nb_kernel112_ia32_3dnow.s

最著名最快的分子模拟软件

        punpckldq %mm4,%mm4

        movd nb112_qqOH(%esp),%mm6
        movq nb112_qqHH(%esp),%mm7

        pfsubr nb112_ixO(%esp),%mm0
        pfsubr nb112_izO(%esp),%mm1

        movq  %mm0,nb112_dxO(%esp)
        pfmul %mm0,%mm0
        movd  %mm1,nb112_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 nb112_ixH(%esp),%mm2
        pfsubr nb112_iyH(%esp),%mm3
        pfsubr nb112_izH(%esp),%mm4   ## mm2-mm4 is dxH-dzH 

        movq %mm2,nb112_dxH(%esp)
        movq %mm3,nb112_dyH(%esp)
        movq %mm4,nb112_dzH(%esp)
        pfmul %mm2,%mm2
        pfmul %mm3,%mm3
        pfmul %mm4,%mm4

        pfadd %mm2,%mm3
        pfadd %mm4,%mm3         ## mm3=rsqH 

        pfrsqrt %mm0,%mm1

        movq %mm1,%mm2
        pfmul %mm1,%mm1
        pfrsqit1 %mm0,%mm1
        pfrcpit2 %mm2,%mm1      ## mm1=invsqrt 
        movq  %mm1,%mm4
        pfmul %mm4,%mm4         ## mm4=invsq 
        ## calculate potential and scalar force 
        pfmul %mm1,%mm6         ## mm6=vcoul 
        pfmul %mm6,%mm4         ## mm4=fscalar  

        pfrsqrt %mm3,%mm5
        pswapd %mm3,%mm3
        pfrsqrt %mm3,%mm2
        pswapd %mm3,%mm3
        punpckldq %mm2,%mm5     ## seeds are in mm5 now, and rsq in mm3 

        movq %mm5,%mm2
        pfmul %mm5,%mm5
        pfrsqit1 %mm3,%mm5
        pfrcpit2 %mm2,%mm5      ## mm5=invsqrt 
        movq %mm5,%mm3
        pfmul %mm3,%mm3         ## mm3=invsq 
        pfmul %mm5,%mm7         ## mm7=vcoul 
        pfmul %mm7,%mm3         ## mm3=fscal for the two H's. 

        ## update vctot 
        pfadd %mm6,%mm7
        pfadd nb112_vctot(%esp),%mm7
        movq %mm7,nb112_vctot(%esp)

        ## spread oxygen fscalar to both positions 
        punpckldq %mm4,%mm4
        ## calc vectorial force for O 
        movq nb112_dxO(%esp),%mm0
        movd nb112_dzO(%esp),%mm1
        pfmul %mm4,%mm0
        pfmul %mm4,%mm1

        ## calc vectorial force for H's 
        movq nb112_dxH(%esp),%mm5
        movq nb112_dyH(%esp),%mm6
        movq nb112_dzH(%esp),%mm7
        pfmul %mm3,%mm5
        pfmul %mm3,%mm6
        pfmul %mm3,%mm7

        ## update iO particle force 
        movq nb112_fixO(%esp),%mm2
        movd nb112_fizO(%esp),%mm3
        pfadd %mm0,%mm2
        pfadd %mm1,%mm3
        movq %mm2,nb112_fixO(%esp)
        movd %mm3,nb112_fizO(%esp)

        ## update iH forces 
        movq nb112_fixH(%esp),%mm2
        movq nb112_fiyH(%esp),%mm3
        movq nb112_fizH(%esp),%mm4
        pfadd %mm5,%mm2
        pfadd %mm6,%mm3
        pfadd %mm7,%mm4
        movq %mm2,nb112_fixH(%esp)
        movq %mm3,nb112_fiyH(%esp)
        movq %mm4,nb112_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

        movd nb112_qqOH(%esp),%mm6
        movq nb112_qqHH(%esp),%mm7

        pfsubr nb112_ixO(%esp),%mm0
        pfsubr nb112_izO(%esp),%mm1

        movq  %mm0,nb112_dxO(%esp)
        pfmul %mm0,%mm0
        movd  %mm1,nb112_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 nb112_ixH(%esp),%mm2
        pfsubr nb112_iyH(%esp),%mm3
        pfsubr nb112_izH(%esp),%mm4   ## mm2-mm4 is dxH-dzH 

        movq %mm2,nb112_dxH(%esp)
        movq %mm3,nb112_dyH(%esp)
        movq %mm4,nb112_dzH(%esp)
        pfmul %mm2,%mm2
        pfmul %mm3,%mm3
        pfmul %mm4,%mm4

        pfadd %mm2,%mm3
        pfadd %mm4,%mm3         ## mm3=rsqH 

        pfrsqrt %mm0,%mm1

        movq %mm1,%mm2
        pfmul %mm1,%mm1
        pfrsqit1 %mm0,%mm1
        pfrcpit2 %mm2,%mm1      ## mm1=invsqrt 
        movq  %mm1,%mm4
        pfmul %mm4,%mm4         ## mm4=invsq 
        ## calculate potential and scalar force 
        pfmul %mm1,%mm6         ## mm6=vcoul 
        pfmul %mm6,%mm4         ## mm4=fscalar  

        pfrsqrt %mm3,%mm5
        pswapd %mm3,%mm3
        pfrsqrt %mm3,%mm2
        pswapd %mm3,%mm3
        punpckldq %mm2,%mm5     ## seeds are in mm5 now, and rsq in mm3. 

        movq %mm5,%mm2
        pfmul %mm5,%mm5
        pfrsqit1 %mm3,%mm5
        pfrcpit2 %mm2,%mm5      ## mm5=invsqrt 
        movq %mm5,%mm3
        pfmul %mm3,%mm3         ## mm3=invsq 
        pfmul %mm5,%mm7         ## mm7=vcoul 
        pfmul %mm7,%mm3         ## mm3=fscal for the two H's. 

        ## update vctot 
        pfadd %mm6,%mm7
        pfadd nb112_vctot(%esp),%mm7
        movq %mm7,nb112_vctot(%esp)

        ## spread oxygen fscalar to both positions 
        punpckldq %mm4,%mm4
        ## calc vectorial force for O 
        movq nb112_dxO(%esp),%mm0
        movd nb112_dzO(%esp),%mm1
        pfmul %mm4,%mm0
        pfmul %mm4,%mm1

        ## calc vectorial force for H's 
        movq nb112_dxH(%esp),%mm5
        movq nb112_dyH(%esp),%mm6
        movq nb112_dzH(%esp),%mm7
        pfmul %mm3,%mm5
        pfmul %mm3,%mm6
        pfmul %mm3,%mm7

        ## update iO particle force 
        movq nb112_fixO(%esp),%mm2
        movd nb112_fizO(%esp),%mm3
        pfadd %mm0,%mm2
        pfadd %mm1,%mm3
        movq %mm2,nb112_fixO(%esp)
        movd %mm3,nb112_fizO(%esp)

        ## update iH forces 
        movq nb112_fixH(%esp),%mm2
        movq nb112_fiyH(%esp),%mm3
        movq nb112_fizH(%esp),%mm4
        pfadd %mm5,%mm2
        pfadd %mm6,%mm3
        pfadd %mm7,%mm4
        movq %mm2,nb112_fixH(%esp)
        movq %mm3,nb112_fiyH(%esp)
        movq %mm4,nb112_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 24(%edi,%eax,4),%mm2
        movd 32(%edi,%eax,4),%mm3
        pfsub %mm0,%mm2
        pfsub %mm1,%mm3
        movq %mm2,24(%edi,%eax,4)
        movd %mm3,32(%edi,%eax,4)

        ##  done  - one more? 
        decl nb112_innerk(%esp)
        jz  _nb_kernel112_ia32_3dnow.nb112_updateouterdata
        jmp _nb_kernel112_ia32_3dnow.nb112_inner_loop
_nb_kernel112_ia32_3dnow.nb112_updateouterdata: 
        movl  nb112_ii3(%esp),%ecx

        movq  (%edi,%ecx,4),%mm6       ## increment iO force  
        movd  8(%edi,%ecx,4),%mm7
        pfadd nb112_fixO(%esp),%mm6
        pfadd nb112_fizO(%esp),%mm7
        movq  %mm6,(%edi,%ecx,4)
        movd  %mm7,8(%edi,%ecx,4)

        movq  nb112_fixH(%esp),%mm0
        movq  nb112_fiyH(%esp),%mm3
        movq  nb112_fizH(%esp),%mm1
        movq  %mm0,%mm2
        punpckldq %mm3,%mm0     ## mm0(l)=fxH1, mm0(h)=fyH1 
        punpckhdq %mm3,%mm2     ## mm2(l)=fxH2, mm2(h)=fyH2 
        movq %mm1,%mm3
        pswapd %mm3,%mm3
        ## mm1 is fzH1 
        ## mm3 is fzH2 

        movq  12(%edi,%ecx,4),%mm6          ## increment iH1 force  
        movd  20(%edi,%ecx,4),%mm7
        pfadd %mm0,%mm6
        pfadd %mm1,%mm7
        movq  %mm6,12(%edi,%ecx,4)
        movd  %mm7,20(%edi,%ecx,4)

        movq  24(%edi,%ecx,4),%mm6          ## increment iH2 force 
        movd  32(%edi,%ecx,4),%mm7
        pfadd %mm2,%mm6
        pfadd %mm3,%mm7
        movq  %mm6,24(%edi,%ecx,4)
        movd  %mm7,32(%edi,%ecx,4)


        movl  nb112_fshift(%ebp),%ebx      ## increment fshift force 
        movl  nb112_is3(%esp),%edx

        movq  (%ebx,%edx,4),%mm6
        movd  8(%ebx,%edx,4),%mm7
        pfadd nb112_fixO(%esp),%mm6
        pfadd nb112_fizO(%esp),%mm7
        pfadd %mm0,%mm6
        pfadd %mm1,%mm7
        pfadd %mm2,%mm6
        pfadd %mm3,%mm7
        movq  %mm6,(%ebx,%edx,4)
        movd  %mm7,8(%ebx,%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]

        movq  nb112_vctot(%esp),%mm7
        pfacc %mm7,%mm7           ## get and sum the two parts of total potential 

        movl  nb112_Vc(%ebp),%eax
        movd  (%eax,%edx,4),%mm6
        pfadd %mm7,%mm6
        movd  %mm6,(%eax,%edx,4)          ## increment vc[gid] 

        movq  nb112_Vvdwtot(%esp),%mm7
        pfacc %mm7,%mm7           ## get and sum the two parts of total potential 

        movl  nb112_Vvdw(%ebp),%eax
        movd  (%eax,%edx,4),%mm6
        pfadd %mm7,%mm6
        movd  %mm6,(%eax,%edx,4)          ## increment Vvdwtot[gid] 
       ## finish if last 
        movl nb112_nn1(%esp),%ecx
        ## esi already loaded with n
        incl %esi
        subl %esi,%ecx
        jz _nb_kernel112_ia32_3dnow.nb112_outerend

        ## not last, iterate outer loop once more!  
        movl %esi,nb112_n(%esp)
        jmp _nb_kernel112_ia32_3dnow.nb112_outer
_nb_kernel112_ia32_3dnow.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_3dnow.nb112_end
        ## non-zero, do one more workunit
        jmp   _nb_kernel112_ia32_3dnow.nb112_threadloop
_nb_kernel112_ia32_3dnow.nb112_end: 
        femms

        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)

        addl $220,%esp
        popl %edi
        popl %esi
        popl %edx
        popl %ecx
        popl %ebx
        popl %eax
        leave
        ret


.globl nb_kernel112nf_ia32_3dnow
.globl _nb_kernel112nf_ia32_3dnow
nb_kernel112nf_ia32_3dnow:      
_nb_kernel112nf_ia32_3dnow:     
.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_p_krf, 56
.set nb112nf_p_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 
.set nb112nf_is3, 0
.set nb112nf_ii3, 4
.set nb112nf_ixO, 8
.set nb112nf_iyO, 12
.set nb112nf_izO, 16
.set nb112nf_ixH, 20
.set nb112nf_iyH, 28
.set nb112nf_izH, 36
.set nb112nf_qqOO, 44
.set nb112nf_qqOH, 52
.set nb112nf_qqHH, 60
.set nb112nf_c6, 68
.set nb112nf_c12, 76
.set nb112nf_vctot, 84
.set nb112nf_Vvdwtot, 92
.set nb112nf_innerjjnr, 100
.set nb112nf_innerk, 104
.set nb112nf_n, 108                         ## idx for outer loop
.set nb112nf_nn1, 112                       ## number of outer iterations
.set nb112nf_nri, 116
.set nb112nf_ntype, 120
.set nb112nf_nouter, 124
.set nb112nf_ninner, 128
        pushl %ebp
        movl %esp,%ebp
        pushl %eax
        pushl %ebx
        pushl %ecx
        pushl %edx
        pushl %esi
        pushl %edi