nb_kernel132_x86_64_sse.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,146 行 · 第 1/5 页

    subps nb132_ixH2(%rsp),%xmm6
    subps nb132_iyH2(%rsp),%xmm7
    subps nb132_izH2(%rsp),%xmm8

        movaps %xmm0,nb132_dxOO(%rsp)
        movaps %xmm1,nb132_dyOO(%rsp)
        movaps %xmm2,nb132_dzOO(%rsp)
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        movaps %xmm3,nb132_dxH1O(%rsp)
        movaps %xmm4,nb132_dyH1O(%rsp)
        movaps %xmm5,nb132_dzH1O(%rsp)
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        movaps %xmm6,nb132_dxH2O(%rsp)
        movaps %xmm7,nb132_dyH2O(%rsp)
        movaps %xmm8,nb132_dzH2O(%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 jO 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  nb132_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  nb132_half(%rsp),%xmm2
        mulps   %xmm2,%xmm9 ## rinvOO 
        mulps   %xmm2,%xmm10 ## rinvH1O
    mulps   %xmm2,%xmm11 ## rinvH2O

        ## O interactions 
    ## rsq in xmm0,xmm3,xmm6  
    ## rinv in xmm9, xmm10, xmm11
    movaps %xmm0,nb132_rsqOO(%rsp)
    movaps %xmm3,nb132_rsqOH1(%rsp)
    movaps %xmm6,nb132_rsqOH2(%rsp)
    movaps %xmm9,nb132_rinvOO(%rsp)
    movaps %xmm10,nb132_rinvOH1(%rsp)
    movaps %xmm11,nb132_rinvOH2(%rsp)

    ## table LJ interaction
    mulps  %xmm9,%xmm0
    mulps  nb132_tsc(%rsp),%xmm0   ## rtab

    ## truncate and convert to integers
    cvttps2dq %xmm0,%xmm1

    ## convert back to float
    cvtdq2ps  %xmm1,%xmm2

    ## multiply by 8
    pslld   $3,%xmm1

    ## move to integer registers
    movhlps %xmm1,%xmm13
    movd    %xmm1,%r8d
    movd    %xmm13,%r10d
    shufps $1,%xmm1,%xmm1
    shufps $1,%xmm13,%xmm13
    movd    %xmm1,%r9d
    movd    %xmm13,%r11d

    ## calculate eps
    subps     %xmm2,%xmm0
    movq nb132_VFtab(%rbp),%rsi

    movlps (%rsi,%r8,4),%xmm5
        movlps 16(%rsi,%r8,4),%xmm9

        movlps (%rsi,%r10,4),%xmm7
        movlps 16(%rsi,%r10,4),%xmm11

        movhps (%rsi,%r9,4),%xmm5
        movhps 16(%rsi,%r9,4),%xmm9

        movhps (%rsi,%r11,4),%xmm7
        movhps 16(%rsi,%r11,4),%xmm11

    movaps %xmm5,%xmm4
    movaps %xmm9,%xmm8
        shufps $136,%xmm7,%xmm4 ## 10001000
        shufps $136,%xmm11,%xmm8 ## 10001000
        shufps $221,%xmm7,%xmm5 ## 11011101
        shufps $221,%xmm11,%xmm9 ## 11011101

        movlps 8(%rsi,%r8,4),%xmm7
        movlps 24(%rsi,%r8,4),%xmm11

        movlps 8(%rsi,%r10,4),%xmm13
        movlps 24(%rsi,%r10,4),%xmm14

        movhps 8(%rsi,%r9,4),%xmm7
        movhps 24(%rsi,%r9,4),%xmm11

        movhps 8(%rsi,%r11,4),%xmm13
        movhps 24(%rsi,%r11,4),%xmm14

    movaps %xmm7,%xmm6
    movaps %xmm11,%xmm10

        shufps $136,%xmm13,%xmm6 ## 10001000
        shufps $136,%xmm14,%xmm10 ## 10001000
        shufps $221,%xmm13,%xmm7 ## 11011101
        shufps $221,%xmm14,%xmm11 ## 11011101
    ## dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11

    mulps  %xmm0,%xmm7   ## Heps
    mulps  %xmm0,%xmm11
    mulps  %xmm0,%xmm6  ## Geps
    mulps  %xmm0,%xmm10
    mulps  %xmm0,%xmm7  ## Heps2
    mulps  %xmm0,%xmm11
    addps  %xmm6,%xmm5 ## F+Geps
    addps  %xmm10,%xmm9
    addps  %xmm7,%xmm5  ## F+Geps+Heps2 = Fp
    addps  %xmm11,%xmm9
    addps  %xmm7,%xmm7   ## 2*Heps2
    addps  %xmm11,%xmm11
    addps  %xmm6,%xmm7  ## 2*Heps2+Geps
    addps  %xmm10,%xmm11

    addps  %xmm5,%xmm7 ## FF = Fp + 2*Heps2 + Geps
    addps  %xmm9,%xmm11
    mulps  %xmm0,%xmm5 ## eps*Fp
    mulps  %xmm0,%xmm9
    movaps nb132_c6(%rsp),%xmm12
    movaps nb132_c12(%rsp),%xmm13
    addps  %xmm4,%xmm5 ## VV
    addps  %xmm8,%xmm9

    mulps  %xmm12,%xmm5 ## VV*c6 = vnb6
    mulps  %xmm13,%xmm9 ## VV*c12 = vnb12
    addps  %xmm9,%xmm5
    addps  nb132_Vvdwtot(%rsp),%xmm5
    movaps %xmm5,nb132_Vvdwtot(%rsp)

    mulps  %xmm12,%xmm7  ## FF*c6 = fnb6
    mulps  %xmm13,%xmm11  ## FF*c12  = fnb12
    addps  %xmm11,%xmm7
    mulps  nb132_tsc(%rsp),%xmm7

    movaps nb132_rinvOO(%rsp),%xmm9
    movaps nb132_rinvOH1(%rsp),%xmm10
    movaps nb132_rinvOH2(%rsp),%xmm11
    movaps %xmm9,%xmm0  ## rinv
    movaps %xmm10,%xmm1
    movaps %xmm11,%xmm2

    mulps  %xmm10,%xmm10 ## rinvsq
    mulps  %xmm11,%xmm11
    mulps  nb132_qqOO(%rsp),%xmm0
    mulps  nb132_qqOH(%rsp),%xmm1
    mulps  nb132_qqOH(%rsp),%xmm2
    mulps  %xmm0,%xmm9
    mulps  %xmm1,%xmm10
    mulps  %xmm2,%xmm11

    subps  %xmm7,%xmm9
    mulps  nb132_rinvOO(%rsp),%xmm9

    addps nb132_vctot(%rsp),%xmm0
    addps %xmm2,%xmm1
    addps %xmm1,%xmm0
    movaps %xmm0,nb132_vctot(%rsp)

        ## move j O forces to local temp variables 
        movq nb132_faction(%rbp),%rdi
    movlps (%rdi,%rax,4),%xmm0 ## jxOa jyOa  -   -
    movlps (%rdi,%rcx,4),%xmm1 ## jxOc jyOc  -   -
    movhps (%rdi,%rbx,4),%xmm0 ## jxOa jyOa jxOb jyOb 
    movhps (%rdi,%rdx,4),%xmm1 ## jxOc jyOc jxOd jyOd 

    movss  8(%rdi,%rax,4),%xmm2    ## jzOa  -  -  -
    movss  8(%rdi,%rcx,4),%xmm3    ## jzOc  -  -  -
    movhps 8(%rdi,%rbx,4),%xmm2    ## jzOa  -  jzOb -
    movhps 8(%rdi,%rdx,4),%xmm3    ## jzOc  -  jzOd -

    shufps $136,%xmm3,%xmm2 ## 10001000 => jzOa jzOb jzOc jzOd

    ## xmm0: jxOa jyOa jxOb jyOb 
    ## xmm1: jxOc jyOc jxOd jyOd
    ## xmm2: jzOa jzOb jzOc jzOd

    movaps %xmm9,%xmm7
    movaps %xmm9,%xmm8
    movaps %xmm11,%xmm13
    movaps %xmm11,%xmm14
    movaps %xmm11,%xmm15
    movaps %xmm10,%xmm11
    movaps %xmm10,%xmm12

        mulps nb132_dxOO(%rsp),%xmm7
        mulps nb132_dyOO(%rsp),%xmm8
        mulps nb132_dzOO(%rsp),%xmm9
        mulps nb132_dxH1O(%rsp),%xmm10
        mulps nb132_dyH1O(%rsp),%xmm11
        mulps nb132_dzH1O(%rsp),%xmm12
        mulps nb132_dxH2O(%rsp),%xmm13
        mulps nb132_dyH2O(%rsp),%xmm14
        mulps nb132_dzH2O(%rsp),%xmm15

    movaps %xmm7,%xmm3
    movaps %xmm8,%xmm4
    addps %xmm9,%xmm2
    addps nb132_fixO(%rsp),%xmm7
    addps nb132_fiyO(%rsp),%xmm8
    addps nb132_fizO(%rsp),%xmm9

    addps %xmm10,%xmm3
    addps %xmm11,%xmm4
    addps %xmm12,%xmm2
    addps nb132_fixH1(%rsp),%xmm10
    addps nb132_fiyH1(%rsp),%xmm11
    addps nb132_fizH1(%rsp),%xmm12

    addps %xmm13,%xmm3
    addps %xmm14,%xmm4
    addps %xmm15,%xmm2
    addps nb132_fixH2(%rsp),%xmm13
    addps nb132_fiyH2(%rsp),%xmm14
    addps nb132_fizH2(%rsp),%xmm15

    movaps %xmm7,nb132_fixO(%rsp)
    movaps %xmm8,nb132_fiyO(%rsp)
    movaps %xmm9,nb132_fizO(%rsp)
    movaps %xmm10,nb132_fixH1(%rsp)
    movaps %xmm11,nb132_fiyH1(%rsp)
    movaps %xmm12,nb132_fizH1(%rsp)
    movaps %xmm13,nb132_fixH2(%rsp)
    movaps %xmm14,nb132_fiyH2(%rsp)
    movaps %xmm15,nb132_fizH2(%rsp)

    ## xmm0 = fOx
    ## xmm1 = fOy
    ## xmm2 = fOz
    movaps %xmm3,%xmm5
    unpcklps %xmm4,%xmm3
    unpckhps %xmm4,%xmm5

    addps %xmm3,%xmm0
    addps %xmm5,%xmm1

    movhlps  %xmm2,%xmm3 ## fOzc fOzd

    movlps %xmm0,(%rdi,%rax,4)
    movhps %xmm0,(%rdi,%rbx,4)
    movlps %xmm1,(%rdi,%rcx,4)
    movhps %xmm1,(%rdi,%rdx,4)
    movss  %xmm2,8(%rdi,%rax,4)
    movss  %xmm3,8(%rdi,%rcx,4)
    shufps $1,%xmm2,%xmm2
    shufps $1,%xmm3,%xmm3
    movss  %xmm2,8(%rdi,%rbx,4)
    movss  %xmm3,8(%rdi,%rdx,4)


        ## move j H1 coordinates to local temp variables 
        movq nb132_pos(%rbp),%rsi        ## base of pos[] 
    movlps 12(%rsi,%rax,4),%xmm0    ## jxH1a jyH1a  -   -
    movlps 12(%rsi,%rcx,4),%xmm1    ## jxH1c jyH1c  -   -
    movhps 12(%rsi,%rbx,4),%xmm0    ## jxH1a jyH1a jxH1b jyH1b 
    movhps 12(%rsi,%rdx,4),%xmm1    ## jxH1c jyH1c jxH1d jyH1d 

    movss  20(%rsi,%rax,4),%xmm2    ## jzH1a  -  -  -
    movss  20(%rsi,%rcx,4),%xmm3    ## jzH1c  -  -  -
    movhps 20(%rsi,%rbx,4),%xmm2    ## jzH1a  -  jzH1b  -
    movhps 20(%rsi,%rdx,4),%xmm3    ## jzH1c  -  jzH1d -

    movaps %xmm0,%xmm4
    unpcklps %xmm1,%xmm0 ## jxH1a jxH1c jyH1a jyH1c        
    unpckhps %xmm1,%xmm4 ## jxH1b jxH1d jyH1b jyH1d
    movaps %xmm0,%xmm1
    unpcklps %xmm4,%xmm0 ## x
    unpckhps %xmm4,%xmm1 ## y

    shufps  $136,%xmm3,%xmm2  ## 10001000 => jzH1a jzH1b jzH1c jzH1d

    ## xmm0 = H1x
    ## xmm1 = H1y
    ## xmm2 = H1z

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

    subps nb132_ixO(%rsp),%xmm0
    subps nb132_iyO(%rsp),%xmm1
    subps nb132_izO(%rsp),%xmm2
    subps nb132_ixH1(%rsp),%xmm3
    subps nb132_iyH1(%rsp),%xmm4
    subps nb132_izH1(%rsp),%xmm5
    subps nb132_ixH2(%rsp),%xmm6
    subps nb132_iyH2(%rsp),%xmm7
    subps nb132_izH2(%rsp),%xmm8

        movaps %xmm0,nb132_dxOH1(%rsp)
        movaps %xmm1,nb132_dyOH1(%rsp)
        movaps %xmm2,nb132_dzOH1(%rsp)
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        movaps %xmm3,nb132_dxH1H1(%rsp)
        movaps %xmm4,nb132_dyH1H1(%rsp)
        movaps %xmm5,nb132_dzH1H1(%rsp)
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        movaps %xmm6,nb132_dxH2H1(%rsp)
        movaps %xmm7,nb132_dyH2H1(%rsp)
        movaps %xmm8,nb132_dzH2H1(%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 jH1 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  nb132_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  nb132_half(%rsp),%xmm0
        mulps   %xmm0,%xmm9 ## rinvOH1
        mulps   %xmm0,%xmm10 ## rinvH1H1
    mulps   %xmm0,%xmm11 ## rinvH2H1

        ## H1 interactions 
    movaps %xmm9,%xmm0  ## rinv
    movaps %xmm10,%xmm1
    movaps %xmm11,%xmm2
    mulps  %xmm9,%xmm9   ## rinvsq
    mulps  %xmm10,%xmm10
    mulps  %xmm11,%xmm11
    mulps  nb132_qqOH(%rsp),%xmm0      ## vcoul
    mulps  nb132_qqHH(%rsp),%xmm1
    mulps  nb132_qqHH(%rsp),%xmm2
    mulps  %xmm0,%xmm9
    mulps  %xmm1,%xmm10
    mulps  %xmm2,%xmm11

    addps nb132_vctot(%rsp),%xmm0
    addps %xmm2,%xmm1
    addps %xmm1,%xmm0
    movaps %xmm0,nb132_vctot(%rsp)

        ## move j H1 forces to local temp variables 
        movq nb132_faction(%rbp),%rdi        ## base of faction[] 
    movlps 12(%rdi,%rax,4),%xmm0    ## jxH1a jyH1a  -   -
    movlps 12(%rdi,%rcx,4),%xmm1    ## jxH1c jyH1c  -   -
    movhps 12(%rdi,%rbx,4),%xmm0    ## jxH1a jyH1a jxH1b jyH1b 
    movhps 12(%rdi,%rdx,4),%xmm1    ## jxH1c jyH1c jxH1d jyH1d 

    movss  20(%rdi,%rax,4),%xmm2    ## jzH1a  -  -  -
    movss  20(%rdi,%rcx,4),%xmm3    ## jzH1c  -  -  -
    movhps 20(%rdi,%rbx,4),%xmm2    ## jzH1a  -  jzH1b  -
    movhps 20(%rdi,%rdx,4),%xmm3    ## jzH1c  -  jzH1d -

    shufps $136,%xmm3,%xmm2 ## 10001000 => jzH1a jzH1b jzH1c jzH1d

    ## xmm0: jxH1a jyH1a jxH1b jyH1b 
    ## xmm1: jxH1c jyH1c jxH1d jyH1d
    ## xmm2: jzH1a jzH1b jzH1c jzH1d

    movaps %xmm9,%xmm7
    movaps %xmm9,%xmm8