nb_kernel102_ia32_sse.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 1,986 行 · 第 1/5 页

        ## update forces H1 - j water 
        movaps  nb102_fjxO(%esp),%xmm3
        movaps  nb102_fjyO(%esp),%xmm4
        movaps  nb102_fjzO(%esp),%xmm5
        subps   %xmm0,%xmm3
        subps   %xmm1,%xmm4
        subps   %xmm2,%xmm5
        movaps  %xmm3,nb102_fjxO(%esp)
        movaps  %xmm4,nb102_fjyO(%esp)
        movaps  %xmm5,nb102_fjzO(%esp)
        addps   nb102_fixH1(%esp),%xmm0
        addps   nb102_fiyH1(%esp),%xmm1
        addps   nb102_fizH1(%esp),%xmm2
        movaps  %xmm0,nb102_fixH1(%esp)
        movaps  %xmm1,nb102_fiyH1(%esp)
        movaps  %xmm2,nb102_fizH1(%esp)
        ## do forces H2 - j water 
        movaps %xmm7,%xmm0
        movaps %xmm7,%xmm1
        movaps %xmm7,%xmm2
        mulps   nb102_dxH2O(%esp),%xmm0
        mulps   nb102_dyH2O(%esp),%xmm1
        mulps   nb102_dzH2O(%esp),%xmm2
        movaps  nb102_fjxO(%esp),%xmm3
        movaps  nb102_fjyO(%esp),%xmm4
        movaps  nb102_fjzO(%esp),%xmm5
        subps   %xmm0,%xmm3
        subps   %xmm1,%xmm4
        subps   %xmm2,%xmm5
        movl    nb102_faction(%ebp),%esi
        movaps  %xmm3,nb102_fjxO(%esp)
        movaps  %xmm4,nb102_fjyO(%esp)
        movaps  %xmm5,nb102_fjzO(%esp)
        addps   nb102_fixH2(%esp),%xmm0
        addps   nb102_fiyH2(%esp),%xmm1
        addps   nb102_fizH2(%esp),%xmm2
        movaps  %xmm0,nb102_fixH2(%esp)
        movaps  %xmm1,nb102_fiyH2(%esp)
        movaps  %xmm2,nb102_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  nb102_fjxO(%esp),%xmm3
        movaps  nb102_fjyO(%esp),%xmm4
        movaps  nb102_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  nb102_innerk(%esp)
        jz    _nb_kernel102_ia32_sse.nb102_updateouterdata
        jmp   _nb_kernel102_ia32_sse.nb102_single_loop
_nb_kernel102_ia32_sse.nb102_updateouterdata: 
        movl  nb102_ii3(%esp),%ecx
        movl  nb102_faction(%ebp),%edi
        movl  nb102_fshift(%ebp),%esi
        movl  nb102_is3(%esp),%edx

        ## accumulate Oi forces in xmm0, xmm1, xmm2 
        movaps nb102_fixO(%esp),%xmm0
        movaps nb102_fiyO(%esp),%xmm1
        movaps nb102_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 nb102_fixH1(%esp),%xmm0
        movaps nb102_fiyH1(%esp),%xmm1
        movaps nb102_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 nb102_fixH2(%esp),%xmm0
        movaps nb102_fiyH2(%esp),%xmm1
        movaps nb102_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 nb102_n(%esp),%esi
        ## get group index for i particle 
        movl  nb102_gid(%ebp),%edx              ## base of gid[]
        movl  (%edx,%esi,4),%edx                ## ggid=gid[n]

        ## accumulate total potential energy and update it 
        movaps nb102_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  nb102_Vc(%ebp),%eax
        addss (%eax,%edx,4),%xmm7
        ## move back to mem 
        movss %xmm7,(%eax,%edx,4)

        ## finish if last 
        movl nb102_nn1(%esp),%ecx
        ## esi already loaded with n
        incl %esi
        subl %esi,%ecx
        jz _nb_kernel102_ia32_sse.nb102_outerend

        ## not last, iterate outer loop once more!  
        movl %esi,nb102_n(%esp)
        jmp _nb_kernel102_ia32_sse.nb102_outer
_nb_kernel102_ia32_sse.nb102_outerend: 
        ## check if more outer neighborlists remain
        movl  nb102_nri(%esp),%ecx
        ## esi already loaded with n above
        subl  %esi,%ecx
        jz _nb_kernel102_ia32_sse.nb102_end
        ## non-zero, do one more workunit
        jmp   _nb_kernel102_ia32_sse.nb102_threadloop
_nb_kernel102_ia32_sse.nb102_end: 
        emms

        movl nb102_nouter(%esp),%eax
        movl nb102_ninner(%esp),%ebx
        movl nb102_outeriter(%ebp),%ecx
        movl nb102_inneriter(%ebp),%edx
        movl %eax,(%ecx)
        movl %ebx,(%edx)

        movl nb102_salign(%esp),%eax
        addl %eax,%esp
        addl $1432,%esp
        popl %edi
        popl %esi
        popl %edx
        popl %ecx
        popl %ebx
        popl %eax
        leave
        ret





.globl nb_kernel102nf_ia32_sse
.globl _nb_kernel102nf_ia32_sse
nb_kernel102nf_ia32_sse:        
_nb_kernel102nf_ia32_sse:       
.set nb102nf_p_nri, 8
.set nb102nf_iinr, 12
.set nb102nf_jindex, 16
.set nb102nf_jjnr, 20
.set nb102nf_shift, 24
.set nb102nf_shiftvec, 28
.set nb102nf_fshift, 32
.set nb102nf_gid, 36
.set nb102nf_pos, 40
.set nb102nf_faction, 44
.set nb102nf_charge, 48
.set nb102nf_p_facel, 52
.set nb102nf_p_krf, 56
.set nb102nf_p_crf, 60
.set nb102nf_Vc, 64
.set nb102nf_type, 68
.set nb102nf_p_ntype, 72
.set nb102nf_vdwparam, 76
.set nb102nf_Vvdw, 80
.set nb102nf_p_tabscale, 84
.set nb102nf_VFtab, 88
.set nb102nf_invsqrta, 92
.set nb102nf_dvda, 96
.set nb102nf_p_gbtabscale, 100
.set nb102nf_GBtab, 104
.set nb102nf_p_nthreads, 108
.set nb102nf_count, 112
.set nb102nf_mtx, 116
.set nb102nf_outeriter, 120
.set nb102nf_inneriter, 124
.set nb102nf_work, 128
        ## stack offsets for local variables  
        ## bottom of stack is cache-aligned for sse use         
.set nb102nf_ixO, 0
.set nb102nf_iyO, 16
.set nb102nf_izO, 32
.set nb102nf_ixH1, 48
.set nb102nf_iyH1, 64
.set nb102nf_izH1, 80
.set nb102nf_ixH2, 96
.set nb102nf_iyH2, 112
.set nb102nf_izH2, 128
.set nb102nf_jxO, 144
.set nb102nf_jyO, 160
.set nb102nf_jzO, 176
.set nb102nf_jxH1, 192
.set nb102nf_jyH1, 208
.set nb102nf_jzH1, 224
.set nb102nf_jxH2, 240
.set nb102nf_jyH2, 256
.set nb102nf_jzH2, 272
.set nb102nf_qqOO, 288
.set nb102nf_qqOH, 304
.set nb102nf_qqHH, 320
.set nb102nf_vctot, 336
.set nb102nf_half, 352
.set nb102nf_three, 368
.set nb102nf_rsqOO, 384
.set nb102nf_rsqOH1, 400
.set nb102nf_rsqOH2, 416
.set nb102nf_rsqH1O, 432
.set nb102nf_rsqH1H1, 448
.set nb102nf_rsqH1H2, 464
.set nb102nf_rsqH2O, 480
.set nb102nf_rsqH2H1, 496
.set nb102nf_rsqH2H2, 512
.set nb102nf_rinvOO, 528
.set nb102nf_rinvOH1, 544
.set nb102nf_rinvOH2, 560
.set nb102nf_rinvH1O, 576
.set nb102nf_rinvH1H1, 592
.set nb102nf_rinvH1H2, 608
.set nb102nf_rinvH2O, 624
.set nb102nf_rinvH2H1, 640
.set nb102nf_rinvH2H2, 656
.set nb102nf_is3, 672
.set nb102nf_ii3, 676
.set nb102nf_innerjjnr, 680
.set nb102nf_innerk, 684
.set nb102nf_n, 688
.set nb102nf_nn1, 692
.set nb102nf_nri, 696
.set nb102nf_nouter, 700
.set nb102nf_ninner, 704
.set nb102nf_salign, 708
        pushl %ebp
        movl %esp,%ebp
        pushl %eax
        pushl %ebx
        pushl %ecx
        pushl %edx
        pushl %esi
        pushl %edi
        subl $712,%esp          ## local stack space 
        movl %esp,%eax
        andl $0xf,%eax
        subl %eax,%esp
        movl %eax,nb102nf_salign(%esp)

        emms

        ## Move args passed by reference to stack
        movl nb102nf_p_nri(%ebp),%ecx
        movl (%ecx),%ecx
        movl %ecx,nb102nf_nri(%esp)

        ## zero iteration counters
        movl $0,%eax
        movl %eax,nb102nf_nouter(%esp)
        movl %eax,nb102nf_ninner(%esp)


        ## assume we have at least one i particle - start directly 
        movl  nb102nf_iinr(%ebp),%ecx         ## ecx = pointer into iinr[]      
        movl  (%ecx),%ebx           ## ebx =ii 

        movl  nb102nf_charge(%ebp),%edx
        movss (%edx,%ebx,4),%xmm3
        movss %xmm3,%xmm4
        movss 4(%edx,%ebx,4),%xmm5
        movl nb102nf_p_facel(%ebp),%esi
        movss (%esi),%xmm6
        mulss  %xmm3,%xmm3
        mulss  %xmm5,%xmm4
        mulss  %xmm5,%xmm5
        mulss  %xmm6,%xmm3