nb_kernel134_ia32_sse.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,068 行 · 第 1/5 页

        subps   %xmm0,%xmm3
        subps   %xmm1,%xmm4
        subps   %xmm2,%xmm5
        movaps  %xmm3,nb134_fjxO(%esp)
        movaps  %xmm4,nb134_fjyO(%esp)
        movaps  %xmm5,nb134_fjzO(%esp)
        addps   nb134_fixH2(%esp),%xmm0
        addps   nb134_fiyH2(%esp),%xmm1
        addps   nb134_fizH2(%esp),%xmm2
        movaps  %xmm0,nb134_fixH2(%esp)
        movaps  %xmm1,nb134_fiyH2(%esp)
        movaps  %xmm2,nb134_fizH2(%esp)

        movl    nb134_faction(%ebp),%esi
        ## update j water forces from local variables.
        ## transpose back first
        movaps  nb134_fjxO(%esp),%xmm0   ## Ox H1x H2x Mx 
        movaps  nb134_fjyO(%esp),%xmm1   ## Oy H1y H2y My
        movaps  nb134_fjzO(%esp),%xmm2   ## Oz H1z H2z Mz

        movaps  %xmm0,%xmm3
        movaps  %xmm0,%xmm4
        unpcklps %xmm1,%xmm3            ## Ox Oy - -
        shufps $0x1,%xmm2,%xmm4        ## h1x - Oz -
        movaps  %xmm1,%xmm5
        movaps  %xmm0,%xmm6
        unpcklps %xmm2,%xmm5            ## - - H1y H1z
        unpckhps %xmm1,%xmm6            ## h2x h2y - - 
        unpckhps %xmm2,%xmm1            ## - - My Mz

        shufps  $50,%xmm0,%xmm2 ## (00110010) h2z - Mx -
        shufps  $36,%xmm4,%xmm3 ## constant 00100100 ;# Ox Oy Oz H1x 
        shufps  $78,%xmm6,%xmm5 ## constant 01001110 ;# h1y h1z h2x h2y
        shufps  $232,%xmm1,%xmm2 ## constant 11101000 ;# h2z mx my mz

        movlps  (%esi,%eax,4),%xmm0
        movlps  16(%esi,%eax,4),%xmm1
        movlps  32(%esi,%eax,4),%xmm4
        movhps  8(%esi,%eax,4),%xmm0
        movhps  24(%esi,%eax,4),%xmm1
        movhps  40(%esi,%eax,4),%xmm4
        addps   %xmm3,%xmm0
        addps   %xmm5,%xmm1
        addps   %xmm2,%xmm4
        movlps   %xmm0,(%esi,%eax,4)
        movlps   %xmm1,16(%esi,%eax,4)
        movlps   %xmm4,32(%esi,%eax,4)
        movhps   %xmm0,8(%esi,%eax,4)
        movhps   %xmm1,24(%esi,%eax,4)
        movhps   %xmm4,40(%esi,%eax,4)

        decl nb134_innerk(%esp)
        jz    _nb_kernel134_ia32_sse.nb134_updateouterdata
        jmp   _nb_kernel134_ia32_sse.nb134_single_loop
_nb_kernel134_ia32_sse.nb134_updateouterdata: 
        movl  nb134_ii3(%esp),%ecx
        movl  nb134_faction(%ebp),%edi
        movl  nb134_fshift(%ebp),%esi
        movl  nb134_is3(%esp),%edx

        ## accumulate  Oi forces in xmm0, xmm1, xmm2 
        movaps nb134_fixO(%esp),%xmm0
        movaps nb134_fiyO(%esp),%xmm1
        movaps nb134_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 nb134_fixH1(%esp),%xmm0
        movaps nb134_fiyH1(%esp),%xmm1
        movaps nb134_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 nb134_fixH2(%esp),%xmm0
        movaps nb134_fiyH2(%esp),%xmm1
        movaps nb134_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 Mi forces in xmm0, xmm1, xmm2 
        movaps nb134_fixM(%esp),%xmm0
        movaps nb134_fiyM(%esp),%xmm1
        movaps nb134_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 nb134_n(%esp),%esi
        ## get group index for i particle 
        movl  nb134_gid(%ebp),%edx              ## base of gid[]
        movl  (%edx,%esi,4),%edx                ## ggid=gid[n]

        ## accumulate total potential energy and update it 
        movaps nb134_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  nb134_Vc(%ebp),%eax
        addss (%eax,%edx,4),%xmm7
        ## move back to mem 
        movss %xmm7,(%eax,%edx,4)

        ## accumulate total lj energy and update it 
        movaps nb134_Vvdwtot(%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  nb134_Vvdw(%ebp),%eax
        addss (%eax,%edx,4),%xmm7
        ## move back to mem 
        movss %xmm7,(%eax,%edx,4)

        ## finish if last 
        movl nb134_nn1(%esp),%ecx
        ## esi already loaded with n
        incl %esi
        subl %esi,%ecx
        jz _nb_kernel134_ia32_sse.nb134_outerend

        ## not last, iterate outer loop once more!  
        movl %esi,nb134_n(%esp)
        jmp _nb_kernel134_ia32_sse.nb134_outer
_nb_kernel134_ia32_sse.nb134_outerend: 
        ## check if more outer neighborlists remain
        movl  nb134_nri(%esp),%ecx
        ## esi already loaded with n above
        subl  %esi,%ecx
        jz _nb_kernel134_ia32_sse.nb134_end
        ## non-zero, do one more workunit
        jmp   _nb_kernel134_ia32_sse.nb134_threadloop
_nb_kernel134_ia32_sse.nb134_end: 
        emms

        movl nb134_nouter(%esp),%eax
        movl nb134_ninner(%esp),%ebx
        movl nb134_outeriter(%ebp),%ecx
        movl nb134_inneriter(%ebp),%edx
        movl %eax,(%ecx)
        movl %ebx,(%edx)

        movl nb134_salign(%esp),%eax
        addl %eax,%esp
        addl $1848,%esp
        popl %edi
        popl %esi
        popl %edx
        popl %ecx
        popl %ebx
        popl %eax
        leave
        ret



.globl nb_kernel134nf_ia32_sse
.globl _nb_kernel134nf_ia32_sse
nb_kernel134nf_ia32_sse:        
_nb_kernel134nf_ia32_sse:       
.set nb134nf_p_nri, 8
.set nb134nf_iinr, 12
.set nb134nf_jindex, 16
.set nb134nf_jjnr, 20
.set nb134nf_shift, 24
.set nb134nf_shiftvec, 28
.set nb134nf_fshift, 32
.set nb134nf_gid, 36
.set nb134nf_pos, 40
.set nb134nf_faction, 44
.set nb134nf_charge, 48
.set nb134nf_p_facel, 52
.set nb134nf_argkrf, 56
.set nb134nf_argcrf, 60
.set nb134nf_Vc, 64
.set nb134nf_type, 68
.set nb134nf_p_ntype, 72
.set nb134nf_vdwparam, 76
.set nb134nf_Vvdw, 80
.set nb134nf_p_tabscale, 84
.set nb134nf_VFtab, 88
.set nb134nf_invsqrta, 92
.set nb134nf_dvda, 96
.set nb134nf_p_gbtabscale, 100
.set nb134nf_GBtab, 104
.set nb134nf_p_nthreads, 108
.set nb134nf_count, 112
.set nb134nf_mtx, 116
.set nb134nf_outeriter, 120
.set nb134nf_inneriter, 124
.set nb134nf_work, 128
        ## stack offsets for local variables  
        ## bottom of stack is cache-aligned for sse use 
.set nb134nf_ixO, 0
.set nb134nf_iyO, 16
.set nb134nf_izO, 32
.set nb134nf_ixH1, 48
.set nb134nf_iyH1, 64
.set nb134nf_izH1, 80
.set nb134nf_ixH2, 96
.set nb134nf_iyH2, 112
.set nb134nf_izH2, 128
.set nb134nf_ixM, 144
.set nb134nf_iyM, 160
.set nb134nf_izM, 176
.set nb134nf_jxO, 192
.set nb134nf_jyO, 208
.set nb134nf_jzO, 224
.set nb134nf_jxH1, 240
.set nb134nf_jyH1, 256
.set nb134nf_jzH1, 272
.set nb134nf_jxH2, 288
.set nb134nf_jyH2, 304
.set nb134nf_jzH2, 320
.set nb134nf_jxM, 336
.set nb134nf_jyM, 352
.set nb134nf_jzM, 368
.set nb134nf_qqMM, 384
.set nb134nf_qqMH, 400
.set nb134nf_qqHH, 416
.set nb134nf_c6, 432
.set nb134nf_c12, 448
.set nb134nf_tsc, 464
.set nb134nf_vctot, 480
.set nb134nf_Vvdwtot, 496
.set nb134nf_three, 512
.set nb134nf_rsqOO, 528
.set nb134nf_rsqH1H1, 544
.set nb134nf_rsqH1H2, 560
.set nb134nf_rsqH1M, 576
.set nb134nf_rsqH2H1, 592
.set nb134nf_rsqH2H2, 608
.set nb134nf_rsqH2M, 624
.set nb134nf_rsqMH1, 640
.set nb134nf_rsqMH2, 656
.set nb134nf_rsqMM, 672
.set nb134nf_rinvOO, 688
.set nb134nf_rinvH1H1, 704
.set nb134nf_rinvH1H2, 720
.set nb134nf_rinvH1M, 736
.set nb134nf_rinvH2H1, 752
.set nb134nf_rinvH2H2, 768
.set nb134nf_rinvH2M, 784
.set nb134nf_rinvMH1, 800
.set nb134nf_rinvMH2, 816
.set nb134nf_rinvMM, 832
.set nb134nf_half, 880
.set nb134nf_is3, 896
.set nb134nf_ii3, 900
.set nb134nf_innerjjnr, 904
.set nb134nf_innerk, 908
.set nb134nf_n, 912
.set nb134nf_nn1, 916
.set nb134nf_nri, 920
.set nb134nf_nouter, 924
.set nb134nf_ninner, 928
.set nb134nf_salign, 932
        pushl %ebp
        movl %esp,%ebp
        pushl %eax
        pushl %ebx
        pushl %ecx