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

    movhlps  %xmm11,%xmm12 ## fH1zc fH1zd

    movlps %xmm9,12(%rdi,%rax,4)
    movhps %xmm9,12(%rdi,%rbx,4)
    movlps %xmm10,12(%rdi,%rcx,4)
    movhps %xmm10,12(%rdi,%rdx,4)
    movss  %xmm11,20(%rdi,%rax,4)
    movss  %xmm12,20(%rdi,%rcx,4)
    shufps $1,%xmm11,%xmm11
    shufps $1,%xmm12,%xmm12
    movss  %xmm11,20(%rdi,%rbx,4)
    movss  %xmm12,20(%rdi,%rdx,4)

        ## move j H2 coordinates to local temp variables 
        movq nb234_pos(%rbp),%rsi       ## base of pos[] 
    movlps 24(%rsi,%rax,4),%xmm0    ## jxH2a jyH2a  -   -
    movlps 24(%rsi,%rcx,4),%xmm1    ## jxH2c jyH2c  -   -
    movhps 24(%rsi,%rbx,4),%xmm0    ## jxH2a jyH2a jxH2b jyH2b 
    movhps 24(%rsi,%rdx,4),%xmm1    ## jxH2c jyH2c jxH2d jyH2d 

    movss  32(%rsi,%rax,4),%xmm2    ## jzH2a  -  -  -
    movss  32(%rsi,%rcx,4),%xmm3    ## jzH2c  -  -  -
    movhps 32(%rsi,%rbx,4),%xmm2    ## jzH2a  -  jzH2b  -
    movhps 32(%rsi,%rdx,4),%xmm3    ## jzH2c  -  jzH2d -

    movaps %xmm0,%xmm4
    unpcklps %xmm1,%xmm0 ## jxH2a jxH2c jyH2a jyH2c        
    unpckhps %xmm1,%xmm4 ## jxH2b jxH2d jyH2b jyH2d
    movaps %xmm0,%xmm1
    unpcklps %xmm4,%xmm0 ## x
    unpckhps %xmm4,%xmm1 ## y

    shufps  $136,%xmm3,%xmm2  ## 10001000 => jzH2a jzH2b jzH2c jzH2d

    ## xmm0 = H2x
    ## xmm1 = H2y
    ## xmm2 = H2z

    movaps %xmm0,%xmm3
    movaps %xmm1,%xmm4
    movaps %xmm2,%xmm5
    movaps %xmm0,%xmm6
    movaps %xmm1,%xmm7
    movaps %xmm2,%xmm8


    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
    subps nb234_ixM(%rsp),%xmm6
    subps nb234_iyM(%rsp),%xmm7
    subps nb234_izM(%rsp),%xmm8

        movaps %xmm0,nb234_dxH1H2(%rsp)
        movaps %xmm1,nb234_dyH1H2(%rsp)
        movaps %xmm2,nb234_dzH1H2(%rsp)
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        movaps %xmm3,nb234_dxH2H2(%rsp)
        movaps %xmm4,nb234_dyH2H2(%rsp)
        movaps %xmm5,nb234_dzH2H2(%rsp)
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        movaps %xmm6,nb234_dxMH2(%rsp)
        movaps %xmm7,nb234_dyMH2(%rsp)
        movaps %xmm8,nb234_dzMH2(%rsp)
        mulps  %xmm6,%xmm6
        mulps  %xmm7,%xmm7
        mulps  %xmm8,%xmm8
        addps  %xmm1,%xmm0
        addps  %xmm2,%xmm0
        addps  %xmm4,%xmm3
        addps  %xmm5,%xmm3
    addps  %xmm7,%xmm6
    addps  %xmm8,%xmm6

        ## start doing invsqrt for jH2 atoms
        rsqrtps %xmm0,%xmm1
        rsqrtps %xmm3,%xmm4
    rsqrtps %xmm6,%xmm7

        movaps  %xmm1,%xmm2
        movaps  %xmm4,%xmm5
    movaps  %xmm7,%xmm8

        mulps   %xmm1,%xmm1 ## lu*lu
        mulps   %xmm4,%xmm4 ## lu*lu
    mulps   %xmm7,%xmm7 ## lu*lu

        movaps  nb234_three(%rsp),%xmm9
        movaps  %xmm9,%xmm10
    movaps  %xmm9,%xmm11

        mulps   %xmm0,%xmm1 ## rsq*lu*lu
        mulps   %xmm3,%xmm4 ## rsq*lu*lu 
    mulps   %xmm6,%xmm7 ## rsq*lu*lu

        subps   %xmm1,%xmm9
        subps   %xmm4,%xmm10
    subps   %xmm7,%xmm11 ## 3-rsq*lu*lu

        mulps   %xmm2,%xmm9
        mulps   %xmm5,%xmm10
    mulps   %xmm8,%xmm11 ## lu*(3-rsq*lu*lu)

        movaps  nb234_half(%rsp),%xmm4
        mulps   %xmm4,%xmm9 ## rinvH1H2
        mulps   %xmm4,%xmm10 ## rinvH2H2
    mulps   %xmm4,%xmm11 ## rinvMH2

        ## H2 interactions 
    ## rsq in xmm0,xmm3,xmm6  
    ## rinv in xmm9, xmm10, xmm11

    movaps %xmm9,%xmm1 ## copy of rinv
    movaps %xmm10,%xmm4
    movaps %xmm11,%xmm7
    movaps nb234_krf(%rsp),%xmm2
    mulps  %xmm9,%xmm9  ## rinvsq
    mulps  %xmm10,%xmm10
    mulps  %xmm11,%xmm11
    mulps  %xmm2,%xmm0 ## k*rsq
    mulps  %xmm2,%xmm3
    mulps  %xmm2,%xmm6
    movaps %xmm0,%xmm2 ## copy of k*rsq
    movaps %xmm3,%xmm5
    movaps %xmm6,%xmm8
    addps  %xmm1,%xmm2 ## rinv+krsq
    addps  %xmm4,%xmm5
    addps  %xmm7,%xmm8
    movaps nb234_crf(%rsp),%xmm14
    subps  %xmm14,%xmm2  ## rinv+krsq-crf
    subps  %xmm14,%xmm5
    subps  %xmm14,%xmm8
    movaps nb234_qqHH(%rsp),%xmm12
    movaps nb234_qqMH(%rsp),%xmm13
    mulps  %xmm12,%xmm2 ## xmm6=voul=qq*(rinv+ krsq-crf)
    mulps  %xmm12,%xmm5 ## xmm6=voul=qq*(rinv+ krsq-crf)
    mulps  %xmm13,%xmm8 ## xmm6=voul=qq*(rinv+ krsq-crf)
    addps  %xmm0,%xmm0 ## 2*krsq
    addps  %xmm3,%xmm3
    addps  %xmm6,%xmm6
    subps  %xmm0,%xmm1 ## rinv-2*krsq
    subps  %xmm3,%xmm4
    subps  %xmm6,%xmm7
    mulps  %xmm12,%xmm1  ## (rinv-2*krsq)*qq
    mulps  %xmm12,%xmm4
    mulps  %xmm13,%xmm7
    addps  %xmm2,%xmm15
    addps  %xmm8,%xmm5
    addps  %xmm5,%xmm15

    mulps  %xmm9,%xmm1  ## fscal
    mulps  %xmm10,%xmm4
    mulps  %xmm11,%xmm7

        ## move j H2 forces to local temp variables 
        movq nb234_faction(%rbp),%rdi
    movlps 24(%rdi,%rax,4),%xmm9    ## jxH2a jyH2a  -   -
    movlps 24(%rdi,%rcx,4),%xmm10    ## jxH2c jyH2c  -   -
    movhps 24(%rdi,%rbx,4),%xmm9    ## jxH2a jyH2a jxH2b jyH2b 
    movhps 24(%rdi,%rdx,4),%xmm10    ## jxH2c jyH2c jxH2d jyH2d 

    movss  32(%rdi,%rax,4),%xmm11    ## jzH2a  -  -  -
    movss  32(%rdi,%rcx,4),%xmm12    ## jzH2c  -  -  -
    movhps 32(%rdi,%rbx,4),%xmm11    ## jzH2a  -  jzH2b  -
    movhps 32(%rdi,%rdx,4),%xmm12    ## jzH2c  -  jzH2d -

    shufps $136,%xmm12,%xmm11 ## 10001000 => jzH2a jzH2b jzH2c jzH2d

    ## xmm9: jxH2a jyH2a jxH2b jyH2b 
    ## xmm10: jxH2c jyH2c jxH2d jyH2d
    ## xmm11: jzH2a jzH2b jzH2c jzH2d

    movaps %xmm1,%xmm0
    movaps %xmm1,%xmm2
    movaps %xmm4,%xmm3
    movaps %xmm4,%xmm5
    movaps %xmm7,%xmm6
    movaps %xmm7,%xmm8

        mulps nb234_dxH1H2(%rsp),%xmm0
        mulps nb234_dyH1H2(%rsp),%xmm1
        mulps nb234_dzH1H2(%rsp),%xmm2
        mulps nb234_dxH2H2(%rsp),%xmm3
        mulps nb234_dyH2H2(%rsp),%xmm4
        mulps nb234_dzH2H2(%rsp),%xmm5
        mulps nb234_dxMH2(%rsp),%xmm6
        mulps nb234_dyMH2(%rsp),%xmm7
        mulps nb234_dzMH2(%rsp),%xmm8

    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)

    ## xmm9  = fH2x
    ## xmm10 = fH2y
    ## xmm11 = fH2z
    movaps %xmm13,%xmm0
    unpcklps %xmm14,%xmm13
    unpckhps %xmm14,%xmm0

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

    movhlps  %xmm11,%xmm12 ## fH2zc fH2zd

    movlps %xmm9,24(%rdi,%rax,4)
    movhps %xmm9,24(%rdi,%rbx,4)
    movlps %xmm10,24(%rdi,%rcx,4)
    movhps %xmm10,24(%rdi,%rdx,4)
    movss  %xmm11,32(%rdi,%rax,4)
    movss  %xmm12,32(%rdi,%rcx,4)
    shufps $1,%xmm11,%xmm11
    shufps $1,%xmm12,%xmm12
    movss  %xmm11,32(%rdi,%rbx,4)
    movss  %xmm12,32(%rdi,%rdx,4)

        ## move j M coordinates to local temp variables 
    movlps 36(%rsi,%rax,4),%xmm0    ## jxMa jyMa  -   -
    movlps 36(%rsi,%rcx,4),%xmm1    ## jxMc jyMc  -   -
    movhps 36(%rsi,%rbx,4),%xmm0    ## jxMa jyMa jxMb jyMb 
    movhps 36(%rsi,%rdx,4),%xmm1    ## jxMc jyMc jxMd jyMd 

    movss  44(%rsi,%rax,4),%xmm2    ## jzMa  -  -  -
    movss  44(%rsi,%rcx,4),%xmm3    ## jzMc  -  -  -
    movss  44(%rsi,%rbx,4),%xmm5    ## jzMb  -  -  -
    movss  44(%rsi,%rdx,4),%xmm6    ## jzMd  -  -  -
    movlhps %xmm5,%xmm2 ## jzMa  -  jzMb  -
    movlhps %xmm6,%xmm3 ## jzMc  -  jzMd -

    movaps %xmm0,%xmm4
    unpcklps %xmm1,%xmm0 ## jxMa jxMc jyMa jyMc        
    unpckhps %xmm1,%xmm4 ## jxMb jxMd jyMb jyMd
    movaps %xmm0,%xmm1
    unpcklps %xmm4,%xmm0 ## x
    unpckhps %xmm4,%xmm1 ## y

    shufps $136,%xmm3,%xmm2 ## 10001000 => jzMa jzMb jzMc jzMd

    ## xmm0 = Mx
    ## xmm1 = My
    ## xmm2 = Mz

    movaps %xmm0,%xmm3
    movaps %xmm1,%xmm4
    movaps %xmm2,%xmm5
    movaps %xmm0,%xmm6
    movaps %xmm1,%xmm7
    movaps %xmm2,%xmm8

    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
    subps nb234_ixM(%rsp),%xmm6
    subps nb234_iyM(%rsp),%xmm7
    subps nb234_izM(%rsp),%xmm8

        movaps %xmm0,nb234_dxH1M(%rsp)
        movaps %xmm1,nb234_dyH1M(%rsp)
        movaps %xmm2,nb234_dzH1M(%rsp)
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        movaps %xmm3,nb234_dxH2M(%rsp)
        movaps %xmm4,nb234_dyH2M(%rsp)
        movaps %xmm5,nb234_dzH2M(%rsp)
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        movaps %xmm6,nb234_dxMM(%rsp)
        movaps %xmm7,nb234_dyMM(%rsp)
        movaps %xmm8,nb234_dzMM(%rsp)
        mulps  %xmm6,%xmm6
        mulps  %xmm7,%xmm7
        mulps  %xmm8,%xmm8
        addps  %xmm1,%xmm0
        addps  %xmm2,%xmm0
        addps  %xmm4,%xmm3
        addps  %xmm5,%xmm3
    addps  %xmm7,%xmm6
    addps  %xmm8,%xmm6

        ## start doing invsqrt for jM atoms
        rsqrtps %xmm0,%xmm1
        rsqrtps %xmm3,%xmm4
    rsqrtps %xmm6,%xmm7

        movaps  %xmm1,%xmm2
        movaps  %xmm4,%xmm5
    movaps  %xmm7,%xmm8

        mulps   %xmm1,%xmm1 ## lu*lu
        mulps   %xmm4,%xmm4 ## lu*lu
    mulps   %xmm7,%xmm7 ## lu*lu

        movaps  nb234_three(%rsp),%xmm9
        movaps  %xmm9,%xmm10
    movaps  %xmm9,%xmm11

        mulps   %xmm0,%xmm1 ## rsq*lu*lu
        mulps   %xmm3,%xmm4 ## rsq*lu*lu 
    mulps   %xmm6,%xmm7 ## rsq*lu*lu

        subps   %xmm1,%xmm9
        subps   %xmm4,%xmm10
    subps   %xmm7,%xmm11 ## 3-rsq*lu*lu

        mulps   %xmm2,%xmm9
        mulps   %xmm5,%xmm10
    mulps   %xmm8,%xmm11 ## lu*(3-rsq*lu*lu)

        movaps  nb234_half(%rsp),%xmm4
        mulps   %xmm4,%xmm9 ## rinvH1M
        mulps   %xmm4,%xmm10 ## rinvH2M
    mulps   %xmm4,%xmm11 ## rinvMM

        ## M interactions 
    ## rsq in xmm0,xmm3,xmm6  
    ## rinv in xmm9, xmm10, xmm11

    movaps %xmm9,%xmm1 ## copy of rinv
    movaps %xmm10,%xmm4
    movaps %xmm11,%xmm7
    movaps nb234_krf(%rsp),%xmm2
    mulps  %xmm9,%xmm9  ## rinvsq
    mulps  %xmm10,%xmm10
    mulps  %xmm11,%xmm11
    mulps  %xmm2,%xmm0 ## k*rsq
    mulps  %xmm2,%xmm3
    mulps  %xmm2,%xmm6
    movaps %xmm0,%xmm2 ## copy of k*rsq
    movaps %xmm3,%xmm5
    movaps %xmm6,%xmm8
    addps  %xmm1,%xmm2 ## rinv+krsq
    addps  %xmm4,%xmm5
    addps  %xmm7,%xmm8
    movaps nb234_crf(%rsp),%xmm14
    subps  %xmm14,%xmm2  ## rinv+krsq-crf
    subps  %xmm14,%xmm5
    subps  %xmm14,%xmm8
    movaps nb234_qqMH(%rsp),%xmm12
    movaps nb234_qqMM(%rsp),%xmm13
    mulps  %xmm12,%xmm2 ## xmm6=voul=qq*(rinv+ krsq-crf)
    mulps  %xmm12,%xmm5 ## xmm6=voul=qq*(rinv+ krsq-crf)
    mulps  %xmm13,%xmm8 ## xmm6=voul=qq*(rinv+ krsq-crf)
    addps  %xmm0,%xmm0 ## 2*krsq
    addps  %xmm3,%xmm3
    addps  %xmm6,%xmm6
    subps  %xmm0,%xmm1 ## rinv-2*krsq
    subps  %xmm3,%xmm4
    subps  %xmm6,%xmm7
    mulps  %xmm12,%xmm1  ## (rinv-2*krsq)*qq
    mulps  %xmm12,%xmm4
    mulps  %xmm13,%xmm7
    addps  %xmm8,%xmm5
    addps  %xmm15,%xmm2
    addps  %xmm5,%xmm2
    movaps %xmm2,nb234_vctot(%rsp)

    mulps  %xmm9,%xmm1  ## fscal
    mulps  %xmm10,%xmm4
    mulps  %xmm11,%xmm7

        ## move j M forces to local temp variables 
        movq nb234_faction(%rbp),%rdi
    movlps 36(%rdi,%rax,4),%xmm9    ## jxMa jyMa  -   -
    movlps 36(%rdi,%rcx,4),%xmm10    ## jxMc jyMc  -   -
    movhps 36(%rdi,%rbx,4),%xmm9    ## jxMa jyMa jxMb jyMb 
    movhps 36(%rdi,%rdx,4),%xmm10    ## jxMc jyMc jxMd jyMd 

    movss  44(%rdi,%rax,4),%xmm11    ## jzMa  -  -  -
    movss  44(%rdi,%rcx,4),%xmm12    ## jzMc  -  -  -
    movss  44(%rdi,%rbx,4),%xmm2     ## jzMb  -  -  -
    movss  44(%rdi,%rdx,4),%xmm3     ## jzMd  -  -  -
    movlhps %xmm2,%xmm11 ## jzMa  -  jzMb  -
    movlhps %xmm3,%xmm12 ## jzMc  -  jzMd -

    shufps $136,%xmm12,%xmm11 ## 10001000 => jzMa jzMb jzMc jzMd

    ## xmm9: jxMa jyMa jxMb jyMb 
    ## xmm10: jxMc jyMc jxMd jyMd
    ## xmm11: jzMa jzMb jzMc jzMd

    movaps %xmm1,%xmm0
    movaps %xmm1,%xmm2
    movaps %xmm4,%xmm3
    movaps %xmm4,%xmm5
    movaps %xmm7,%xmm6
    movaps %xmm7,%xmm8

        mulps nb234_dxH1M(%rsp),%xmm0
        mulps nb234_dyH1M(%rsp),%xmm1
        mulps nb234_dzH1M(%rsp),%xmm2
        mulps nb234_dxH2M(%rsp),%xmm3
        mulps nb234_dyH2M(%rsp),%xmm4
        mulps nb234_dzH2M(%rsp),%xmm5
        mulps nb234_dxMM(%rsp),%xmm6
        mulps nb234_dyMM(%rsp),%xmm7
        mulps nb234_dzMM(%rsp),%xmm8