nb_kernel104_ia32_sse.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 1,968 行 · 第 1/5 页

        addps   nb104_vctot(%esp),%xmm2
        mulps   %xmm4,%xmm7     ## fscal 
        movaps  %xmm2,nb104_vctot(%esp)
        movaps  %xmm0,%xmm1
        movaps  %xmm0,%xmm2
        mulps   nb104_dxH1M(%esp),%xmm0
        mulps   nb104_dyH1M(%esp),%xmm1
        mulps   nb104_dzH1M(%esp),%xmm2
        ## update forces H1 - j water 
        movaps  nb104_fjxM(%esp),%xmm3
        movaps  nb104_fjyM(%esp),%xmm4
        movaps  nb104_fjzM(%esp),%xmm5
        subps   %xmm0,%xmm3
        subps   %xmm1,%xmm4
        subps   %xmm2,%xmm5
        movaps  %xmm3,nb104_fjxM(%esp)
        movaps  %xmm4,nb104_fjyM(%esp)
        movaps  %xmm5,nb104_fjzM(%esp)
        addps   nb104_fixH1(%esp),%xmm0
        addps   nb104_fiyH1(%esp),%xmm1
        addps   nb104_fizH1(%esp),%xmm2
        movaps  %xmm0,nb104_fixH1(%esp)
        movaps  %xmm1,nb104_fiyH1(%esp)
        movaps  %xmm2,nb104_fizH1(%esp)
        ## do forces H2 - j water 
        movaps %xmm7,%xmm0
        movaps %xmm7,%xmm1
        movaps %xmm7,%xmm2
        mulps   nb104_dxH2M(%esp),%xmm0
        mulps   nb104_dyH2M(%esp),%xmm1
        mulps   nb104_dzH2M(%esp),%xmm2
        movaps  nb104_fjxM(%esp),%xmm3
        movaps  nb104_fjyM(%esp),%xmm4
        movaps  nb104_fjzM(%esp),%xmm5
        subps   %xmm0,%xmm3
        subps   %xmm1,%xmm4
        subps   %xmm2,%xmm5
        movl    nb104_faction(%ebp),%esi
        movaps  %xmm3,nb104_fjxM(%esp)
        movaps  %xmm4,nb104_fjyM(%esp)
        movaps  %xmm5,nb104_fjzM(%esp)
        addps   nb104_fixH2(%esp),%xmm0
        addps   nb104_fiyH2(%esp),%xmm1
        addps   nb104_fizH2(%esp),%xmm2
        movaps  %xmm0,nb104_fixH2(%esp)
        movaps  %xmm1,nb104_fiyH2(%esp)
        movaps  %xmm2,nb104_fizH2(%esp)

        ## update j water forces from local variables 
        movlps  36(%esi,%eax,4),%xmm0
        movlps  12(%esi,%eax,4),%xmm1
        movhps  24(%esi,%eax,4),%xmm1
        movaps  nb104_fjxM(%esp),%xmm3
        movaps  nb104_fjyM(%esp),%xmm4
        movaps  nb104_fjzM(%esp),%xmm5
        movaps  %xmm5,%xmm6
        movaps  %xmm5,%xmm7
        shufps $2,%xmm6,%xmm6 ## constant 00000010
        shufps $3,%xmm7,%xmm7 ## constant 00000011
        addss   44(%esi,%eax,4),%xmm5
        addss   20(%esi,%eax,4),%xmm6
        addss   32(%esi,%eax,4),%xmm7
        movss   %xmm5,44(%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  %xmm0,36(%esi,%eax,4)
        movlps  %xmm1,12(%esi,%eax,4)
        movhps  %xmm1,24(%esi,%eax,4)

        decl  nb104_innerk(%esp)
        jz    _nb_kernel104_ia32_sse.nb104_updateouterdata
        jmp   _nb_kernel104_ia32_sse.nb104_single_loop
_nb_kernel104_ia32_sse.nb104_updateouterdata: 
        movl  nb104_ii3(%esp),%ecx
        movl  nb104_faction(%ebp),%edi
        movl  nb104_fshift(%ebp),%esi
        movl  nb104_is3(%esp),%edx

        ## accumulate H1i forces in xmm0, xmm1, xmm2 
        movaps nb104_fixH1(%esp),%xmm0
        movaps nb104_fiyH1(%esp),%xmm1
        movaps nb104_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 
        movaps %xmm0,%xmm6
        movss %xmm2,%xmm7
        movlhps %xmm1,%xmm6
        shufps $8,%xmm6,%xmm6 ## constant 00001000      

        ## accumulate H2i forces in xmm0, xmm1, xmm2 
        movaps nb104_fixH2(%esp),%xmm0
        movaps nb104_fiyH2(%esp),%xmm1
        movaps nb104_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

        ## accumulate M i forces in xmm0, xmm1, xmm2 
        movaps nb104_fixM(%esp),%xmm0
        movaps nb104_fiyM(%esp),%xmm1
        movaps nb104_fizM(%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  36(%edi,%ecx,4),%xmm3
        movss  40(%edi,%ecx,4),%xmm4
        movss  44(%edi,%ecx,4),%xmm5
        addss  %xmm0,%xmm3
        addss  %xmm1,%xmm4
        addss  %xmm2,%xmm5
        movss  %xmm3,36(%edi,%ecx,4)
        movss  %xmm4,40(%edi,%ecx,4)
        movss  %xmm5,44(%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 nb104_n(%esp),%esi
        ## get group index for i particle 
        movl  nb104_gid(%ebp),%edx              ## base of gid[]
        movl  (%edx,%esi,4),%edx                ## ggid=gid[n]

        ## accumulate total potential energy and update it 
        movaps nb104_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  nb104_Vc(%ebp),%eax
        addss (%eax,%edx,4),%xmm7
        ## move back to mem 
        movss %xmm7,(%eax,%edx,4)

        ## finish if last 
        movl nb104_nn1(%esp),%ecx
        ## esi already loaded with n
        incl %esi
        subl %esi,%ecx
        jz _nb_kernel104_ia32_sse.nb104_outerend

        ## not last, iterate outer loop once more!  
        movl %esi,nb104_n(%esp)
        jmp _nb_kernel104_ia32_sse.nb104_outer
_nb_kernel104_ia32_sse.nb104_outerend: 
        ## check if more outer neighborlists remain
        movl  nb104_nri(%esp),%ecx
        ## esi already loaded with n above
        subl  %esi,%ecx
        jz _nb_kernel104_ia32_sse.nb104_end
        ## non-zero, do one more workunit
        jmp   _nb_kernel104_ia32_sse.nb104_threadloop
_nb_kernel104_ia32_sse.nb104_end: 
        emms

        movl nb104_nouter(%esp),%eax
        movl nb104_ninner(%esp),%ebx
        movl nb104_outeriter(%ebp),%ecx
        movl nb104_inneriter(%ebp),%edx
        movl %eax,(%ecx)
        movl %ebx,(%edx)

        movl nb104_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_kernel104nf_ia32_sse
.globl _nb_kernel104nf_ia32_sse
nb_kernel104nf_ia32_sse:        
_nb_kernel104nf_ia32_sse:       
.set nb104nf_p_nri, 8
.set nb104nf_iinr, 12
.set nb104nf_jindex, 16
.set nb104nf_jjnr, 20
.set nb104nf_shift, 24
.set nb104nf_shiftvec, 28
.set nb104nf_fshift, 32
.set nb104nf_gid, 36
.set nb104nf_pos, 40
.set nb104nf_faction, 44
.set nb104nf_charge, 48
.set nb104nf_p_facel, 52
.set nb104nf_p_krf, 56
.set nb104nf_p_crf, 60
.set nb104nf_Vc, 64
.set nb104nf_type, 68
.set nb104nf_p_ntype, 72
.set nb104nf_vdwparam, 76
.set nb104nf_Vvdw, 80
.set nb104nf_p_tabscale, 84
.set nb104nf_VFtab, 88
.set nb104nf_invsqrta, 92
.set nb104nf_dvda, 96
.set nb104nf_p_gbtabscale, 100
.set nb104nf_GBtab, 104
.set nb104nf_p_nthreads, 108
.set nb104nf_count, 112
.set nb104nf_mtx, 116
.set nb104nf_outeriter, 120
.set nb104nf_inneriter, 124
.set nb104nf_work, 128
        ## stack offsets for local variables  
        ## bottom of stack is cache-aligned for sse use         
.set nb104nf_ixH1, 0
.set nb104nf_iyH1, 16
.set nb104nf_izH1, 32
.set nb104nf_ixH2, 48
.set nb104nf_iyH2, 64
.set nb104nf_izH2, 80
.set nb104nf_ixM, 96
.set nb104nf_iyM, 112
.set nb104nf_izM, 128
.set nb104nf_jxH1, 144
.set nb104nf_jyH1, 160
.set nb104nf_jzH1, 176
.set nb104nf_jxH2, 192
.set nb104nf_jyH2, 208
.set nb104nf_jzH2, 224
.set nb104nf_jxM, 240
.set nb104nf_jyM, 256
.set nb104nf_jzM, 272
.set nb104nf_dxMM, 288
.set nb104nf_dyMM, 304
.set nb104nf_dzMM, 320
.set nb104nf_qqHH, 336
.set nb104nf_qqMH, 352
.set nb104nf_qqMM, 368
.set nb104nf_vctot, 384
.set nb104nf_half, 400
.set nb104nf_three, 416
.set nb104nf_rsqH1H1, 432
.set nb104nf_rsqH1H2, 448
.set nb104nf_rsqH1M, 464
.set nb104nf_rsqH2H1, 480
.set nb104nf_rsqH2H2, 496
.set nb104nf_rsqH2M, 512
.set nb104nf_rsqMH1, 528
.set nb104nf_rsqMH2, 544
.set nb104nf_rsqMM, 560
.set nb104nf_rinvH1H1, 576
.set nb104nf_rinvH1H2, 592
.set nb104nf_rinvH1M, 608
.set nb104nf_rinvH2H1, 624
.set nb104nf_rinvH2H2, 640
.set nb104nf_rinvH2M, 656
.set nb104nf_rinvMH1, 672
.set nb104nf_rinvMH2, 688
.set nb104nf_rinvMM, 704
.set nb104nf_is3, 720
.set nb104nf_ii3, 724
.set nb104nf_innerjjnr, 728
.set nb104nf_innerk, 732
.set nb104nf_n, 736
.set nb104nf_nn1, 740
.set nb104nf_nri, 744
.set nb104nf_nouter, 748
.set nb104nf_ninner, 752
.set nb104nf_salign, 756
        pushl %ebp
        movl %esp,%ebp
        pushl %eax
        pushl %ebx
        pushl %ecx
        pushl %edx
        pushl %esi
        pushl %edi
        subl $760,%esp          ## local stack space 
        movl %esp,%eax
        andl $0xf,%eax
        subl %eax,%esp
        movl %eax,nb104nf_salign(%esp)

        emms

        ## Move args passed by reference to stack
        movl nb104nf_p_nri(%ebp),%ecx
        movl (%ecx),%ecx
        movl %ecx,nb104nf_nri(%esp)

        ## zero iteration counters
        movl $0,%eax
        movl %eax,nb104nf_nouter(%esp)
        movl %eax,nb104nf_ninner(%esp)