nb_kernel234_x86_64_sse.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,177 行 · 第 1/5 页

    movaps %xmm0,%xmm13
    movaps %xmm1,%xmm14
    addps %xmm2,%xmm11
    addps nb234_fixH1(%rsp),%xmm0
    addps nb234_fiyH1(%rsp),%xmm1
    addps nb234_fizH1(%rsp),%xmm2

    addps %xmm3,%xmm13
    addps %xmm4,%xmm14
    addps %xmm5,%xmm11
    addps nb234_fixH2(%rsp),%xmm3
    addps nb234_fiyH2(%rsp),%xmm4
    addps nb234_fizH2(%rsp),%xmm5

    addps %xmm6,%xmm13
    addps %xmm7,%xmm14
    addps %xmm8,%xmm11
    addps nb234_fixM(%rsp),%xmm6
    addps nb234_fiyM(%rsp),%xmm7
    addps nb234_fizM(%rsp),%xmm8

    movaps %xmm0,nb234_fixH1(%rsp)
    movaps %xmm1,nb234_fiyH1(%rsp)
    movaps %xmm2,nb234_fizH1(%rsp)
    movaps %xmm3,nb234_fixH2(%rsp)
    movaps %xmm4,nb234_fiyH2(%rsp)
    movaps %xmm5,nb234_fizH2(%rsp)
    movaps %xmm6,nb234_fixM(%rsp)
    movaps %xmm7,nb234_fiyM(%rsp)
    movaps %xmm8,nb234_fizM(%rsp)

    ## xmm0 = fMx
    ## xmm1 = fMy
    ## xmm2 = fMz
    movaps %xmm13,%xmm0
    unpcklps %xmm14,%xmm13
    unpckhps %xmm14,%xmm0

    addps %xmm13,%xmm9
    addps %xmm0,%xmm10

    movhlps  %xmm11,%xmm12 ## fMzc fMzd

    movlps %xmm9,36(%rdi,%rax,4)
    movhps %xmm9,36(%rdi,%rbx,4)
    movlps %xmm10,36(%rdi,%rcx,4)
    movhps %xmm10,36(%rdi,%rdx,4)
    movss  %xmm11,44(%rdi,%rax,4)
    movss  %xmm12,44(%rdi,%rcx,4)
    shufps $1,%xmm11,%xmm11
    shufps $1,%xmm12,%xmm12
    movss  %xmm11,44(%rdi,%rbx,4)
    movss  %xmm12,44(%rdi,%rdx,4)

        ## should we do one more iteration? 
        subl $4,nb234_innerk(%rsp)
        jl    _nb_kernel234_x86_64_sse.nb234_single_check
        jmp   _nb_kernel234_x86_64_sse.nb234_unroll_loop
_nb_kernel234_x86_64_sse.nb234_single_check: 
        addl $4,nb234_innerk(%rsp)
        jnz   _nb_kernel234_x86_64_sse.nb234_single_loop
        jmp   _nb_kernel234_x86_64_sse.nb234_updateouterdata
_nb_kernel234_x86_64_sse.nb234_single_loop: 
        movq  nb234_innerjjnr(%rsp),%rdx        ## pointer to jjnr[k] 
        movl  (%rdx),%eax
        addq $4,nb234_innerjjnr(%rsp)

        movq nb234_pos(%rbp),%rsi
        lea  (%rax,%rax,2),%rax

        ## fetch j coordinates
        movlps (%rsi,%rax,4),%xmm3              ##  Ox  Oy  
        movlps 16(%rsi,%rax,4),%xmm4            ## H1y H1z 
        movlps 32(%rsi,%rax,4),%xmm5            ## H2z  Mx 
        movhps 8(%rsi,%rax,4),%xmm3             ##  Ox  Oy  Oz H1x
        movhps 24(%rsi,%rax,4),%xmm4            ## H1y H1z H2x H2y
        movhps 40(%rsi,%rax,4),%xmm5            ## H2z  Mx  My  Mz
        ## transpose
        movaps %xmm4,%xmm0
        movaps %xmm3,%xmm1
        movaps %xmm4,%xmm2
        movaps %xmm3,%xmm6
        shufps $18,%xmm5,%xmm4 ## (00010010)  h2x - Mx  - 
        shufps $193,%xmm0,%xmm3 ## (11000001)  Oy  - H1y - 
        shufps $35,%xmm5,%xmm2 ## (00100011) H2y - My  - 
        shufps $18,%xmm0,%xmm1 ## (00010010)  Oz  - H1z - 
        ##  xmm6: Ox - - H1x   xmm5: H2z - - Mz 
        shufps $140,%xmm4,%xmm6 ## (10001100) Ox H1x H2x Mx 
        shufps $136,%xmm2,%xmm3 ## (10001000) Oy H1y H2y My 
        shufps $200,%xmm5,%xmm1 ## (11001000) Oz H1z H2z Mz

        ## store all j coordinates in jO  
        movaps %xmm6,nb234_jxO(%rsp)
        movaps %xmm3,nb234_jyO(%rsp)
        movaps %xmm1,nb234_jzO(%rsp)

    movaps %xmm6,%xmm0  ## jxO
    movaps %xmm1,%xmm2  ## jzO
    movaps %xmm3,%xmm1  ## jyO
    movaps %xmm3,%xmm4  ## jyO
    movaps %xmm6,%xmm3  ## jxO
    movaps %xmm2,%xmm5  ## jzO

        ## do O and M in parallel
        subps  nb234_ixO(%rsp),%xmm0
        subps  nb234_iyO(%rsp),%xmm1
        subps  nb234_izO(%rsp),%xmm2
        subps  nb234_ixM(%rsp),%xmm3
        subps  nb234_iyM(%rsp),%xmm4
        subps  nb234_izM(%rsp),%xmm5

        movaps %xmm0,nb234_dxOO(%rsp)
        movaps %xmm1,nb234_dyOO(%rsp)
        movaps %xmm2,nb234_dzOO(%rsp)
        movaps %xmm3,nb234_dxMM(%rsp)
        movaps %xmm4,nb234_dyMM(%rsp)
        movaps %xmm5,nb234_dzMM(%rsp)

        mulps %xmm0,%xmm0
        mulps %xmm1,%xmm1
        mulps %xmm2,%xmm2
        addps %xmm1,%xmm0
        addps %xmm2,%xmm0       ## have rsq in xmm0 
        mulps %xmm3,%xmm3
        mulps %xmm4,%xmm4
        mulps %xmm5,%xmm5
        addps %xmm3,%xmm4
        addps %xmm5,%xmm4       ## have rsq in xmm4
        ## Save data 
        movaps %xmm0,nb234_rsqOO(%rsp)
        movaps %xmm4,nb234_rsqMM(%rsp)

        ## do 1/x for O
        rsqrtps %xmm0,%xmm1
        movaps  %xmm1,%xmm2
        mulps   %xmm1,%xmm1
        movaps  nb234_three(%rsp),%xmm3
        mulps   %xmm0,%xmm1     ## rsq*lu*lu 
        subps   %xmm1,%xmm3     ## constant 30-rsq*lu*lu 
        mulps   %xmm2,%xmm3     ## lu*(3-rsq*lu*lu) 
        mulps   nb234_half(%rsp),%xmm3
        movaps  %xmm3,nb234_rinvOO(%rsp)        ## rinvH2 

        ## 1/sqrt(x) for M
        rsqrtps %xmm4,%xmm5
        movaps  %xmm5,%xmm6
        mulps   %xmm5,%xmm5
        movaps  nb234_three(%rsp),%xmm7
        mulps   %xmm4,%xmm5
        subps   %xmm5,%xmm7
        mulps   %xmm6,%xmm7
        mulps   nb234_half(%rsp),%xmm7   ## rinv iH1 - j water 
        movaps %xmm7,nb234_rinvMM(%rsp)


        ## LJ table interaction
        movaps nb234_rinvOO(%rsp),%xmm0
        movss %xmm0,%xmm1
        mulss  nb234_rsqOO(%rsp),%xmm1   ## xmm1=r 
        mulss  nb234_tsc(%rsp),%xmm1

    cvttps2pi %xmm1,%mm6
    cvtpi2ps %mm6,%xmm3
        subss    %xmm3,%xmm1    ## xmm1=eps 
    movss %xmm1,%xmm2
    mulss  %xmm2,%xmm2      ## xmm2=eps2 
    pslld $3,%mm6

    movq nb234_VFtab(%rbp),%rsi
    movd %mm6,%r8d

    ## dispersion 
    movlps (%rsi,%r8,4),%xmm5
    movaps %xmm5,%xmm4
    shufps $136,%xmm7,%xmm4 ## constant 10001000
    shufps $221,%xmm7,%xmm5 ## constant 11011101

    movlps 8(%rsi,%r8,4),%xmm7
    movaps %xmm7,%xmm6
    shufps $136,%xmm3,%xmm6 ## constant 10001000
    shufps $221,%xmm3,%xmm7 ## constant 11011101
    ## dispersion table ready, in xmm4-xmm7 
    mulss  %xmm1,%xmm6      ## xmm6=Geps 
    mulss  %xmm2,%xmm7      ## xmm7=Heps2 
    addss  %xmm6,%xmm5
    addss  %xmm7,%xmm5      ## xmm5=Fp 
    mulss  nb234_two(%rsp),%xmm7         ## two*Heps2 
    addss  %xmm6,%xmm7
    addss  %xmm5,%xmm7 ## xmm7=FF 
    mulss  %xmm1,%xmm5 ## xmm5=eps*Fp 
    addss  %xmm4,%xmm5 ## xmm5=VV 

    movss nb234_c6(%rsp),%xmm4
    mulss  %xmm4,%xmm7   ## fijD 
    mulss  %xmm4,%xmm5   ## Vvdw6 
        xorps  %xmm3,%xmm3

        mulps  nb234_tsc(%rsp),%xmm7
        subss  %xmm7,%xmm3
        movss  %xmm3,nb234_fstmp(%rsp)

    addss  nb234_Vvdwtot(%rsp),%xmm5
    movss %xmm5,nb234_Vvdwtot(%rsp)

    ## repulsion 
    movlps 16(%rsi,%r8,4),%xmm5
    movaps %xmm5,%xmm4
    shufps $136,%xmm7,%xmm4 ## constant 10001000
    shufps $221,%xmm7,%xmm5 ## constant 11011101

    movlps 24(%rsi,%r8,4),%xmm7
    movaps %xmm7,%xmm6
    shufps $136,%xmm3,%xmm6 ## constant 10001000
    shufps $221,%xmm3,%xmm7 ## constant 11011101
    ## table ready, in xmm4-xmm7 
    mulss  %xmm1,%xmm6      ## xmm6=Geps 
    mulss  %xmm2,%xmm7      ## xmm7=Heps2 
    addss  %xmm6,%xmm5
    addss  %xmm7,%xmm5      ## xmm5=Fp 
    mulss  nb234_two(%rsp),%xmm7         ## two*Heps2 
    addss  %xmm6,%xmm7
    addss  %xmm5,%xmm7 ## xmm7=FF 
    mulss  %xmm1,%xmm5 ## xmm5=eps*Fp 
    addss  %xmm4,%xmm5 ## xmm5=VV 

    movss nb234_c12(%rsp),%xmm4
    mulss  %xmm4,%xmm7 ## fijR 
    mulss  %xmm4,%xmm5 ## Vvdw12 
        movaps nb234_fstmp(%rsp),%xmm3
        mulss  nb234_tsc(%rsp),%xmm7
        subss  %xmm7,%xmm3

    addss  nb234_Vvdwtot(%rsp),%xmm5
    movss %xmm5,nb234_Vvdwtot(%rsp)

        mulss %xmm3,%xmm0
        movaps %xmm0,%xmm1
        movaps %xmm0,%xmm2

        mulss  nb234_dxOO(%rsp),%xmm0
        mulss  nb234_dyOO(%rsp),%xmm1
        mulss  nb234_dzOO(%rsp),%xmm2
        xorps   %xmm3,%xmm3
        xorps   %xmm4,%xmm4
        xorps   %xmm5,%xmm5
        addss   %xmm0,%xmm3
        addss   %xmm1,%xmm4
        addss   %xmm2,%xmm5
        movaps  %xmm3,nb234_fjxO(%rsp)
        movaps  %xmm4,nb234_fjyO(%rsp)
        movaps  %xmm5,nb234_fjzO(%rsp)
        addss   nb234_fixO(%rsp),%xmm0
        addss   nb234_fiyO(%rsp),%xmm1
        addss   nb234_fizO(%rsp),%xmm2
        movss  %xmm0,nb234_fixO(%rsp)
        movss  %xmm1,nb234_fiyO(%rsp)
        movss  %xmm2,nb234_fizO(%rsp)

        ## do  M coulomb interaction
        movaps nb234_rinvMM(%rsp),%xmm0
        movaps %xmm0,%xmm7      ## xmm7=rinv 
        movaps nb234_krf(%rsp),%xmm5
        mulps  %xmm0,%xmm0      ## xmm0=rinvsq 

        ## fetch charges to xmm3 (temporary) 
        xorps  %xmm3,%xmm3
        movss   nb234_qqMH(%rsp),%xmm3
        movhps  nb234_qqMM(%rsp),%xmm3
        shufps $193,%xmm3,%xmm3 ## constant 11000001 

        mulps  nb234_rsqMM(%rsp),%xmm5   ## xmm5=krsq 
        movaps %xmm5,%xmm6
        addps  %xmm7,%xmm6      ## xmm6=rinv+ krsq 
        subps  nb234_crf(%rsp),%xmm6
        mulps  %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) 
        mulps nb234_two(%rsp),%xmm5
        subps  %xmm5,%xmm7      ## xmm7=rinv-2*krsq 
        mulps  %xmm3,%xmm7 ## xmm7 = coul part of fscal 

        addps  nb234_vctot(%rsp),%xmm6
        movaps %xmm6,nb234_vctot(%rsp)
        mulps  %xmm7,%xmm0

        movaps %xmm0,%xmm1
        movaps %xmm0,%xmm2

        mulps   nb234_dxMM(%rsp),%xmm0
        mulps   nb234_dyMM(%rsp),%xmm1
        mulps   nb234_dzMM(%rsp),%xmm2
        ## update forces M - j water 
        movaps  nb234_fjxO(%rsp),%xmm3
        movaps  nb234_fjyO(%rsp),%xmm4
        movaps  nb234_fjzO(%rsp),%xmm5
        addps   %xmm0,%xmm3
        addps   %xmm1,%xmm4
        addps   %xmm2,%xmm5
        movaps  %xmm3,nb234_fjxO(%rsp)
        movaps  %xmm4,nb234_fjyO(%rsp)
        movaps  %xmm5,nb234_fjzO(%rsp)
        addps   nb234_fixM(%rsp),%xmm0
        addps   nb234_fiyM(%rsp),%xmm1
        addps   nb234_fizM(%rsp),%xmm2
        movaps  %xmm0,nb234_fixM(%rsp)
        movaps  %xmm1,nb234_fiyM(%rsp)
        movaps  %xmm2,nb234_fizM(%rsp)

        ## i H1 & H2 simultaneously first get i particle coords: 
    movaps  nb234_jxO(%rsp),%xmm0
    movaps  nb234_jyO(%rsp),%xmm1
    movaps  nb234_jzO(%rsp),%xmm2
    movaps  %xmm0,%xmm3
    movaps  %xmm1,%xmm4
    movaps  %xmm2,%xmm5
        subps   nb234_ixH1(%rsp),%xmm0
        subps   nb234_iyH1(%rsp),%xmm1
        subps   nb234_izH1(%rsp),%xmm2
        subps   nb234_ixH2(%rsp),%xmm3
        subps   nb234_iyH2(%rsp),%xmm4
        subps   nb234_izH2(%rsp),%xmm5
        movaps %xmm0,nb234_dxH1H1(%rsp)
        movaps %xmm1,nb234_dyH1H1(%rsp)
        movaps %xmm2,nb234_dzH1H1(%rsp)
        movaps %xmm3,nb234_dxH2H2(%rsp)
        movaps %xmm4,nb234_dyH2H2(%rsp)
        movaps %xmm5,nb234_dzH2H2(%rsp)
        mulps %xmm0,%xmm0
        mulps %xmm1,%xmm1
        mulps %xmm2,%xmm2
        mulps %xmm3,%xmm3
        mulps %xmm4,%xmm4
        mulps %xmm5,%xmm5
        addps %xmm1,%xmm0
        addps %xmm3,%xmm4
        addps %xmm2,%xmm0       ## have rsqH1 in xmm0 
        addps %xmm5,%xmm4       ## have rsqH2 in xmm4 
        movaps  %xmm0,nb234_rsqH1H1(%rsp)
        movaps  %xmm4,nb234_rsqH2H2(%rsp)

        ## start doing invsqrt use rsq values in xmm0, xmm4 
        rsqrtps %xmm0,%xmm1
        rsqrtps %xmm4,%xmm5
        movaps  %xmm1,%xmm2
        movaps  %xmm5,%xmm6
        mulps   %xmm1,%xmm1
        mulps   %xmm5,%xmm5
        movaps  nb234_three(%rsp),%xmm3
        movaps  %xmm3,%xmm7
        mulps   %xmm0,%xmm1
        mulps   %xmm4,%xmm5
        subps   %xmm1,%xmm3
        subps   %xmm5,%xmm7
        mulps   %xmm2,%xmm3
        mulps   %xmm6,%xmm7
        mulps   nb234_half(%rsp),%xmm3   ## rinvH1H1
        mulps   nb234_half(%rsp),%xmm7   ## rinvH2H2
        movaps  %xmm3,nb234_rinvH1H1(%rsp)
        movaps  %xmm7,nb234_rinvH2H2(%rsp)

        ## Do H1 coulomb interaction
        movaps nb234_rinvH1H1(%rsp),%xmm0
        movaps %xmm0,%xmm7      ## xmm7=rinv 
        movaps nb234_krf(%rsp),%xmm5
        mulps  %xmm0,%xmm0      ## xmm0=rinvsq 

        ## fetch charges to xmm3 (temporary) 
        xorps  %xmm3,%xmm3
        movss   nb234_qqHH(%rsp),%xmm3
        movhps  nb234_qqMH(%rsp),%xmm3
        shufps $193,%xmm3,%xmm3 ## constant 11000001 

        mulps  nb234_rsqH1H1(%rsp),%xmm5   ## xmm5=krsq 
        movaps %xmm5,%xmm6
        addps  %xmm7,%xmm6      ## xmm6=rinv+ krsq 
        subps  nb234_crf(%rsp),%xmm6
        mulps  %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) 
        mulps nb234_two(%rsp),%xmm5
        subps  %xmm5,%xmm7      ## xmm7=rinv-2*krsq 
        mulps  %xmm3,%xmm7 ## xmm7 = coul part of fscal 

        addps  nb234_vctot(%rsp),%xmm6
        movaps %xmm6,nb234_vctot(%rsp)

        mulps  %xmm7,%xmm0

        movaps %xmm0,%xmm1
        movaps %xmm0,%xmm2

        mulps   nb234_dxH1H1(%rsp),%xmm0
        mulps   nb234_dyH1H1(%rsp),%xmm1
        mulps   nb234_dzH1H1(%rsp),%xmm2
        ## update forces H1 - j water 
        movaps  nb234_fjxO(%rsp),%xmm3
        movaps  nb234_fjyO(%rsp),%xmm4
        movaps  nb234_fjzO(%rsp),%xmm5
        addps   %xmm0,%xmm3
        addps   %xmm1,%xmm4
        addps   %xmm2,%xmm5
        movaps  %xmm3,nb234_fjxO(%rsp)
        movaps  %xmm4,nb234_fjyO(%rsp)
        movaps  %xmm5,nb234_fjzO(%rsp)
        addps   nb234_fixH1(%rsp),%xmm0
        addps   nb234_fiyH1(%rsp),%xmm1
        addps   nb234_fizH1(%rsp),%xmm2
        movaps  %xmm0,nb234_fixH1(%rsp)
        movaps  %xmm1,nb234_fiyH1(%rsp)
        movaps  %xmm2,nb234_fizH1(%rsp)

        ## H2 Coulomb
        movaps nb234_rinvH2H2(%rsp),%xmm0
        movaps %xmm0,%xmm7      ## xmm7=rinv 
        movaps nb234_krf(%rsp),%xmm5
        mulps  %xmm0,%xmm0      ## xmm0=rinvsq 

        ## fetch charges to xmm3 (temporary) 
        xorps  %xmm3,%xmm3
        movss   nb234_qqHH(%rsp),%xmm3
        movhps  nb234_qqMH(%rsp),%xmm3
        shufps $193,%xmm3,%xmm3 ## constant 11000001 

        mulps  nb234_rsqH2H2(%rsp),%xmm5   ## xmm5=krsq 
        movaps %xmm5,%xmm6
        addps  %xmm7,%xmm6      ## xmm6=rinv+ krsq 
        subps  nb234_crf(%rsp),%xmm6
        mulps  %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) 
        mulps nb234_two(%rsp),%xmm5
        subps  %xmm5,%xmm7      ## xmm7=rinv-2*krsq 
        mulps  %xmm3,%xmm7 ## xmm7 = coul part of fscal 

        addps  nb234_vctot(%rsp),%xmm6   ## local vctot summation variable
        movaps %xmm6,nb234_vctot(%rsp)
        mulps  %xmm7,%xmm0

        movaps %xmm0,%xmm1
        movaps %xmm0,%xmm2