📄 nb_kernel302_ia32_3dnow.s
字号:
## spread oxygen fscalar to both positions punpckldq %mm3,%mm3 ## calc vectorial force for O movq nb302_dxO(%esp),%mm0 movd nb302_dzO(%esp),%mm1 pfmul %mm3,%mm0 pfmul %mm3,%mm1 ## calc vectorial force for H's movq nb302_dxH(%esp),%mm5 movq nb302_dyH(%esp),%mm6 movq nb302_dzH(%esp),%mm7 pfmul %mm4,%mm5 pfmul %mm4,%mm6 pfmul %mm4,%mm7 ## update iO particle force movq nb302_fixO(%esp),%mm2 movd nb302_fizO(%esp),%mm3 pfadd %mm0,%mm2 pfadd %mm1,%mm3 movq %mm2,nb302_fixO(%esp) movd %mm3,nb302_fizO(%esp) ## update iH forces movq nb302_fixH(%esp),%mm2 movq nb302_fiyH(%esp),%mm3 movq nb302_fizH(%esp),%mm4 pfadd %mm5,%mm2 pfadd %mm6,%mm3 pfadd %mm7,%mm4 movq %mm2,nb302_fixH(%esp) movq %mm3,nb302_fiyH(%esp) movq %mm4,nb302_fizH(%esp) ## pack j forces from H in the same form as the oxygen force. pfacc %mm6,%mm5 ## mm5(l)=fjx(H1+ h2) mm5(h)=fjy(H1+ h2) pfacc %mm7,%mm7 ## mm7(l)=fjz(H1+ h2) pfadd %mm5,%mm0 ## add up total force on j particle. pfadd %mm7,%mm1 ## update j particle force movq (%edi,%eax,4),%mm2 movd 8(%edi,%eax,4),%mm3 pfsub %mm0,%mm2 pfsub %mm1,%mm3 movq %mm2,(%edi,%eax,4) movd %mm3,8(%edi,%eax,4) ## interactions with j H1 movq 12(%esi,%eax,4),%mm0 movd 20(%esi,%eax,4),%mm1 ## copy & expand to mm2-mm4 for the H interactions movq %mm0,%mm2 movq %mm0,%mm3 movq %mm1,%mm4 punpckldq %mm2,%mm2 punpckhdq %mm3,%mm3 punpckldq %mm4,%mm4 pfsubr nb302_ixO(%esp),%mm0 pfsubr nb302_izO(%esp),%mm1 movq %mm0,nb302_dxO(%esp) pfmul %mm0,%mm0 movd %mm1,nb302_dzO(%esp) pfmul %mm1,%mm1 pfacc %mm1,%mm0 pfadd %mm1,%mm0 ## mm0=rsqO punpckldq %mm2,%mm2 punpckldq %mm3,%mm3 punpckldq %mm4,%mm4 ## mm2-mm4 is jx-jz pfsubr nb302_ixH(%esp),%mm2 pfsubr nb302_iyH(%esp),%mm3 pfsubr nb302_izH(%esp),%mm4 ## mm2-mm4 is dxH-dzH movq %mm2,nb302_dxH(%esp) movq %mm3,nb302_dyH(%esp) movq %mm4,nb302_dzH(%esp) pfmul %mm2,%mm2 pfmul %mm3,%mm3 pfmul %mm4,%mm4 pfadd %mm2,%mm3 pfadd %mm4,%mm3 ## mm3=rsqH movq %mm3,nb302_tmprsqH(%esp) pfrsqrt %mm0,%mm1 movq %mm1,%mm2 pfmul %mm1,%mm1 pfrsqit1 %mm0,%mm1 pfrcpit2 %mm2,%mm1 ## mm1=invsqrt pfmul %mm1,%mm0 ## mm0=rsq pfmul nb302_tsc(%esp),%mm0 pf2iw %mm0,%mm4 movd %mm4,nb302_n1(%esp) pi2fd %mm4,%mm4 pfsub %mm4,%mm0 ## now mm0 is eps and mm4 n0 movq %mm0,%mm2 pfmul %mm2,%mm2 ## mm0 is eps, mm2 eps2 ## coulomb table movl nb302_VFtab(%ebp),%edx movl nb302_n1(%esp),%ecx shll $2,%ecx ## load all values we need movd (%edx,%ecx,4),%mm4 movd 4(%edx,%ecx,4),%mm5 movd 8(%edx,%ecx,4),%mm6 movd 12(%edx,%ecx,4),%mm7 pfmul %mm0,%mm6 ## mm6 = Geps pfmul %mm2,%mm7 ## mm7 = Heps2 pfadd %mm6,%mm5 pfadd %mm7,%mm5 ## mm5 = Fp pfmul nb302_two(%esp),%mm7 ## two*Heps2 pfadd %mm6,%mm7 pfadd %mm5,%mm7 ## mm7=FF pfmul %mm0,%mm5 ## mm5=eps*Fp pfadd %mm4,%mm5 ## mm5= VV pfmul nb302_qqOH(%esp),%mm5 ## vcoul=qq*VV pfmul nb302_qqOH(%esp),%mm7 ## fijC=qq*FF ## update vctot directly, force is moved to mm3 pfadd nb302_vctot(%esp),%mm5 movq %mm5,nb302_vctot(%esp) pxor %mm3,%mm3 pfsub %mm7,%mm3 pfmul nb302_tsc(%esp),%mm3 pfmul %mm1,%mm3 ## mm3 is total fscal (for the oxygen) now movq nb302_tmprsqH(%esp),%mm0 pfrsqrt %mm0,%mm1 pswapd %mm0,%mm0 pfrsqrt %mm0,%mm2 pswapd %mm0,%mm0 punpckldq %mm2,%mm1 ## seeds are in mm1 now, and rsq in mm0. movq %mm1,%mm2 pfmul %mm1,%mm1 pfrsqit1 %mm0,%mm1 pfrcpit2 %mm2,%mm1 ## mm1=invsqrt pfmul %mm1,%mm0 ## mm0=r pfmul nb302_tsc(%esp),%mm0 pf2iw %mm0,%mm4 movq %mm4,nb302_n1(%esp) pi2fd %mm4,%mm4 pfsub %mm4,%mm0 ## now mm0 is eps and mm4 n0 movq %mm0,%mm2 pfmul %mm2,%mm2 ## mm0 is eps, mm2 eps2 ## coulomb table movl nb302_VFtab(%ebp),%edx movl nb302_n1(%esp),%ecx shll $2,%ecx ## load all values we need movd (%edx,%ecx,4),%mm4 movd 4(%edx,%ecx,4),%mm5 movd 8(%edx,%ecx,4),%mm6 movd 12(%edx,%ecx,4),%mm7 movl nb302_n1+4(%esp),%ecx shll $2,%ecx punpckldq (%edx,%ecx,4),%mm4 punpckldq 4(%edx,%ecx,4),%mm5 punpckldq 8(%edx,%ecx,4),%mm6 punpckldq 12(%edx,%ecx,4),%mm7 pfmul %mm0,%mm6 ## mm6 = Geps pfmul %mm2,%mm7 ## mm7 = Heps2 pfadd %mm6,%mm5 pfadd %mm7,%mm5 ## mm5 = Fp pfmul nb302_two(%esp),%mm7 ## two*Heps2 pfadd %mm6,%mm7 pfadd %mm5,%mm7 ## mm7=FF pfmul %mm0,%mm5 ## mm5=eps*Fp pfadd %mm4,%mm5 ## mm5= VV pfmul nb302_qqHH(%esp),%mm5 ## vcoul=qq*VV pfmul nb302_qqHH(%esp),%mm7 ## fijC=qq*FF ## update vctot pfadd nb302_vctot(%esp),%mm5 movq %mm5,nb302_vctot(%esp) ## change sign of fijC and multiply by rinv pxor %mm4,%mm4 pfsub %mm7,%mm4 pfmul nb302_tsc(%esp),%mm4 pfmul %mm1,%mm4 ## mm4 is total fscal (for the hydrogens) now ## spread oxygen fscalar to both positions punpckldq %mm3,%mm3 ## calc vectorial force for O movq nb302_dxO(%esp),%mm0 movd nb302_dzO(%esp),%mm1 pfmul %mm3,%mm0 pfmul %mm3,%mm1 ## calc vectorial force for H's movq nb302_dxH(%esp),%mm5 movq nb302_dyH(%esp),%mm6 movq nb302_dzH(%esp),%mm7 pfmul %mm4,%mm5 pfmul %mm4,%mm6 pfmul %mm4,%mm7 ## update iO particle force movq nb302_fixO(%esp),%mm2 movd nb302_fizO(%esp),%mm3 pfadd %mm0,%mm2 pfadd %mm1,%mm3 movq %mm2,nb302_fixO(%esp) movd %mm3,nb302_fizO(%esp) ## update iH forces movq nb302_fixH(%esp),%mm2 movq nb302_fiyH(%esp),%mm3 movq nb302_fizH(%esp),%mm4 pfadd %mm5,%mm2 pfadd %mm6,%mm3 pfadd %mm7,%mm4 movq %mm2,nb302_fixH(%esp) movq %mm3,nb302_fiyH(%esp) movq %mm4,nb302_fizH(%esp) ## pack j forces from H in the same form as the oxygen force. pfacc %mm6,%mm5 ## mm5(l)=fjx(H1+ h2) mm5(h)=fjy(H1+ h2) pfacc %mm7,%mm7 ## mm7(l)=fjz(H1+ h2) pfadd %mm5,%mm0 ## add up total force on j particle. pfadd %mm7,%mm1 ## update j particle force movq 12(%edi,%eax,4),%mm2 movd 20(%edi,%eax,4),%mm3 pfsub %mm0,%mm2 pfsub %mm1,%mm3 movq %mm2,12(%edi,%eax,4) movd %mm3,20(%edi,%eax,4) ## interactions with j H2 movq 24(%esi,%eax,4),%mm0 movd 32(%esi,%eax,4),%mm1 ## copy & expand to mm2-mm4 for the H interactions movq %mm0,%mm2 movq %mm0,%mm3 movq %mm1,%mm4 punpckldq %mm2,%mm2 punpckhdq %mm3,%mm3 punpckldq %mm4,%mm4 pfsubr nb302_ixO(%esp),%mm0 pfsubr nb302_izO(%esp),%mm1 movq %mm0,nb302_dxO(%esp) pfmul %mm0,%mm0 movd %mm1,nb302_dzO(%esp) pfmul %mm1,%mm1 pfacc %mm1,%mm0 pfadd %mm1,%mm0 ## mm0=rsqO punpckldq %mm2,%mm2 punpckldq %mm3,%mm3 punpckldq %mm4,%mm4 ## mm2-mm4 is jx-jz pfsubr nb302_ixH(%esp),%mm2 pfsubr nb302_iyH(%esp),%mm3 pfsubr nb302_izH(%esp),%mm4 ## mm2-mm4 is dxH-dzH movq %mm2,nb302_dxH(%esp) movq %mm3,nb302_dyH(%esp) movq %mm4,nb302_dzH(%esp) pfmul %mm2,%mm2 pfmul %mm3,%mm3 pfmul %mm4,%mm4 pfadd %mm2,%mm3 pfadd %mm4,%mm3 ## mm3=rsqH movq %mm3,nb302_tmprsqH(%esp) pfrsqrt %mm0,%mm1 movq %mm1,%mm2 pfmul %mm1,%mm1 pfrsqit1 %mm0,%mm1 pfrcpit2 %mm2,%mm1 ## mm1=invsqrt pfmul %mm1,%mm0 pfmul nb302_tsc(%esp),%mm0 pf2iw %mm0,%mm4 movd %mm4,nb302_n1(%esp) pi2fd %mm4,%mm4 pfsub %mm4,%mm0 ## now mm0 is eps and mm4 n0 movq %mm0,%mm2 pfmul %mm2,%mm2 ## mm0 is eps, mm2 eps2 ## coulomb table movl nb302_VFtab(%ebp),%edx movl nb302_n1(%esp),%ecx shll $2,%ecx ## load all values we need movd (%edx,%ecx,4),%mm4 movd 4(%edx,%ecx,4),%mm5 movd 8(%edx,%ecx,4),%mm6 movd 12(%edx,%ecx,4),%mm7 pfmul %mm0,%mm6 ## mm6 = Geps pfmul %mm2,%mm7 ## mm7 = Heps2 pfadd %mm6,%mm5 pfadd %mm7,%mm5 ## mm5 = Fp pfmul nb302_two(%esp),%mm7 ## two*Heps2 pfadd %mm6,%mm7 pfadd %mm5,%mm7 ## mm7=FF pfmul %mm0,%mm5 ## mm5=eps*Fp pfadd %mm4,%mm5 ## mm5= VV pfmul nb302_qqOH(%esp),%mm5 ## vcoul=qq*VV pfmul nb302_qqOH(%esp),%mm7 ## fijC=qq*FF ## update vctot directly, use mm3 for fscal sum. pfadd nb302_vctot(%esp),%mm5 movq %mm5,nb302_vctot(%esp) pxor %mm3,%mm3 pfsub %mm7,%mm3 pfmul nb302_tsc(%esp),%mm3 pfmul %mm1,%mm3 ## mm3 is total fscal (for the oxygen) now movq nb302_tmprsqH(%esp),%mm0 pfrsqrt %mm0,%mm1 pswapd %mm0,%mm0 pfrsqrt %mm0,%mm2 pswapd %mm0,%mm0 punpckldq %mm2,%mm1 ## seeds are in mm1 now, and rsq in mm0. movq %mm1,%mm2 pfmul %mm1,%mm1 pfrsqit1 %mm0,%mm1 pfrcpit2 %mm2,%mm1 ## mm1=invsqrt pfmul %mm1,%mm0 ## mm0=r pfmul nb302_tsc(%esp),%mm0 pf2iw %mm0,%mm4 movq %mm4,nb302_n1(%esp) pi2fd %mm4,%mm4 pfsub %mm4,%mm0 ## now mm0 is eps and mm4 n0 movq %mm0,%mm2 pfmul %mm2,%mm2 ## mm0 is eps, mm2 eps2 ## coulomb table movl nb302_VFtab(%ebp),%edx movl nb302_n1(%esp),%ecx shll $2,%ecx ## load all values we need movd (%edx,%ecx,4),%mm4 movd 4(%edx,%ecx,4),%mm5 movd 8(%edx,%ecx,4),%mm6 movd 12(%edx,%ecx,4),%mm7 movl nb302_n1+4(%esp),%ecx shll $2,%ecx punpckldq (%edx,%ecx,4),%mm4 punpckldq 4(%edx,%ecx,4),%mm5 punpckldq 8(%edx,%ecx,4),%mm6 punpckldq 12(%edx,%ecx,4),%mm7 pfmul %mm0,%mm6 ## mm6 = Geps pfmul %mm2,%mm7 ## mm7 = Heps2 pfadd %mm6,%mm5 pfadd %mm7,%mm5 ## mm5 = Fp pfmul nb302_two(%esp),%mm7 ## two*Heps2 pfadd %mm6,%mm7 pfadd %mm5,%mm7 ## mm7=FF pfmul %mm0,%mm5 ## mm5=eps*Fp pfadd %mm4,%mm5 ## mm5= VV pfmul nb302_qqHH(%esp),%mm5 ## vcoul=qq*VV pfmul nb302_qqHH(%esp),%mm7 ## fijC=qq*FF ## update vctot pfadd nb302_vctot(%esp),%mm5 movq %mm5,nb302_vctot(%esp) ## change sign of fijC and multiply by rinv pxor %mm4,%mm4 pfsub %mm7,%mm4 pfmul nb302_tsc(%esp),%mm4 pfmul %mm1,%mm4 ## mm4 is total fscal (for the hydrogens) now
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -