ecx),%ebx addl $8,nb210_innerjjnr(%esp) xorps %xmm3,%xmm3 movss (%esi,%eax,4),%xmm3 movss (%esi,%ebx,4),%xmm6 shufps $12,%xmm6,%xmm3 ## constant 00001100 shufps $88,%xmm3,%xmm3 ## constant 01011000 ;# xmm3(0,1) has the charges movl nb210_type(%ebp),%esi movl %eax,%ecx movl %ebx,%edx movl (%esi,%ecx,4),%ecx movl (%esi,%edx,4),%edx movl nb210_vdwparam(%ebp),%esi shll %ecx shll %edx movl nb210_ntia(%esp),%edi addl %edi,%ecx addl %edi,%edx movlps (%esi,%ecx,4),%xmm6 movhps (%esi,%edx,4),%xmm6 movl nb210_pos(%ebp),%edi xorps %xmm7,%xmm7 movaps %xmm6,%xmm4 shufps $8,%xmm4,%xmm4 ## constant 00001000 shufps $13,%xmm6,%xmm6 ## constant 00001101 movlhps %xmm7,%xmm4 movlhps %xmm7,%xmm6 movaps %xmm4,nb210_c6(%esp) movaps %xmm6,nb210_c12(%esp) 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 nb210_iq(%esp),%xmm3 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 nb210_faction(%ebp),%edi ## move ix-iz to xmm4-xmm6 xorps %xmm7,%xmm7 movaps nb210_ix(%esp),%xmm4 movaps nb210_iy(%esp),%xmm5 movaps nb210_iz(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb210_dx(%esp) movaps %xmm5,nb210_dy(%esp) movaps %xmm6,nb210_dz(%esp) ## square it mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 mulps %xmm6,%xmm6 addps %xmm5,%xmm4 addps %xmm6,%xmm4 ## rsq in xmm4 movaps nb210_krf(%esp),%xmm7 rsqrtps %xmm4,%xmm5 ## lookup seed in xmm5 movaps %xmm5,%xmm2 mulps %xmm5,%xmm5 movaps nb210_three(%esp),%xmm1 mulps %xmm4,%xmm5 ## rsq*lu*lu movaps nb210_half(%esp),%xmm0 mulps %xmm4,%xmm7 ## xmm7=krsq 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 %xmm0,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq movaps %xmm4,%xmm1 subps nb210_crf(%esp),%xmm6 mulps %xmm4,%xmm1 mulps %xmm4,%xmm1 ## xmm1=rinvsix movaps %xmm1,%xmm2 mulps %xmm2,%xmm2 ## xmm2=rinvtwelve mulps %xmm3,%xmm6 ## xmm6=vcoul=qq*(rinv+ krsq-crf) mulps nb210_two(%esp),%xmm7 mulps nb210_c6(%esp),%xmm1 mulps nb210_c12(%esp),%xmm2 movaps %xmm2,%xmm5 subps %xmm1,%xmm5 ## Vvdw=Vvdw12-Vvdw6 addps nb210_Vvdwtot(%esp),%xmm5 mulps nb210_six(%esp),%xmm1 mulps nb210_twelve(%esp),%xmm2 subps %xmm1,%xmm2 subps %xmm7,%xmm0 mulps %xmm0,%xmm3 addps %xmm3,%xmm2 mulps %xmm2,%xmm4 ## xmm4=total fscal addps nb210_vctot(%esp),%xmm6 movaps nb210_dx(%esp),%xmm0 movaps nb210_dy(%esp),%xmm1 movaps nb210_dz(%esp),%xmm2 movaps %xmm6,nb210_vctot(%esp) movaps %xmm5,nb210_Vvdwtot(%esp) mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movaps nb210_fix(%esp),%xmm3 movaps nb210_fiy(%esp),%xmm4 movaps nb210_fiz(%esp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb210_fix(%esp) movaps %xmm4,nb210_fiy(%esp) movaps %xmm5,nb210_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_kernel210_ia32_sse.nb210_checksingle: movl nb210_innerk(%esp),%edx andl $1,%edx jnz _nb_kernel210_ia32_sse.nb210_dosingle jmp _nb_kernel210_ia32_sse.nb210_updateouterdata_nb_kernel210_ia32_sse.nb210_dosingle: movl nb210_charge(%ebp),%esi movl nb210_pos(%ebp),%edi movl nb210_innerjjnr(%esp),%ecx xorps %xmm3,%xmm3 movl (%ecx),%eax movss (%esi,%eax,4),%xmm3 ## xmm3(0) has the charge movl nb210_type(%ebp),%esi movl %eax,%ecx movl (%esi,%ecx,4),%ecx movl nb210_vdwparam(%ebp),%esi shll %ecx addl nb210_ntia(%esp),%ecx xorps %xmm6,%xmm6 movlps (%esi,%ecx,4),%xmm6 movaps %xmm6,%xmm4 shufps $252,%xmm4,%xmm4 ## constant 11111100 shufps $253,%xmm6,%xmm6 ## constant 11111101 movaps %xmm4,nb210_c6(%esp) movaps %xmm6,nb210_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 mulps nb210_iq(%esp),%xmm3 xorps %xmm7,%xmm7 movaps nb210_ix(%esp),%xmm4 movaps nb210_iy(%esp),%xmm5 movaps nb210_iz(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb210_dx(%esp) movaps %xmm5,nb210_dy(%esp) movaps %xmm6,nb210_dz(%esp) ## square it mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 mulps %xmm6,%xmm6 addps %xmm5,%xmm4 addps %xmm6,%xmm4 ## rsq in xmm4 movaps nb210_krf(%esp),%xmm7 rsqrtps %xmm4,%xmm5 ## lookup seed in xmm5 movaps %xmm5,%xmm2 mulps %xmm5,%xmm5 movaps nb210_three(%esp),%xmm1 mulps %xmm4,%xmm5 ## rsq*lu*lu movaps nb210_half(%esp),%xmm0 mulps %xmm4,%xmm7 ## xmm7=krsq 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 %xmm0,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq movaps %xmm4,%xmm1 subps nb210_crf(%esp),%xmm6 mulps %xmm4,%xmm1 mulps %xmm4,%xmm1 ## xmm1=rinvsix movaps %xmm1,%xmm2 mulps %xmm2,%xmm2 ## xmm2=rinvtwelve mulps %xmm3,%xmm6 ## xmm6=vcoul mulps nb210_two(%esp),%xmm7 mulps nb210_c6(%esp),%xmm1 mulps nb210_c12(%esp),%xmm2 movaps %xmm2,%xmm5 subps %xmm1,%xmm5 ## Vvdw=Vvdw12-Vvdw6 addss nb210_Vvdwtot(%esp),%xmm5 mulps nb210_six(%esp),%xmm1 mulps nb210_twelve(%esp),%xmm2 subps %xmm1,%xmm2 subps %xmm7,%xmm0 mulps %xmm0,%xmm3 addps %xmm3,%xmm2 mulps %xmm2,%xmm4 ## xmm4=total fscal addss nb210_vctot(%esp),%xmm6 movl nb210_faction(%ebp),%edi movaps nb210_dx(%esp),%xmm0 movaps nb210_dy(%esp),%xmm1 movaps nb210_dz(%esp),%xmm2 movss %xmm6,nb210_vctot(%esp) movss %xmm5,nb210_Vvdwtot(%esp) mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movaps nb210_fix(%esp),%xmm3 movaps nb210_fiy(%esp),%xmm4 movaps nb210_fiz(%esp),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movaps %xmm3,nb210_fix(%esp) movaps %xmm4,nb210_fiy(%esp) movaps %xmm5,nb210_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)⌨️ 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?