nb_kernel204_x86_64_sse2.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,256 行 · 第 1/5 页

    cvtps2pd %xmm4,%xmm4
    cvtps2pd %xmm7,%xmm7

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

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

        movapd  nb204_three(%rsp),%xmm9
        movapd  %xmm9,%xmm10
    movapd  %xmm9,%xmm11

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

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

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

        movapd  nb204_half(%rsp),%xmm15
        mulpd   %xmm15,%xmm9 ## first iteration for rinvH1M 
        mulpd   %xmm15,%xmm10 ## first iteration for rinvH2M
    mulpd   %xmm15,%xmm11 ## first iteration for rinvMM

    ## second iteration step    
        movapd  %xmm9,%xmm2
        movapd  %xmm10,%xmm5
    movapd  %xmm11,%xmm8

        mulpd   %xmm2,%xmm2 ## lu*lu
        mulpd   %xmm5,%xmm5 ## lu*lu
    mulpd   %xmm8,%xmm8 ## lu*lu

        movapd  nb204_three(%rsp),%xmm1
        movapd  %xmm1,%xmm4
    movapd  %xmm1,%xmm7

        mulpd   %xmm0,%xmm2 ## rsq*lu*lu
        mulpd   %xmm3,%xmm5 ## rsq*lu*lu 
    mulpd   %xmm6,%xmm8 ## rsq*lu*lu

        subpd   %xmm2,%xmm1
        subpd   %xmm5,%xmm4
    subpd   %xmm8,%xmm7 ## 3-rsq*lu*lu

        mulpd   %xmm1,%xmm9
        mulpd   %xmm4,%xmm10
    mulpd   %xmm7,%xmm11 ## lu*(3-rsq*lu*lu)

        movapd  nb204_half(%rsp),%xmm15
        mulpd   %xmm15,%xmm9 ##  rinvH1M
        mulpd   %xmm15,%xmm10 ##   rinvH2M
    mulpd   %xmm15,%xmm11 ##   rinvMM


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

    movapd %xmm9,%xmm1 ## copy of rinv
    movapd %xmm10,%xmm4
    movapd %xmm11,%xmm7
    movapd nb204_krf(%rsp),%xmm2
    mulpd  %xmm9,%xmm9  ## rinvsq
    mulpd  %xmm10,%xmm10
    mulpd  %xmm11,%xmm11
    mulpd  %xmm2,%xmm0 ## k*rsq
    mulpd  %xmm2,%xmm3
    mulpd  %xmm2,%xmm6
    movapd %xmm0,%xmm2 ## copy of k*rsq
    movapd %xmm3,%xmm5
    movapd %xmm6,%xmm8
    addpd  %xmm1,%xmm2 ## rinv+krsq
    addpd  %xmm4,%xmm5
    addpd  %xmm7,%xmm8
    movapd nb204_crf(%rsp),%xmm14
    subpd  %xmm14,%xmm2  ## rinv+krsq-crf
    subpd  %xmm14,%xmm5
    subpd  %xmm14,%xmm8
    movapd nb204_qqMH(%rsp),%xmm12
    movapd nb204_qqMM(%rsp),%xmm13
    mulpd  %xmm12,%xmm2 ## xmm6=voul=qq*(rinv+ krsq-crf)
    mulpd  %xmm12,%xmm5 ## xmm6=voul=qq*(rinv+ krsq-crf)
    mulpd  %xmm13,%xmm8 ## xmm6=voul=qq*(rinv+ krsq-crf)
    addpd  %xmm0,%xmm0 ## 2*krsq
    addpd  %xmm3,%xmm3
    addpd  %xmm6,%xmm6
    subpd  %xmm0,%xmm1 ## rinv-2*krsq
    subpd  %xmm3,%xmm4
    subpd  %xmm6,%xmm7
    mulpd  %xmm12,%xmm1  ## (rinv-2*krsq)*qq
    mulpd  %xmm12,%xmm4
    mulpd  %xmm13,%xmm7
    addpd  nb204_vctot(%rsp),%xmm2
    addpd  %xmm8,%xmm5
    addpd  %xmm5,%xmm2
    movapd  %xmm2,nb204_vctot(%rsp)

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

    ## move j M forces to xmm0-xmm2
        movlpd 72(%rdi,%rax,8),%xmm0
        movlpd 80(%rdi,%rax,8),%xmm1
        movlpd 88(%rdi,%rax,8),%xmm2
        movhpd 72(%rdi,%rbx,8),%xmm0
        movhpd 80(%rdi,%rbx,8),%xmm1
        movhpd 88(%rdi,%rbx,8),%xmm2

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

        mulpd nb204_dxH1M(%rsp),%xmm7
        mulpd nb204_dyH1M(%rsp),%xmm8
        mulpd nb204_dzH1M(%rsp),%xmm9
        mulpd nb204_dxH2M(%rsp),%xmm10
        mulpd nb204_dyH2M(%rsp),%xmm11
        mulpd nb204_dzH2M(%rsp),%xmm12
        mulpd nb204_dxMM(%rsp),%xmm13
        mulpd nb204_dyMM(%rsp),%xmm14
        mulpd nb204_dzMM(%rsp),%xmm15

    addpd %xmm7,%xmm0
    addpd %xmm8,%xmm1
    addpd %xmm9,%xmm2
    addpd nb204_fixH1(%rsp),%xmm7
    addpd nb204_fiyH1(%rsp),%xmm8
    addpd nb204_fizH1(%rsp),%xmm9

    addpd %xmm10,%xmm0
    addpd %xmm11,%xmm1
    addpd %xmm12,%xmm2
    addpd nb204_fixH2(%rsp),%xmm10
    addpd nb204_fiyH2(%rsp),%xmm11
    addpd nb204_fizH2(%rsp),%xmm12

    addpd %xmm13,%xmm0
    addpd %xmm14,%xmm1
    addpd %xmm15,%xmm2
    addpd nb204_fixM(%rsp),%xmm13
    addpd nb204_fiyM(%rsp),%xmm14
    addpd nb204_fizM(%rsp),%xmm15

    movapd %xmm7,nb204_fixH1(%rsp)
    movapd %xmm8,nb204_fiyH1(%rsp)
    movapd %xmm9,nb204_fizH1(%rsp)
    movapd %xmm10,nb204_fixH2(%rsp)
    movapd %xmm11,nb204_fiyH2(%rsp)
    movapd %xmm12,nb204_fizH2(%rsp)
    movapd %xmm13,nb204_fixM(%rsp)
    movapd %xmm14,nb204_fiyM(%rsp)
    movapd %xmm15,nb204_fizM(%rsp)

    ## store back j M forces from xmm0-xmm2
        movlpd %xmm0,72(%rdi,%rax,8)
        movlpd %xmm1,80(%rdi,%rax,8)
        movlpd %xmm2,88(%rdi,%rax,8)
        movhpd %xmm0,72(%rdi,%rbx,8)
        movhpd %xmm1,80(%rdi,%rbx,8)
        movhpd %xmm2,88(%rdi,%rbx,8)

        ## should we do one more iteration? 
        subl $2,nb204_innerk(%rsp)
        jl    _nb_kernel204_x86_64_sse2.nb204_checksingle
        jmp   _nb_kernel204_x86_64_sse2.nb204_unroll_loop
_nb_kernel204_x86_64_sse2.nb204_checksingle: 
        movl  nb204_innerk(%rsp),%edx
        andl  $1,%edx
        jnz   _nb_kernel204_x86_64_sse2.nb204_dosingle
        jmp   _nb_kernel204_x86_64_sse2.nb204_updateouterdata
_nb_kernel204_x86_64_sse2.nb204_dosingle: 
        movq  nb204_innerjjnr(%rsp),%rdx       ## pointer to jjnr[k] 
        movl  (%rdx),%eax

        movq nb204_pos(%rbp),%rsi
        lea  (%rax,%rax,2),%rax


        ## move j H1 coordinates to local temp variables 
    movsd 24(%rsi,%rax,8),%xmm0
    movsd 32(%rsi,%rax,8),%xmm1
    movsd 40(%rsi,%rax,8),%xmm2

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

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

    subsd nb204_ixH1(%rsp),%xmm0
    subsd nb204_iyH1(%rsp),%xmm1
    subsd nb204_izH1(%rsp),%xmm2
    subsd nb204_ixH2(%rsp),%xmm3
    subsd nb204_iyH2(%rsp),%xmm4
    subsd nb204_izH2(%rsp),%xmm5
    subsd nb204_ixM(%rsp),%xmm6
    subsd nb204_iyM(%rsp),%xmm7
    subsd nb204_izM(%rsp),%xmm8

        movsd %xmm0,nb204_dxH1H1(%rsp)
        movsd %xmm1,nb204_dyH1H1(%rsp)
        movsd %xmm2,nb204_dzH1H1(%rsp)
        mulsd  %xmm0,%xmm0
        mulsd  %xmm1,%xmm1
        mulsd  %xmm2,%xmm2
        movsd %xmm3,nb204_dxH2H1(%rsp)
        movsd %xmm4,nb204_dyH2H1(%rsp)
        movsd %xmm5,nb204_dzH2H1(%rsp)
        mulsd  %xmm3,%xmm3
        mulsd  %xmm4,%xmm4
        mulsd  %xmm5,%xmm5
        movsd %xmm6,nb204_dxMH1(%rsp)
        movsd %xmm7,nb204_dyMH1(%rsp)
        movsd %xmm8,nb204_dzMH1(%rsp)
        mulsd  %xmm6,%xmm6
        mulsd  %xmm7,%xmm7
        mulsd  %xmm8,%xmm8
        addsd  %xmm1,%xmm0
        addsd  %xmm2,%xmm0
        addsd  %xmm4,%xmm3
        addsd  %xmm5,%xmm3
    addsd  %xmm7,%xmm6
    addsd  %xmm8,%xmm6

        ## start doing invsqrt for jH1 atoms
    cvtsd2ss %xmm0,%xmm1
    cvtsd2ss %xmm3,%xmm4
    cvtsd2ss %xmm6,%xmm7
        rsqrtss %xmm1,%xmm1
        rsqrtss %xmm4,%xmm4
    rsqrtss %xmm7,%xmm7
    cvtss2sd %xmm1,%xmm1
    cvtss2sd %xmm4,%xmm4
    cvtss2sd %xmm7,%xmm7

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

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

        movsd  nb204_three(%rsp),%xmm9
        movsd  %xmm9,%xmm10
    movsd  %xmm9,%xmm11

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

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

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

        movsd  nb204_half(%rsp),%xmm15
        mulsd   %xmm15,%xmm9 ## first iteration for rinvH1H1 
        mulsd   %xmm15,%xmm10 ## first iteration for rinvH2H1
    mulsd   %xmm15,%xmm11 ## first iteration for rinvMH1 

    ## second iteration step    
        movsd  %xmm9,%xmm2
        movsd  %xmm10,%xmm5
    movsd  %xmm11,%xmm8

        mulsd   %xmm2,%xmm2 ## lu*lu
        mulsd   %xmm5,%xmm5 ## lu*lu
    mulsd   %xmm8,%xmm8 ## lu*lu

        movsd  nb204_three(%rsp),%xmm1
        movsd  %xmm1,%xmm4
    movsd  %xmm1,%xmm7

        mulsd   %xmm0,%xmm2 ## rsq*lu*lu
        mulsd   %xmm3,%xmm5 ## rsq*lu*lu 
    mulsd   %xmm6,%xmm8 ## rsq*lu*lu

        subsd   %xmm2,%xmm1
        subsd   %xmm5,%xmm4
    subsd   %xmm8,%xmm7 ## 3-rsq*lu*lu

        mulsd   %xmm1,%xmm9
        mulsd   %xmm4,%xmm10
    mulsd   %xmm7,%xmm11 ## lu*(3-rsq*lu*lu)

        movsd  nb204_half(%rsp),%xmm15
        mulsd   %xmm15,%xmm9 ##  rinvH1H1 
        mulsd   %xmm15,%xmm10 ##   rinvH2H1
    mulsd   %xmm15,%xmm11 ##   rinvMH1

        ## H1 interactions 
    ## rsq in xmm0,xmm3,xmm6  
    ## rinv in xmm9, xmm10, xmm11

    movsd %xmm9,%xmm1 ## copy of rinv
    movsd %xmm10,%xmm4
    movsd %xmm11,%xmm7
    movsd nb204_krf(%rsp),%xmm2
    mulsd  %xmm9,%xmm9  ## rinvsq
    mulsd  %xmm10,%xmm10
    mulsd  %xmm11,%xmm11
    mulsd  %xmm2,%xmm0 ## k*rsq
    mulsd  %xmm2,%xmm3
    mulsd  %xmm2,%xmm6
    movsd  %xmm0,%xmm2 ## copy of k*rsq
    movsd  %xmm3,%xmm5
    movsd  %xmm6,%xmm8
    addsd  %xmm1,%xmm2 ## rinv+krsq
    addsd  %xmm4,%xmm5
    addsd  %xmm7,%xmm8
    movsd  nb204_crf(%rsp),%xmm14
    subsd  %xmm14,%xmm2  ## rinv+krsq-crf
    subsd  %xmm14,%xmm5
    subsd  %xmm14,%xmm8
    movsd  nb204_qqHH(%rsp),%xmm12
    movsd  nb204_qqMH(%rsp),%xmm13
    mulsd  %xmm12,%xmm2 ## voul=qq*(rinv+ krsq-crf)
    mulsd  %xmm12,%xmm5 ## voul=qq*(rinv+ krsq-crf)
    mulsd  %xmm13,%xmm8 ## voul=qq*(rinv+ krsq-crf)
    addsd  %xmm0,%xmm0 ## 2*krsq
    addsd  %xmm3,%xmm3
    addsd  %xmm6,%xmm6
    subsd  %xmm0,%xmm1 ## rinv-2*krsq
    subsd  %xmm3,%xmm4
    subsd  %xmm6,%xmm7
    mulsd  %xmm12,%xmm1  ## (rinv-2*krsq)*qq
    mulsd  %xmm12,%xmm4
    mulsd  %xmm13,%xmm7
    addsd  nb204_vctot(%rsp),%xmm2
    addsd  %xmm8,%xmm5
    addsd  %xmm5,%xmm2
    movsd  %xmm2,nb204_vctot(%rsp)

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

    ## move j H1 forces to xmm0-xmm2
        movsd 24(%rdi,%rax,8),%xmm0
        movsd 32(%rdi,%rax,8),%xmm1
        movsd 40(%rdi,%rax,8),%xmm2

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

        mulsd nb204_dxH1H1(%rsp),%xmm7
        mulsd nb204_dyH1H1(%rsp),%xmm8
        mulsd nb204_dzH1H1(%rsp),%xmm9
        mulsd nb204_dxH2H1(%rsp),%xmm10
        mulsd nb204_dyH2H1(%rsp),%xmm11
        mulsd nb204_dzH2H1(%rsp),%xmm12
        mulsd nb204_dxMH1(%rsp),%xmm13
        mulsd nb204_dyMH1(%rsp),%xmm14
        mulsd nb204_dzMH1(%rsp),%xmm15

    addsd %xmm7,%xmm0
    addsd %xmm8,%xmm1
    addsd %xmm9,%xmm2
    addsd nb204_fixH1(%rsp),%xmm7
    addsd nb204_fiyH1(%rsp),%xmm8
    addsd nb204_fizH1(%rsp),%xmm9

    addsd %xmm10,%xmm0
    addsd %xmm11,%xmm1
    addsd %xmm12,%xmm2
    addsd nb204_fixH2(%rsp),%xmm10
    addsd nb204_fiyH2(%rsp),%xmm11
    addsd nb204_fizH2(%rsp),%xmm12

    addsd %xmm13,%xmm0
    addsd %xmm14,%xmm1
    addsd %xmm15,%xmm2
    addsd nb204_fixM(%rsp),%xmm13
    addsd nb204_fiyM(%rsp),%xmm14
    addsd nb204_fizM(%rsp),%xmm15

    movsd %xmm7,nb204_fixH1(%rsp)
    movsd %xmm8,nb204_fiyH1(%rsp)
    movsd %xmm9,nb204_fizH1(%rsp)
    movsd %xmm10,nb204_fixH2(%rsp)
    movsd %xmm11,nb204_fiyH2(%rsp)
    movsd %xmm12,nb204_fizH2(%rsp)
    movsd %xmm13,nb204_fixM(%rsp)
    movsd %xmm14,nb204_fiyM(%rsp)
    movsd %xmm15,nb204_fizM(%rsp)

    ## store back j H1 forces from xmm0-xmm2
        movsd %xmm0,24(%rdi,%rax,8)
        movsd %xmm1,32(%rdi,%rax,8)
        movsd %xmm2,40(%rdi,%rax,8)

        ## move j H2 coordinates to local temp variables 
    movsd 48(%rsi,%rax,8),%xmm0
    movsd 56(%rsi,%rax,8),%xmm1
    movsd 64(%rsi,%rax,8),%xmm2

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

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

    subsd nb204_ixH1(%rsp),%xmm0
    subsd nb204_iyH1(%rsp),%xmm1
    subsd nb204_izH1(%rsp),%xmm2
    subsd nb204_ixH2(%rsp),%xmm3
    subsd nb204_iyH2(%rsp),%xmm4
    subsd nb204_izH2(%rsp),%xmm5
    subsd nb204_ixM(%rsp),%xmm6
    subsd nb204_iyM(%rsp),%xmm7
    subsd nb204_izM(%rsp),%xmm8

        movsd %xmm0,nb204_dxH1H2(%rsp)
        movsd %xmm1,nb204_dyH1H2(%rsp)
        movsd %xmm2,nb204_dzH1H2(%rsp)
        mulsd  %xmm0,%xmm0
        mulsd  %xmm1,%xmm1
        mulsd  %xmm2,%xmm2
        movsd %xmm3,nb204_dxH2H2(%rsp)