nb_kernel212_x86_64_sse2.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,263 行 · 第 1/5 页

    subpd nb212_izH1(%rsp),%xmm5
    subpd nb212_ixH2(%rsp),%xmm6
    subpd nb212_iyH2(%rsp),%xmm7
    subpd nb212_izH2(%rsp),%xmm8

        movapd %xmm0,nb212_dxOH2(%rsp)
        movapd %xmm1,nb212_dyOH2(%rsp)
        movapd %xmm2,nb212_dzOH2(%rsp)
        mulpd  %xmm0,%xmm0
        mulpd  %xmm1,%xmm1
        mulpd  %xmm2,%xmm2
        movapd %xmm3,nb212_dxH1H2(%rsp)
        movapd %xmm4,nb212_dyH1H2(%rsp)
        movapd %xmm5,nb212_dzH1H2(%rsp)
        mulpd  %xmm3,%xmm3
        mulpd  %xmm4,%xmm4
        mulpd  %xmm5,%xmm5
        movapd %xmm6,nb212_dxH2H2(%rsp)
        movapd %xmm7,nb212_dyH2H2(%rsp)
        movapd %xmm8,nb212_dzH2H2(%rsp)
        mulpd  %xmm6,%xmm6
        mulpd  %xmm7,%xmm7
        mulpd  %xmm8,%xmm8
        addpd  %xmm1,%xmm0
        addpd  %xmm2,%xmm0
        addpd  %xmm4,%xmm3
        addpd  %xmm5,%xmm3
    addpd  %xmm7,%xmm6
    addpd  %xmm8,%xmm6

        ## start doing invsqrt for jH2 atoms
    cvtpd2ps %xmm0,%xmm1
    cvtpd2ps %xmm3,%xmm4
    cvtpd2ps %xmm6,%xmm7
        rsqrtps %xmm1,%xmm1
        rsqrtps %xmm4,%xmm4
    rsqrtps %xmm7,%xmm7
    cvtps2pd %xmm1,%xmm1
    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  nb212_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  nb212_half(%rsp),%xmm15
        mulpd   %xmm15,%xmm9 ## first iteration for rinvOH2 
        mulpd   %xmm15,%xmm10 ## first iteration for rinvH1H2
    mulpd   %xmm15,%xmm11 ## first iteration for rinvH2H2

    ## 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  nb212_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  nb212_half(%rsp),%xmm15
        mulpd   %xmm15,%xmm9 ##  rinvOH2
        mulpd   %xmm15,%xmm10 ##   rinvH1H2
    mulpd   %xmm15,%xmm11 ##   rinvH2H2


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

    movapd %xmm9,%xmm1 ## copy of rinv
    movapd %xmm10,%xmm4
    movapd %xmm11,%xmm7
    movapd nb212_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 nb212_crf(%rsp),%xmm14
    subpd  %xmm14,%xmm2  ## rinv+krsq-crf
    subpd  %xmm14,%xmm5
    subpd  %xmm14,%xmm8
    movapd nb212_qqOH(%rsp),%xmm12
    movapd nb212_qqHH(%rsp),%xmm13
    mulpd  %xmm12,%xmm2 ## xmm6=voul=qq*(rinv+ krsq-crf)
    mulpd  %xmm13,%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  %xmm13,%xmm4
    mulpd  %xmm13,%xmm7
    addpd  nb212_vctot(%rsp),%xmm2
    addpd  %xmm8,%xmm5
    addpd  %xmm5,%xmm2
    movapd %xmm2,nb212_vctot(%rsp)

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

    ## move j H2 forces to xmm0-xmm2
        movlpd 48(%rdi,%rax,8),%xmm0
        movlpd 56(%rdi,%rax,8),%xmm1
        movlpd 64(%rdi,%rax,8),%xmm2
        movhpd 48(%rdi,%rbx,8),%xmm0
        movhpd 56(%rdi,%rbx,8),%xmm1
        movhpd 64(%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 nb212_dxOH2(%rsp),%xmm7
        mulpd nb212_dyOH2(%rsp),%xmm8
        mulpd nb212_dzOH2(%rsp),%xmm9
        mulpd nb212_dxH1H2(%rsp),%xmm10
        mulpd nb212_dyH1H2(%rsp),%xmm11
        mulpd nb212_dzH1H2(%rsp),%xmm12
        mulpd nb212_dxH2H2(%rsp),%xmm13
        mulpd nb212_dyH2H2(%rsp),%xmm14
        mulpd nb212_dzH2H2(%rsp),%xmm15

    addpd %xmm7,%xmm0
    addpd %xmm8,%xmm1
    addpd %xmm9,%xmm2
    addpd nb212_fixO(%rsp),%xmm7
    addpd nb212_fiyO(%rsp),%xmm8
    addpd nb212_fizO(%rsp),%xmm9

    addpd %xmm10,%xmm0
    addpd %xmm11,%xmm1
    addpd %xmm12,%xmm2
    addpd nb212_fixH1(%rsp),%xmm10
    addpd nb212_fiyH1(%rsp),%xmm11
    addpd nb212_fizH1(%rsp),%xmm12

    addpd %xmm13,%xmm0
    addpd %xmm14,%xmm1
    addpd %xmm15,%xmm2
    addpd nb212_fixH2(%rsp),%xmm13
    addpd nb212_fiyH2(%rsp),%xmm14
    addpd nb212_fizH2(%rsp),%xmm15

    movapd %xmm7,nb212_fixO(%rsp)
    movapd %xmm8,nb212_fiyO(%rsp)
    movapd %xmm9,nb212_fizO(%rsp)
    movapd %xmm10,nb212_fixH1(%rsp)
    movapd %xmm11,nb212_fiyH1(%rsp)
    movapd %xmm12,nb212_fizH1(%rsp)
    movapd %xmm13,nb212_fixH2(%rsp)
    movapd %xmm14,nb212_fiyH2(%rsp)
    movapd %xmm15,nb212_fizH2(%rsp)

    ## store back j H2 forces from xmm0-xmm2
        movlpd %xmm0,48(%rdi,%rax,8)
        movlpd %xmm1,56(%rdi,%rax,8)
        movlpd %xmm2,64(%rdi,%rax,8)
        movhpd %xmm0,48(%rdi,%rbx,8)
        movhpd %xmm1,56(%rdi,%rbx,8)
        movhpd %xmm2,64(%rdi,%rbx,8)

        ## should we do one more iteration? 
        subl $2,nb212_innerk(%rsp)
        jl    _nb_kernel212_x86_64_sse2.nb212_checksingle
        jmp   _nb_kernel212_x86_64_sse2.nb212_unroll_loop
_nb_kernel212_x86_64_sse2.nb212_checksingle: 
        movl  nb212_innerk(%rsp),%edx
        andl  $1,%edx
        jnz   _nb_kernel212_x86_64_sse2.nb212_dosingle
        jmp   _nb_kernel212_x86_64_sse2.nb212_updateouterdata
_nb_kernel212_x86_64_sse2.nb212_dosingle: 
        movq  nb212_innerjjnr(%rsp),%rdx       ## pointer to jjnr[k] 
        movl  (%rdx),%eax

        movq nb212_pos(%rbp),%rsi
        lea  (%rax,%rax,2),%rax


        ## move j O coordinates to local temp variables 
    movsd (%rsi,%rax,8),%xmm0
    movsd 8(%rsi,%rax,8),%xmm1
    movsd 16(%rsi,%rax,8),%xmm2

    ## xmm0 = Ox
    ## xmm1 = Oy
    ## xmm2 = Oz

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

    subsd nb212_ixO(%rsp),%xmm0
    subsd nb212_iyO(%rsp),%xmm1
    subsd nb212_izO(%rsp),%xmm2
    subsd nb212_ixH1(%rsp),%xmm3
    subsd nb212_iyH1(%rsp),%xmm4
    subsd nb212_izH1(%rsp),%xmm5
    subsd nb212_ixH2(%rsp),%xmm6
    subsd nb212_iyH2(%rsp),%xmm7
    subsd nb212_izH2(%rsp),%xmm8

        movsd %xmm0,nb212_dxOO(%rsp)
        movsd %xmm1,nb212_dyOO(%rsp)
        movsd %xmm2,nb212_dzOO(%rsp)
        mulsd  %xmm0,%xmm0
        mulsd  %xmm1,%xmm1
        mulsd  %xmm2,%xmm2
        movsd %xmm3,nb212_dxH1O(%rsp)
        movsd %xmm4,nb212_dyH1O(%rsp)
        movsd %xmm5,nb212_dzH1O(%rsp)
        mulsd  %xmm3,%xmm3
        mulsd  %xmm4,%xmm4
        mulsd  %xmm5,%xmm5
        movsd %xmm6,nb212_dxH2O(%rsp)
        movsd %xmm7,nb212_dyH2O(%rsp)
        movsd %xmm8,nb212_dzH2O(%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 jO 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  nb212_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  nb212_half(%rsp),%xmm15
        mulsd   %xmm15,%xmm9 ## first iteration for rinvOO 
        mulsd   %xmm15,%xmm10 ## first iteration for rinvH1O
    mulsd   %xmm15,%xmm11 ## first iteration for rinvH2O 

    ## 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  nb212_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  nb212_half(%rsp),%xmm15
        mulsd   %xmm15,%xmm9 ##  rinvOO 
        mulsd   %xmm15,%xmm10 ##   rinvH1O
    mulsd   %xmm15,%xmm11 ##   rinvH2O

        ## O interactions 
    ## rsq in xmm0,xmm3,xmm6  
    ## rinv in xmm9, xmm10, xmm11
    movsd %xmm9,%xmm1 ## copy of rinv
    movsd %xmm10,%xmm4
    movsd %xmm11,%xmm7
    movsd nb212_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 %xmm9,%xmm12
    mulsd  %xmm12,%xmm12 ## rinv4
    mulsd  %xmm9,%xmm12 ## rinv6
    subsd  nb212_crf(%rsp),%xmm2     ## rinv+krsq-crf
    subsd  nb212_crf(%rsp),%xmm5
    subsd  nb212_crf(%rsp),%xmm8
    mulsd  nb212_qqOO(%rsp),%xmm2   ## voul=qq*(rinv+ krsq-crf)
    mulsd  nb212_qqOH(%rsp),%xmm5   ## voul=qq*(rinv+ krsq-crf)
    mulsd  nb212_qqOH(%rsp),%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
    movsd %xmm12,%xmm13 ## rinv6
    mulsd %xmm12,%xmm12 ## rinv12
        mulsd  nb212_c6(%rsp),%xmm13
        mulsd  nb212_c12(%rsp),%xmm12
    movsd %xmm12,%xmm14
    subsd  %xmm13,%xmm14
    mulsd  nb212_qqOO(%rsp),%xmm1     ## (rinv-2*krsq)*qq
    mulsd  nb212_qqOH(%rsp),%xmm4
    mulsd  nb212_qqOH(%rsp),%xmm7
    addsd  nb212_vctot(%rsp),%xmm2
    addsd  %xmm8,%xmm5
    addsd  %xmm5,%xmm2
    movsd  %xmm2,nb212_vctot(%rsp)

        addsd  nb212_Vvdwtot(%rsp),%xmm14
        mulsd  nb212_six(%rsp),%xmm13
        mulsd  nb212_twelve(%rsp),%xmm12
        movsd %xmm14,nb212_Vvdwtot(%rsp)
    subsd  %xmm13,%xmm12 ## LJ fscal        

    addsd %xmm12,%xmm1

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

    ## move j O forces to xmm0-xmm2
        movsd (%rdi,%rax,8),%xmm0
        movsd 8(%rdi,%rax,8),%xmm1
        movsd 16(%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 nb212_dxOO(%rsp),%xmm7
        mulsd nb212_dyOO(%rsp),%xmm8
        mulsd nb212_dzOO(%rsp),%xmm9
        mulsd nb212_dxH1O(%rsp),%xmm10
        mulsd nb212_dyH1O(%rsp),%xmm11
        mulsd nb212_dzH1O(%rsp),%xmm12
        mulsd nb212_dxH2O(%rsp),%xmm13
        mulsd nb212_dyH2O(%rsp),%xmm14
        mulsd nb212_dzH2O(%rsp),%xmm15

    addsd %xmm7,%xmm0
    addsd %xmm8,%xmm1
    addsd %xmm9,%xmm2
    addsd nb212_fixO(%rsp),%xmm7
    addsd nb212_fiyO(%rsp),%xmm8
    addsd nb212_fizO(%rsp),%xmm9

    addsd %xmm10,%xmm0
    addsd %xmm11,%xmm1
    addsd %xmm12,%xmm2
    addsd nb212_fixH1(%rsp),%xmm10
    addsd nb212_fiyH1(%rsp),%xmm11
    addsd nb212_fizH1(%rsp),%xmm12

    addsd %xmm13,%xmm0
    addsd %xmm14,%xmm1
    addsd %xmm15,%xmm2
    addsd nb212_fixH2(%rsp),%xmm13
    addsd nb212_fiyH2(%rsp),%xmm14