nb_kernel112_x86_64_sse.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,103 行 · 第 1/5 页

        lock 
        cmpxchgl %ebx,(%rsi)                    ## write nn1 to *counter,
                                                ## if it hasnt changed.
                                                ## or reread *counter to eax.
        pause                                   ## -> better p4 performance
        jnz _nb_kernel112nf_x86_64_sse.nb112nf_spinlock

        ## if(nn1>nri) nn1=nri
        movl nb112nf_nri(%rsp),%ecx
        movl %ecx,%edx
        subl %ebx,%ecx
        cmovlel %edx,%ebx                       ## if(nn1>nri) nn1=nri
        ## Cleared the spinlock if we got here.
        ## eax contains nn0, ebx contains nn1.
        movl %eax,nb112nf_n(%rsp)
        movl %ebx,nb112nf_nn1(%rsp)
        subl %eax,%ebx                          ## calc number of outer lists
        movl %eax,%esi                          ## copy n to esi
        jg  _nb_kernel112nf_x86_64_sse.nb112nf_outerstart
        jmp _nb_kernel112nf_x86_64_sse.nb112nf_end

_nb_kernel112nf_x86_64_sse.nb112nf_outerstart: 
        ## ebx contains number of outer iterations
        addl nb112nf_nouter(%rsp),%ebx
        movl %ebx,nb112nf_nouter(%rsp)

_nb_kernel112nf_x86_64_sse.nb112nf_outer: 
        movq  nb112nf_shift(%rsp),%rax        ## rax = pointer into shift[] 
        movl  (%rax,%rsi,4),%ebx                ## rbx=shift[n] 

        lea  (%rbx,%rbx,2),%rbx    ## rbx=3*is 
        movl  %ebx,nb112nf_is3(%rsp)            ## store is3 

        movq  nb112nf_shiftvec(%rsp),%rax     ## rax = base of shiftvec[] 

        movss (%rax,%rbx,4),%xmm0
        movss 4(%rax,%rbx,4),%xmm1
        movss 8(%rax,%rbx,4),%xmm2

        movq  nb112nf_iinr(%rsp),%rcx         ## rcx = pointer into iinr[]      
        movl  (%rcx,%rsi,4),%ebx            ## ebx =ii 

        lea  (%rbx,%rbx,2),%rbx        ## rbx = 3*ii=ii3 
        movq  nb112nf_pos(%rbp),%rax      ## rax = base of pos[]  
        movl  %ebx,nb112nf_ii3(%rsp)

        movaps %xmm0,%xmm3
        movaps %xmm1,%xmm4
        movaps %xmm2,%xmm5
        addss (%rax,%rbx,4),%xmm3
        addss 4(%rax,%rbx,4),%xmm4
        addss 8(%rax,%rbx,4),%xmm5
        shufps $0,%xmm3,%xmm3
        shufps $0,%xmm4,%xmm4
        shufps $0,%xmm5,%xmm5
        movaps %xmm3,nb112nf_ixO(%rsp)
        movaps %xmm4,nb112nf_iyO(%rsp)
        movaps %xmm5,nb112nf_izO(%rsp)

        movss %xmm0,%xmm3
        movss %xmm1,%xmm4
        movss %xmm2,%xmm5
        addss 12(%rax,%rbx,4),%xmm0
        addss 16(%rax,%rbx,4),%xmm1
        addss 20(%rax,%rbx,4),%xmm2
        addss 24(%rax,%rbx,4),%xmm3
        addss 28(%rax,%rbx,4),%xmm4
        addss 32(%rax,%rbx,4),%xmm5

        shufps $0,%xmm0,%xmm0
        shufps $0,%xmm1,%xmm1
        shufps $0,%xmm2,%xmm2
        shufps $0,%xmm3,%xmm3
        shufps $0,%xmm4,%xmm4
        shufps $0,%xmm5,%xmm5
        movaps %xmm0,nb112nf_ixH1(%rsp)
        movaps %xmm1,nb112nf_iyH1(%rsp)
        movaps %xmm2,nb112nf_izH1(%rsp)
        movaps %xmm3,nb112nf_ixH2(%rsp)
        movaps %xmm4,nb112nf_iyH2(%rsp)
        movaps %xmm5,nb112nf_izH2(%rsp)

        ## clear vctot and i forces 
        xorps %xmm4,%xmm4
        movaps %xmm4,nb112nf_vctot(%rsp)
        movaps %xmm4,nb112nf_Vvdwtot(%rsp)

        movq  nb112nf_jindex(%rsp),%rax
        movl  (%rax,%rsi,4),%ecx             ## jindex[n] 
        movl  4(%rax,%rsi,4),%edx            ## jindex[n+1] 
        subl  %ecx,%edx              ## number of innerloop atoms 

        movq  nb112nf_pos(%rbp),%rsi
        movq  nb112nf_jjnr(%rsp),%rax
        shll  $2,%ecx
        addq  %rcx,%rax
        movq  %rax,nb112nf_innerjjnr(%rsp)       ## pointer to jjnr[nj0] 
        movl  %edx,%ecx
        subl  $4,%edx
        addl  nb112nf_ninner(%rsp),%ecx
        movl  %ecx,nb112nf_ninner(%rsp)
        addl  $0,%edx
        movl  %edx,nb112nf_innerk(%rsp)      ## number of innerloop atoms 
        jge   _nb_kernel112nf_x86_64_sse.nb112nf_unroll_loop
        jmp   _nb_kernel112nf_x86_64_sse.nb112nf_single_check
_nb_kernel112nf_x86_64_sse.nb112nf_unroll_loop: 
        ## quad-unroll innerloop here 
        movq  nb112nf_innerjjnr(%rsp),%rdx       ## pointer to jjnr[k] 

        movl  (%rdx),%eax
        movl  4(%rdx),%ebx
        movl  8(%rdx),%ecx
        movl  12(%rdx),%edx           ## eax-edx=jnr1-4 

        addq $16,nb112nf_innerjjnr(%rsp)             ## advance pointer (unrolled 4) 

        movq nb112nf_pos(%rbp),%rsi        ## base of pos[] 

        lea  (%rax,%rax,2),%rax     ## replace jnr with j3 
        lea  (%rbx,%rbx,2),%rbx
        lea  (%rcx,%rcx,2),%rcx     ## replace jnr with j3 
        lea  (%rdx,%rdx,2),%rdx

        ## move j coordinates to local temp variables 
        movlps (%rsi,%rax,4),%xmm2
        movlps 12(%rsi,%rax,4),%xmm3
        movlps 24(%rsi,%rax,4),%xmm4

        movlps (%rsi,%rbx,4),%xmm5
        movlps 12(%rsi,%rbx,4),%xmm6
        movlps 24(%rsi,%rbx,4),%xmm7

        movhps (%rsi,%rcx,4),%xmm2
        movhps 12(%rsi,%rcx,4),%xmm3
        movhps 24(%rsi,%rcx,4),%xmm4

        movhps (%rsi,%rdx,4),%xmm5
        movhps 12(%rsi,%rdx,4),%xmm6
        movhps 24(%rsi,%rdx,4),%xmm7

        ## current state:       
        ## xmm2= jxOa  jyOa  jxOc  jyOc 
        ## xmm3= jxH1a jyH1a jxH1c jyH1c 
        ## xmm4= jxH2a jyH2a jxH2c jyH2c 
        ## xmm5= jxOb  jyOb  jxOd  jyOd 
        ## xmm6= jxH1b jyH1b jxH1d jyH1d 
        ## xmm7= jxH2b jyH2b jxH2d jyH2d 

        movaps %xmm2,%xmm0
        movaps %xmm3,%xmm1
        unpcklps %xmm5,%xmm0    ## xmm0= jxOa  jxOb  jyOa  jyOb 
        unpcklps %xmm6,%xmm1    ## xmm1= jxH1a jxH1b jyH1a jyH1b 
        unpckhps %xmm5,%xmm2    ## xmm2= jxOc  jxOd  jyOc  jyOd 
        unpckhps %xmm6,%xmm3    ## xmm3= jxH1c jxH1d jyH1c jyH1d 
        movaps %xmm4,%xmm5
        movaps   %xmm0,%xmm6
        unpcklps %xmm7,%xmm4    ## xmm4= jxH2a jxH2b jyH2a jyH2b                
        unpckhps %xmm7,%xmm5    ## xmm5= jxH2c jxH2d jyH2c jyH2d 
        movaps   %xmm1,%xmm7
        movlhps  %xmm2,%xmm0    ## xmm0= jxOa  jxOb  jxOc  jxOd 
        movaps %xmm0,nb112nf_jxO(%rsp)
        movhlps  %xmm6,%xmm2    ## xmm2= jyOa  jyOb  jyOc  jyOd 
        movaps %xmm2,nb112nf_jyO(%rsp)
        movlhps  %xmm3,%xmm1
        movaps %xmm1,nb112nf_jxH1(%rsp)
        movhlps  %xmm7,%xmm3
        movaps   %xmm4,%xmm6
        movaps %xmm3,nb112nf_jyH1(%rsp)
        movlhps  %xmm5,%xmm4
        movaps %xmm4,nb112nf_jxH2(%rsp)
        movhlps  %xmm6,%xmm5
        movaps %xmm5,nb112nf_jyH2(%rsp)

        movss  8(%rsi,%rax,4),%xmm0
        movss  20(%rsi,%rax,4),%xmm1
        movss  32(%rsi,%rax,4),%xmm2

        movss  8(%rsi,%rcx,4),%xmm3
        movss  20(%rsi,%rcx,4),%xmm4
        movss  32(%rsi,%rcx,4),%xmm5

        movhps 4(%rsi,%rbx,4),%xmm0
        movhps 16(%rsi,%rbx,4),%xmm1
        movhps 28(%rsi,%rbx,4),%xmm2

        movhps 4(%rsi,%rdx,4),%xmm3
        movhps 16(%rsi,%rdx,4),%xmm4
        movhps 28(%rsi,%rdx,4),%xmm5

        shufps $204,%xmm3,%xmm0 ## 11001100
        shufps $204,%xmm4,%xmm1 ## 11001100
        shufps $204,%xmm5,%xmm2 ## 11001100
        movaps %xmm0,nb112nf_jzO(%rsp)
        movaps %xmm1,nb112nf_jzH1(%rsp)
        movaps %xmm2,nb112nf_jzH2(%rsp)

        movaps nb112nf_ixO(%rsp),%xmm0
        movaps nb112nf_iyO(%rsp),%xmm1
        movaps nb112nf_izO(%rsp),%xmm2
        movaps nb112nf_ixO(%rsp),%xmm3
        movaps nb112nf_iyO(%rsp),%xmm4
        movaps nb112nf_izO(%rsp),%xmm5
        subps  nb112nf_jxO(%rsp),%xmm0
        subps  nb112nf_jyO(%rsp),%xmm1
        subps  nb112nf_jzO(%rsp),%xmm2
        subps  nb112nf_jxH1(%rsp),%xmm3
        subps  nb112nf_jyH1(%rsp),%xmm4
        subps  nb112nf_jzH1(%rsp),%xmm5
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        addps  %xmm1,%xmm0
        addps  %xmm2,%xmm0
        addps  %xmm4,%xmm3
        addps  %xmm5,%xmm3
        movaps %xmm0,nb112nf_rsqOO(%rsp)
        movaps %xmm3,nb112nf_rsqOH1(%rsp)

        movaps nb112nf_ixO(%rsp),%xmm0
        movaps nb112nf_iyO(%rsp),%xmm1
        movaps nb112nf_izO(%rsp),%xmm2
        movaps nb112nf_ixH1(%rsp),%xmm3
        movaps nb112nf_iyH1(%rsp),%xmm4
        movaps nb112nf_izH1(%rsp),%xmm5
        subps  nb112nf_jxH2(%rsp),%xmm0
        subps  nb112nf_jyH2(%rsp),%xmm1
        subps  nb112nf_jzH2(%rsp),%xmm2
        subps  nb112nf_jxO(%rsp),%xmm3
        subps  nb112nf_jyO(%rsp),%xmm4
        subps  nb112nf_jzO(%rsp),%xmm5
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        addps  %xmm1,%xmm0
        addps  %xmm2,%xmm0
        addps  %xmm4,%xmm3
        addps  %xmm5,%xmm3
        movaps %xmm0,nb112nf_rsqOH2(%rsp)
        movaps %xmm3,nb112nf_rsqH1O(%rsp)

        movaps nb112nf_ixH1(%rsp),%xmm0
        movaps nb112nf_iyH1(%rsp),%xmm1
        movaps nb112nf_izH1(%rsp),%xmm2
        movaps nb112nf_ixH1(%rsp),%xmm3
        movaps nb112nf_iyH1(%rsp),%xmm4
        movaps nb112nf_izH1(%rsp),%xmm5
        subps  nb112nf_jxH1(%rsp),%xmm0
        subps  nb112nf_jyH1(%rsp),%xmm1
        subps  nb112nf_jzH1(%rsp),%xmm2
        subps  nb112nf_jxH2(%rsp),%xmm3
        subps  nb112nf_jyH2(%rsp),%xmm4
        subps  nb112nf_jzH2(%rsp),%xmm5
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        addps  %xmm1,%xmm0
        addps  %xmm2,%xmm0
        addps  %xmm4,%xmm3
        addps  %xmm5,%xmm3
        movaps %xmm0,nb112nf_rsqH1H1(%rsp)
        movaps %xmm3,nb112nf_rsqH1H2(%rsp)

        movaps nb112nf_ixH2(%rsp),%xmm0
        movaps nb112nf_iyH2(%rsp),%xmm1
        movaps nb112nf_izH2(%rsp),%xmm2
        movaps nb112nf_ixH2(%rsp),%xmm3
        movaps nb112nf_iyH2(%rsp),%xmm4
        movaps nb112nf_izH2(%rsp),%xmm5
        subps  nb112nf_jxO(%rsp),%xmm0
        subps  nb112nf_jyO(%rsp),%xmm1
        subps  nb112nf_jzO(%rsp),%xmm2
        subps  nb112nf_jxH1(%rsp),%xmm3
        subps  nb112nf_jyH1(%rsp),%xmm4
        subps  nb112nf_jzH1(%rsp),%xmm5
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        addps  %xmm1,%xmm0
        addps  %xmm2,%xmm0
        addps  %xmm3,%xmm4
        addps  %xmm5,%xmm4
        movaps %xmm0,nb112nf_rsqH2O(%rsp)
        movaps %xmm4,nb112nf_rsqH2H1(%rsp)

        movaps nb112nf_ixH2(%rsp),%xmm0
        movaps nb112nf_iyH2(%rsp),%xmm1
        movaps nb112nf_izH2(%rsp),%xmm2
        subps  nb112nf_jxH2(%rsp),%xmm0
        subps  nb112nf_jyH2(%rsp),%xmm1
        subps  nb112nf_jzH2(%rsp),%xmm2
        mulps %xmm0,%xmm0
        mulps %xmm1,%xmm1
        mulps %xmm2,%xmm2
        addps %xmm1,%xmm0
        addps %xmm2,%xmm0
        movaps %xmm0,nb112nf_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  nb112nf_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   nb112nf_half(%rsp),%xmm3   ## rinvH2H2 
        mulps   nb112nf_half(%rsp),%xmm7   ## rinvH2H1 
        movaps  %xmm3,nb112nf_rinvH2H2(%rsp)
        movaps  %xmm7,nb112nf_rinvH2H1(%rsp)

        rsqrtps nb112nf_rsqOO(%rsp),%xmm1
        rsqrtps nb112nf_rsqOH1(%rsp),%xmm5
        movaps  %xmm1,%xmm2
        movaps  %xmm5,%xmm6
        mulps   %xmm1,%xmm1
        mulps   %xmm5,%xmm5
        movaps  nb112nf_three(%rsp),%xmm3
        movaps  %xmm3,%xmm7
        mulps   nb112nf_rsqOO(%rsp),%xmm1
        mulps   nb112nf_rsqOH1(%rsp),%xmm5
        subps   %xmm1,%xmm3
        subps   %xmm5,%xmm7
        mulps   %xmm2,%xmm3
        mulps   %xmm6,%xmm7
        mulps   nb112nf_half(%rsp),%xmm3
        mulps   nb112nf_half(%rsp),%xmm7
        movaps  %xmm3,nb112nf_rinvOO(%rsp)
        movaps  %xmm7,nb112nf_rinvOH1(%rsp)

        rsqrtps nb112nf_rsqOH2(%rsp),%xmm1
        rsqrtps nb112nf_rsqH1O(%rsp),%xmm5
        movaps  %xmm1,%xmm2
        movaps  %xmm5,%xmm6
        mulps   %xmm1,%xmm1
        mulps   %xmm5,%xmm5
        movaps  nb112nf_three(%rsp),%xmm3
        movaps  %xmm3,%xmm7
        mulps   nb112nf_rsqOH2(%rsp),%xmm1
        mulps   nb112nf_rsqH1O(%rsp),%xmm5
        subps   %xmm1,%xmm3
        subps   %xmm5,%xmm7
        mulps   %xmm2,%xmm3
        mulps   %xmm6,%xmm7
        mulps   nb112nf_half(%rsp),%xmm3
        mulps   nb112nf_half(%rsp),%xmm7
        movaps  %xmm3,nb112nf_rinvOH2(%rsp)
        movaps  %xmm7,nb112nf_rinvH1O(%rsp)

        rsqrtps nb112nf_rsqH1H1(%rsp),%xmm1
        rsqrtps nb112nf_rsqH1H2(%rsp),%xmm5
        movaps  %xmm1,%xmm2
        movaps  %xmm5,%xmm6
        mulps   %xmm1,%xmm1
        mulps   %xmm5,%xmm5
        movaps  nb112nf_three(%rsp),%xmm3
        movaps  %xmm3,%xmm7
        mulps   nb112nf_rsqH1H1(%rsp),%xmm1
        mulps   nb112nf_rsqH1H2(%rsp),%xmm5
        subps   %xmm1,%xmm3
        subps   %xmm5,%xmm7
        mulps   %xmm2,%xmm3
        mulps   %xmm6,%xmm7
        mulps   nb112nf_half(%rsp),%xmm3
        mulps   nb112nf_half(%rsp),%xmm7
        movaps  %xmm3,nb112nf_rinvH1H1(%rsp)
        movaps  %xmm7,nb112nf_rinvH1H2(%rsp)

        rsqrtps nb112nf_rsqH2O(%rsp),%xmm1
        movaps  %xmm1,%xmm2
        mulps   %xmm1,%xmm1
        movaps  nb112nf_three(%rsp),%xmm3
        mulps   nb112nf_rsqH2O(%rsp),%xmm1
        subps   %xmm1,%xmm3
        mulps   %xmm2,%xmm3
        mulps   nb112nf_half(%rsp),%xmm3
        movaps  %xmm3,nb112nf_rinvH2O(%rsp)

        ## start with OO interaction 
        movaps nb112nf_rinvOO(%rsp),%xmm0
        movaps %xmm0,%xmm7
        mulps  %xmm0,%xmm0
        movaps %xmm0,%xmm1
        mulps  %xmm0,%xmm1
        mulps  %xmm0,%xmm1      ## xmm1=rinvsix 
        mulps  nb112nf_qqOO(%rsp),%xmm7
        movaps %xmm1,%xmm2
        mulps  %xmm2,%xmm2      ## xmm2=rinvtwelve 
        mulps  nb112nf_c6(%rsp),%xmm1
        mulps  nb112nf_c12(%rsp),%xmm2
        subps  %xmm1,%xmm2      ## xmm3=Vvdw12-Vvdw6 
        addps  nb112nf_Vvdwtot(%rsp),%xmm2
        movaps %xmm2,nb112nf_Vvdwtot(%rsp)
        addps  nb112nf_vctot(%rsp),%xmm7

        ## all other interaction 
        movaps nb112nf_rinvOH1(%rsp),%xmm0
        movaps nb112nf_rinvH1H1(%rsp),%xmm1
        addps  nb112nf_rinvOH2(%rsp),%xmm0
        addps  nb112nf_rinvH1H2(%rsp),%xmm1